In 2014, Facebook paid $2 billion for a company with roughly 75 employees.

That company had not shipped a single consumer product.

The founder was 21 years old. Its most famous product was a duct-taped prototype for a VR headset.

Fortunately, a video-games legend introduced that headset to the biggest gaming conference in the world.

And that’s when Mark Zuckerberg wrote that check.

This is the story of how Oculus revolutionized VR. It is the story of Palmer Luckey, a homeschooled kid who built over 50 VR headsets in his parents’ garage.

I’ll also talk about what happened after the check was cashed. The backlash and Palmer’s firing.

And lastly, we have to talk about Meta's billion-dollar losses in the metaverse.

Building Headsets in a Garage

Palmer Luckey’s story is as interesting as the tech he has worked on.

He was born to a car salesman. His mom homeschooled all 4 of her kids. Palmer was the only one who never went back to regular school. He liked the flexibility too much.

From an early age Luckey was obsessed with electronics. He built high-powered lasers in his garage. He built coil guns.

He once electrocuted himself touching a grounded metal bed frame. He burned a gray spot into his vision cleaning an infrared laser.

Normal kids stuff, clearly.

To fund his experiments he fixed and resold broken iPhones. He worked as a groundskeeper and a youth sailing coach. He earned at least $36,000 this way before he was old enough to vote.

But his real obsession was virtual reality.

Luckey had spent years collecting old VR headsets. He had one of the largest private collections of VR hardware in the world at a time when he was a teenager living with his parents.

He had seen what VR was supposed to be and he was not impressed with what was available.

Every headset he owned had similar problems. They were heavy, expensive and didn't immerse you in a world as a good headset should.

So he decided to build his own.

Starting around 2009, when he was 16 or 17, Luckey began prototyping in the garage. He had a simple idea that most in the industry weren’t paying that much attention to.

Smartphone screens were getting really good really fast. They were getting sharper, brighter, and cheaper. What if you used one as the display inside a VR headset?

Then he had another idea. Instead of using expensive, bulky optics, what if you used cheap lenses and fixed distortions in software?

Both ideas together meant you could build a VR headset that was lighter, cheaper, and had a much wider field of view than anything else on the market.

For those who don’t know, the Field of View or FOV is how wide you could see inside a VR headset. A higher FOV means you’d feel more immersed.

Over the next few years Palmer built prototype after prototype. Over 50 of them in total. Each one refining his techniques a little more.

By his sixth generation prototype he had something he called the Oculus Rift. It had 3D, low-latency head tracking, and a 90-degree field of view.

Nobody outside a small VR enthusiast internet forum called MTBS3D had heard of it, but an interesting person was about to join the forum.

What Oculus Actually Was When It Started

In early 2012 Luckey was posting updates on MTBS3D, the niche online community where VR nerds compared notes and argued about specs.

Read more

Some days ago, I was watching one of Linus Tech Tips’ recent videos reviewing Samsung’s Galaxy Book 6 enterprise edition. During the video, Linus also demonstrated Intel vPro, an enterprise platform used by many organizations to manage company-owned computers.

What caught my attention was seeing just how much control administrators could have over a device.

Just using a central dashboard, an IT department could remotely wipe a laptop, deploy software across an entire fleet of computers, troubleshoot devices even when the operating system wasn’t working properly, and enforce security policies without ever physically touching the machine.

If a normal person watches something like that, they’ll probably think, “That’s pretty useful for IT departments.”

But I had a different thought: if you use a work laptop, there’s a good chance your organization has access to many of the same management capabilities.

And that’s what made me write this post.

Many people think of a work laptop as “their laptop” because they’re the ones carrying it around every day. In reality, a work laptop is a part of a centrally managed system designed to protect company data and allow IT administrators to manage hundreds or even thousands of devices at the same time.

The goal of this article is simply awareness.

Once you understand how modern device management works, you’ll probably look at your work laptop a little differently.

What a Managed Laptop Is

To the human eye, a personal laptop and a company-issued laptop can look almost the same because they both run Windows or macOS. They open the same applications and connect to Wi-Fi in the same way.

But underneath the hood, they can be completely different devices.

Most medium and large companies no longer configure computers manually. They don’t have enough time or resources to install software, update security settings or recover stolen laptops across thousands of employees.

Instead, companies rely on different kinds of enterprise device management platforms.

For example, Microsoft Intune allows administrators to configure devices and wipe company data remotely when necessary. Similarly, Microsoft Defender for Endpoint focuses on monitoring devices for security threats and helping companies respond to incidents across their entire fleet.

Other companies follow a similar path as well, just in a different way.

Organizations running Apple devices commonly use Apple Business Manager alongside Mobile Device Management (MDM) platforms such as Jamf. Samsung offers Knox Manage for enterprise device management. Intel provides vPro for hardware-level administration.

Exactly what an administrator can do depends entirely on how an organization has configured these platforms. Some companies enable only basic security features. Others implement far more comprehensive management policies. The level of visibility and control varies significantly between organizations.

That brings us to another important point: using a managed laptop doesn’t automatically mean someone is monitoring everything you do. Those capabilities are entirely dependent on the company’s policy and the management tools being used.

Common Myths About Work Laptops

By now, you probably understand that a work laptop is different from a personal one, but where most people get confused is how they use it.

Many employees assume that using another browser or creating a separate Windows profile somehow turns part of the laptop into their own space.

But enterprise management doesn’t usually operate inside a single browser or user profile. It exists at the device level, where operating system settings, security policies and management software all work together.

That doesn’t mean your employer can automatically see everything you do, but no matter what you do, your company will always own the laptop and will have the upper hand.

Anyway, let’s bust some common myths about work devices.

Myth #1: “I’ll Just Use Another Browser”

One of the most common misconceptions is that switching from Microsoft Edge to Brave or to another browser somehow bypasses company management.

Changing browsers doesn’t change who owns or manages the device.

Organizations can apply policies that affect multiple browsers through enterprise management platforms, installed certificates, DNS settings, VPN configurations, or compliance requirements. Many browsers, including Chrome and Microsoft Edge, also support enterprise policies that allow organizations to configure security settings and other features centrally.

Myth #2: “Incognito Mode Hides Everything”

The good old Incognito mode is one of the most misunderstood features in modern browsers. It was never designed to make browsing invisible or keep you anonymous.

Its primary purpose is to avoid storing browsing history, cookies, cached files, and session information on the local device after you close the window. That’s useful if you’re sharing a computer or don’t want your own browser history saved.

It doesn’t disable enterprise policies, VPN routing, DNS logging, or other security controls that may exist on a managed device or corporate network.

In other words, even the mighty Incognito doesn’t possess the power to stop the organization managing the device.

Myth #3: “It’s My Personal Google or Microsoft Account”

Another common assumption is that signing into a personal account such as a Gmail, Outlook, or Google Drive account somehow separates that data from the company laptop.

In most cases, enterprise management doesn’t automatically give IT administrators access to your personal cloud account simply because you signed in.

Your personal accounts remain your personal accounts, but the confusion comes from what happens on the device itself.

Downloaded files, synchronized folders, temporary files, and locally stored copies may all exist on the company-owned laptop. If the device is remotely wiped, replaced, or returned to the organization, those local copies can disappear along with everything else stored on the machine.

Myth #4: “IT Can See Absolutely Everything”

This misconception is almost as common as the previous ones.

Platforms such as Microsoft Intune, Apple’s Jamf, Samsung Knox, and similar enterprise management tools don’t automatically provide unrestricted access to passwords, encrypted messages, webcam feeds, microphone recordings, or every key you type.

Those capabilities generally require additional software or specialized monitoring products that an organization has deliberately chosen to deploy.

Every company configures its environment differently. Some organizations implement only basic device management. Others have much stricter security requirements because they operate in industries such as healthcare, finance, or government.

Myth #5: “I Trust My Employer”

Trust isn’t the point. Ownership is. This is about understanding who owns the device you’re using.

Most people have perfectly good employers that are simply trying to keep company systems secure. But organizations change over time. Employees leave and IT teams turn over.

But the fact that the laptop was never yours doesn't change.

It’s a company asset that has been assigned to you for work. Thinking about it that way makes many of the questions in this article much easier to answer.

What Happens When You Leave the Company

When people start treating their work laptop like a personal computer, years of personal files slowly accumulate on a device that ultimately belongs to someone else.

Then one day they resign, they’re laid off, or the company simply asks for the laptop back. This is where many people discover that enterprise devices follow a very different lifecycle than personal computers.

As part of the offboarding process, organizations often revoke access to internal systems, reclaim the device, and in many cases remotely wipe company-managed data before preparing the laptop for its next employee.

Exactly how that process works depends on the organization, but one thing is consistent across most companies. Once your employment ends, the laptop is expected to return to the organization.

If personal files exist only on that device, there’s a chance they disappear along with everything else.

Illusion of Remote Work

I work remotely as well.

In my case, I use my own personal computer for work. But many people working remotely receive a company-issued device instead. When you're working from home every day, it's easy to forget that the laptop on your desk is still a corporate device.

It connects to your home Wi-Fi. It sits next to your personal computer. It travels with you. Eventually, it starts feeling like just another device in the house.

Behind the scenes, however, very little has changed. And this also leads to another common misconception.

People think connecting a managed work laptop to their home Wi-Fi exposes everything to the company. That isn't true. Those are separate devices, and the organization manages the laptop, not your entire home network.

What the organization may monitor is traffic generated by the managed device itself, depending on how the organization has configured its security systems.

Why Companies Need This Level of Control

You might be wondering, “Why do companies go to all this trouble in the first place?”

It’s because companies have security responsibilities that most individual users don’t.

A stolen laptop might contain confidential customer information. A ransomware infection on a single employee’s device could spread across an organization’s network. Healthcare providers, financial institutions, government agencies, and many other industries are also required to protect sensitive information under various laws and regulations.

A February 2010 Aberdeen Group study sponsored by Intel found that for every 100 enterprise devices deployed, only 85 come back. 5 are lost or stolen, 1 of those gets recovered, and 11 simply go missing, unaccounted for.

Managing devices centrally allows organizations to respond much more quickly when something goes wrong. A stolen laptop can often be remotely locked or wiped. Critical security updates can be deployed across thousands of devices at the same time. If suspicious activity is detected, administrators can isolate an affected device before the problem spreads further.

Enterprise device management wasn’t created because companies wanted to spy on employees.

It exists because securing a large number of connected devices has become one of the biggest challenges in modern cybersecurity.

Practical Ways to Use a Work Laptop Safely

For many people, none of this is really a choice.

If your employer gives you a work laptop, chances are you’ll be using it every day.

The main goal is to understand how to use the work device without unintentionally mixing your personal and professional lives.

Let’s go over a few habits that can make a surprisingly big difference.

Separate Ownership from Usage

Don’t think of it as your device. Think of it as a computer you’re temporarily allowed to use.

Once you think of it that way, you're much less likely to store years of family photos, financial documents, passwords, or other irreplaceable files on a machine you don't own.

Whenever possible, keep your personal life on your personal devices. Banking, taxes, password managers, and private documents are all better suited to hardware you control.

Plan for the Day You Lose Access

Many employees assume they’ll have plenty of time to copy everything before returning their laptop, but the world doesn’t always work that way.

Resignations, layoffs, security incidents, damaged devices, or simple offboarding procedures can all result in losing access much sooner than expected.

By the time you realize you need something from the laptop, it may already be gone.

That’s why it’s important to have a Plan B.

Treat anything stored only on a work device as temporary.

It’s also worth remembering that synchronization isn’t the same as a backup. A synchronized OneDrive or Google Drive folder managed by your employer doesn’t replace an independent backup that you control yourself.

Minimize Your Digital Footprint

Sometimes logging into a personal account is unavoidable, but that’s very different from making a work laptop the center of your digital life.

Avoid permanently signing into personal password managers, cloud storage services, shopping accounts, streaming platforms, or social media unless you genuinely need them. Every additional account creates another point where your personal and professional lives become intertwined.

It’s also worth taking a few minutes to read your company’s acceptable use policy instead of simply clicking “Accept” during onboarding.

Knowing whether personal use is allowed, how company devices are managed, and what happens during offboarding removes a lot of uncertainty.

What to Do If It’s Your Only Computer

I understand that not everyone has the luxury of keeping separate devices.

Some people travel constantly. Others work remotely full-time or simply can’t justify buying another laptop.

If you are one of those people whose work laptop is the only computer you have, focus on reducing your exposure instead of trying to eliminate it completely.

• Keep personal information to an absolute minimum.

• Maintain independent backups of anything important.

• Avoid storing irreplaceable files locally.

And always assume the laptop won't be with you forever. Organizing your digital life around that assumption can save a lot of frustration later.

Check Other Security Focused Explainers

• The Illusion of Stability: The Internet Runs on a Handful of Companies

• The Illusion of Privacy: Why Your Data Isn’t Secure

• Strong Passwords Still Get Hacked - Here’s Why

• The Illusion of Choice: How Tech Decides Before You Do

• The Death of Ownership: Why Your Tech Isn’t Really Yours Anymore

In Part 1 of the series, we learned how the creator economy was built on the idea that content itself was the business. Views turned into ad revenue, ad revenue turned into a living, and the work stood on its own.

But Part 2 showed why that model is under pressure, why platform revenue has become less reliable, and why trust has become harder to measure.

This piece looks at what creators are doing about it, and whether any of it addresses the deeper problem or just works around it.

What Creators Are Doing

Ad revenue has gotten less predictable. Platform reach has gotten harder to count on. AI has made content cheaper to produce at scale, and manufactured trust has made it harder for audiences to tell a genuine recommendation from a coordinated one.

That’s the environment creators are operating in right now, and it’s pushing them in two directions. Some are moving closer to traditional business by building physical products around their audience. Others are focusing on direct relationships that reduce how much they depend on any single platform.

Neither approach removes the underlying problem. But both show how creators are adapting to a system that’s gotten harder to rely on.

Let’s take a look at both of these approaches that creators are using to survive in today’s economy.

Approach 1: Building Businesses Beyond Content

One of the clearest responses to declining content value has been creators moving into physical products.

Digital content can now be produced and copied at a scale that makes it harder to hold value, but physical products don’t work that way. They require manufacturing, supply chains, and real investment, which makes them harder to copy and easier to defend.

For creators with large audiences, that difficulty becomes the opportunity. Years of built-up attention becomes the distribution channel for a product business.

MrBeast launched Feastables, a chocolate brand built around his audience, in January 2022. Feastables generated $250 million in revenue and $20 million in profit in 2024, which made a chocolate bar brand more profitable than the largest YouTube channel on the planet.

Similarly, Logan Paul and KSI’s Prime Hydration shows how quickly a creator audience can turn into a consumer business. Bloomberg reported that Prime was on pace to surpass $1.2 billion in sales in 2023, up from $250 million in its first year.

But the growth did not last. The story is useful because it shows both the upside and the limits of turning audience attention into a product business.

Read more

I was watching the WAN Show a while back when Linus Sebastian shared this cool thought.

He was talking about Valve’s new Steam Machine, and he made a point nobody else was really making.

See, everybody keeps comparing this thing to the PlayStation and the Xbox. Linus thinks the real rival is the Mac Mini.

That sounds a little crazy at first. A gaming box from Valve going up against Apple’s tiny work computer?

But the more he explained it, the more it clicked for me. And I want to build on that idea today.

Here is the short version. Judge the Steam Machine as a console and it looks bad. Too pricey, too weak.

But judge it as a small, quiet computer that one company built from the chip all the way up to the software, and it suddenly stands almost alone.

The only other famous box that works this way is the Mac Mini. So let’s talk about why these two are secretly rivals, and what this means for the future of desktop computers.

Everybody Reviewed the Wrong Machine

Console pricing works on a trick. Sony and Microsoft sell them at a loss.

They lose money on every PS5 and Xbox that leaves the shelf. They earn it back later by taking a cut of every game you buy. The console is basically bait because the games are where the real money is.

Valve does not want that for Steam Machine. They plainly said in their FAQ that the Steam Machine is priced like a PC, not like a console.

There is no selling it cheap now and clawing the money back later, because Valve does not need to. It already prints money from the Steam store itself.

So when people line the Steam Machine up next to a PS5 on price, they are comparing a computer sold at its true cost against a console sold at a loss. Of course the console looks cheaper.

The moment you stop thinking console and start thinking small desktop computer, everything shifts. And there is really only one famous small desktop that is built the way this one is.

What Valve Actually Built

Let’s cover what the Steam Machine actually is first.

It is a little cube, roughly 6 inches on each side. Valve calls it Steam Machine, the community calls it the GabeCube, after Valve boss Gabe Newell.

It launched on June 29, 2026, starting at $1,049 for the 512GB version.

Inside sits a custom AMD chip built to Valve’s spec. You get a 6-core processor based on AMD’s Zen 4 design that runs up to 4.8 GHz, paired with a graphics chip based on AMD’s RDNA 3 architecture.

The GPU has its own 8GB of video memory. There is 16GB of regular RAM too, and you pick either 512GB or 2TB of storage. It runs Valve’s own SteamOS.

Valve says it is about 6 times more powerful than the Steam Deck, their handheld. The goal is 4K gaming at 60 frames per second, though it leans on a trick called FSR to get there.

FSR is basically smart upscaling. The chip draws the game at a lower resolution, then cleverly fills in the detail so it looks close to 4K without doing all the hard work.

One genuinely nice touch is that you can open this thing. It has a slot to add storage and two slots to add memory, and Valve sells official replacement parts through iFixit.

Even Linus pointed this out in his video and it matters a lot in this comparison as you will see.

Did you know Valve just opened up their SteamOS for anyone to build their own Steam Machine?

Here’s my linked article where I turned my ThinkPad into a Steam Machine in less than 60 minutes.

The Vertical Integration Magic

Vertical integration just means one company makes the whole thing instead of splitting the work. The hardware, the box, and the software all come from the same place, built to fit together.

Think about a normal Windows PC.

Microsoft makes Windows. AMD or Intel makes the chip. Some other company like Dell builds the actual computer.

That is three different companies, none of them really talking, all their parts meeting inside your machine and hoping to get along.

That’s why Windows PCs can feel janky sometimes. A driver breaks. A setting fights you. Two bits of software disagree.

Nobody owns the whole thing, so nobody can make it all just work in a perfect way.

When one company controls everything, they can tune it tight. The software knows the exact chip it is running on. The power settings are built for that exact hardware.

Stuff just works, because one team made sure it does. And this is the real reason the Steam Machine is interesting.

Valve designed the chip, wrote the OS, made the interface, and controls the updates. That is a purpose-built appliance, not a generic beige box.

And there is only one other company known for doing this in a tiny desktop.

The King of Vertical Integration

You already guessed it. It’s Apple.

Look, Apple is the poster child for controlling the whole stack.

Ever since it started making its own chips, called Apple Silicon, the Mac runs on hardware Apple designed, with software Apple wrote, in a box Apple built. Everything from one house.

That is a big reason Macs feel smooth and last so long on battery. The software and the chip were made for each other. No third parties means no guessing games.

The Mac Mini is the purest version of this idea. It is just the computer. A small square box, no screen, no keyboard, no mouse. You bring your own.

It is Apple’s whole approach shrunk down to the smallest, cheapest Mac you can buy.

So here is where we land.

On one side, the Mac Mini, one company controlling everything inside a closed Apple world. On the other, the Steam Machine, one company controlling everything but built on open parts.

These two boxes have a similar idea but opposite personalities. That is why they belong on the same shelf, and it is a shelf with almost nothing else on it.

Now let’s compare the two “boxes.”

Vertical integration is what got us arguably the best value laptop of this decade.

Here’s a link to my article where I discuss the MacBook Neo and how it changed the laptop game for us all.

The Price Argument

We just can’t ignore price because it is the roughest part for Valve.

The base Mac Mini with the M6 chip starts at $899 (once it ships this month).

For that you get Apple’s M6 chip, 16GB of memory, and 256GB of storage, in that tiny 5-inch square. It is still one of the best value computers you can buy.

The Steam Machine starts at $1,049. That is about $150 more than the Mac Mini before you even start.

The funny part though:

The base Mac Mini used to be even cheaper. It went for around $599 until May 2026. That’s when Apple killed off the cheapest version and bumped the entry price to $799.

Then later in August of this year, Apple announced the Mac Mini with the M6 chip and it has gone up $100 in price. It will ship this month.

In a way, the Mac Mini had its price bumped two times just this year.

When a company like Apple has to raise prices in that way, you know things are getting bad!

If you read my article on RAM crisis, I mentioned Micron leaving the consumer memory business and warned that RAM prices were about to punch everyone in the face.

Well, that was the punch. Both Valve and Apple raised prices in 2026 for the same reason. Nobody escaped it.

To be fair to Valve, you are not being ripped off.

GamersNexus priced out building an equal box from parts and landed around $979, so Valve’s $1,049 is close to what a similar machine would cost you to build yourself.

You pay a small premium for something tiny, quiet, and pre-built. But close to fair is not the same as cheap.

And against an $899 Mac Mini, the Steam Machine has to earn that extra cash.

SteamOS Is Secretly a Really Nice Desktop

Here is the part that genuinely surprised people, and it is the most interesting bit of the whole story.

SteamOS, the software on the Steam Machine, is not just a game launcher.

Underneath, it is a full desktop operating system just like Microsoft Windows.

Unlike Windows, though, it is built on Linux, the free and open-source system that quietly runs a huge chunk of the internet’s servers.

On top of that Linux base sits a desktop called KDE Plasma, which looks and behaves a lot like Windows, with a taskbar, windows, and a menu.

Getting to it is easy. You tap a reboot to desktop button, wait a few seconds, and the gaming interface drops you into a normal desktop where you can browse, write, and do computer stuff.

This is the part Linus fell in love with, and his examples say it best.

His oddball high-resolution monitor was detected right away. The settings menus are simple instead of buried. Mouse pointer acceleration is turned off by default, which is exactly what you want.

The biggest thing is that there is no bloatware shoved in your face and no pop-up nagging you to buy a subscription.

And this last part is the quiet killer.

Windows in 2026 keeps pushing ads and pop-ups at you inside your own computer. And plenty of people are sick of it.

Linus put it perfectly. He said a product you pay for should feel comfortable out of the box, and he is right.

SteamOS feels like software that respects you and gets out of your way. That is the whole job of an operating system. When an OS does its job right, you won’t know about it.

Apps come through a system called Flatpak, which is basically an app store for Linux that keeps each program tidy in its own box. Chrome, VLC, OBS, a calculator, the usual stuff.

It is not as stuffed as Windows or Mac for random software, but for everyday needs it covers a lot.

The modern SteamOS wouldn’t exist if it weren’t for a revolutionary piece of hardware. I’m talking about the Steam Deck.

Here’s my linked article that goes over how Steam Deck revolutionized handheld consoles and got us an alternative to Windows.

The Immutable OS Thing Explained

SteamOS is what people call an immutable OS.

In plain terms, the core of the system is locked, and it resets itself back to normal every time you reboot. Install something the normal way and it survives. Install it outside the normal system, and a reboot wipes it.

Your files and games stay where they are, but deep changes to the system itself get wiped.

The upside is that it’s pretty hard to break SteamOS. On a normal computer, you can mess up one setting or delete the wrong file and suddenly nothing boots.

On SteamOS, if things go sideways, a restart puts the core back the way it should be. For a regular person, that is fantastic. It is really hard to shoot yourself in the foot.

The downside shows up if you like to tinker.

Linus talked about this too: he installed a small app outside the normal Flatpak system to get an audio device working, and after a reboot the whole thing reset and stopped.

That is the tradeoff SteamOS makes, and for most people it is the right one. For tinkerers, it is the one thing to know going in.

Rough Edges Around the Cube

I love Valve but I won’t oversell the Steam Machine, so let’s talk about its rough edges.

The biggest miss is the ports.

The Steam Machine does not have USB4 or Thunderbolt, which are the fast connections that let you run a whole desk setup, monitor and drives and everything, through a single cable.

Its fastest USB-C port is quick at 10 gigabits per second, but it cannot even output video to a monitor.

It also uses regular Gigabit Ethernet instead of the faster 2.5 gigabit kind, which can be an issue for some.

Then there are the small Linux annoyances that pop up in daily use.

Pulling video files off an iPhone turned into a real headache, mostly thanks to how Apple locks things down, and Linus ended up doing the transfer on Windows instead.

And middle-click auto-scroll, the handy trick where you click the scroll wheel to glide down a long page, just does not work properly on SteamOS yet.

None of these are dealbreakers on their own though. But stack them together and they remind you this is a gaming-first machine wearing a desktop hat, not just a work computer.

The Mac Mini Isn’t Perfect Either

Just like the Steam Machine, the Mac Mini has its own rough edges.

Its big issue is that you cannot upgrade it at all.

The RAM and storage are soldered in at the factory. Whatever you buy on day one is what you have forever. There is no popping it open to add more later like you can with the Steam Machine.

And Apple has historically charged brutal prices to bump those specs up front.

Want more storage or memory? Apple’s upgrade pricing is famously steep, and it pushes that $899 starting price up fast.

The Steam Machine, with its openable case and swappable parts, clearly wins on repairability and living a long life.

The Mac Mini is also a closed world. You run macOS, and that is that. Apple does not give you an escape hatch.

The Steam Machine is the opposite. You do not like SteamOS? You can wipe it and install Windows, and Valve is completely fine with that. That freedom is a real point in its favor.

And of course, the Mac Mini basically cannot game. For a lot of buyers that does not matter one bit. For others it is the whole reason they showed up.

More Choice for the Rest of Us

This is the crux of this whole discussion.

For years, if you wanted a computer, you basically had two choices.

1. Buy into Apple’s world, where everything is smooth but closed and you play by Apple’s rules.

2. Buy a Windows PC, where you get freedom but also driver issues, the bloat, and now the ads.

The Steam Machine quietly adds a third door.

It takes Apple’s best idea, one company controlling the whole stack, and builds it out of open parts instead of a walled garden.

As an end-user, you get the smoothness of a tightly built machine and the freedom to crack it open and do what you want.

Both Valve and Apple decided the way to make a great small computer is to control the whole stack.

Apple did it with a closed system. Valve did it with an open one.

And for the first time, there is a real desktop that is neither Windows nor Mac, one that a normal person can just plug in and use.

Linus called 2026 the year of the Linux desktop, and watching the Steam Machine, it is hard to argue.

The point isn’t a comparison between the Mac Mini and the Steam Machine.

It is that the choice finally got wider for the end-user.

And more options means more competition, which is always a good thing for all of us.

If you liked reading through this free article, a share to someone as curious as you would be really cool. Just use the blue button below:

Share SK NEXUS

Similar Articles from SK NEXUS

Further Learning

• I Love The Steam Machine More Already - WAN Show June 26, 2026

• Valve Accidentally Made the Best Desktop PC I’ve Ever Used

• Valve Steam Machine Review: GPU & CPU Benchmarks, SteamOS Test, Thermals, Noise, and Price

• Building a PC to BEAT the STEAM Machine!

• Valve Steam Machine, Desktop SteamOS, Steam Frame VR, & Controller

• Even Valve is Disappointed - Steam Machine Review

• The Steam Machine is the most ambitious game console I’ve ever played

• The Steam Machine fits my TV, my desk, and my life

• The Steam Machine is actually a Mac mini competitor, but not in the way you think

• Valve explains why it isn’t subsidizing the Steam Machine

• Apple raises the Mac Mini’s starting price

• Apple was ready for the RAM crisis

• Apple to Raise Prices Due to Memory Chip Crunch, Tim Cook Says

• iPhone Prices to Go up as Memory Chip Costs Keep Climbing

• Immutable operating systems are a blessing and a curse

• Apple’s new Mac Mini has fresh M6 and M5 Pro chip offerings — and higher prices

• Apple’s latest Mac Mini runs on a new M6 chip, and starts at $899

In Part 1, we looked at what the Creator Economy is, how it emerged, and why it became a genuine alternative to traditional media.

For many people, the model worked. Independent writers, filmmakers, and countless others built businesses around their work without relying on publishers or broadcasters.

But the story doesn’t end there.

Over the past few years, the Creator Economy has started changing in ways that many people don’t fully notice. Some of those changes come from platforms themselves. Others come from changes in technology and how people discover information online.

In this article, we’re going to look at some of the biggest challenges creators face today, why the economics have become more difficult, and how artificial intelligence is adding another layer to an already changing system.

Quick Note: Part 1 is free and open to everyone, so if you're just getting here, start there. Parts 2 and on are for paid subscribers only.

With that being said, let’s get started.

Why the Model Was Already Under Pressure Before AI Arrived

If you read the previous part, you’ll remember that the Creator Economy wasn’t a perfect system. It gave creators independence from traditional gatekeepers, but it also introduced and increased the dependency on the platform itself.

Creators built audiences on infrastructure they didn’t own, under rules they didn’t control. For years, that tradeoff made sense because the reach was there and, for many creators, the income followed.

The problem is that platforms don’t stay the same.

One of the biggest changes happened gradually, through recommendation algorithms.

Early social media largely showed people content from accounts they chose to follow. Over time, that moved toward recommendation systems designed to maximize engagement instead. Facebook reduced the reach of Pages throughout the 2010s. Instagram replaced its chronological feed with an algorithmic one in 2016. YouTube increasingly relied on recommendations rather than subscriptions to determine what people watched.

Subscribers and followers still mattered, but they no longer guaranteed distribution.

That meant creators could spend years building an audience and still watch their reach fluctuate because of decisions made by a platform.

Jack Conte, the co-founder and CEO of Patreon, has described this dynamic by saying creators are building businesses on land they don’t own. It’s a useful way to think about the relationship. The audience may belong to the creator, but access to that audience is still mediated by the platform.

I’ve also covered this broader pattern in my article on enshittification. The same incentives that gradually change platforms for users also affect the people building businesses on top of them.

Let’s take a look at the major issues in depth to understand what’s really going on.

Issue 1: Platform Revenue Became Less Reliable

For many creators, advertising was never the only source of income. It acted more like a foundation that made everything else easier. Over time, that foundation became less predictable.

Read more

Sometime ago, I was reading a blog post that was related to football. But turns out the topic moved from football into something completely different.

It mentioned Ronaldo’s YouTube channel growth and how quickly it scaled after launch. That comparison then drifted into MrBeast, PewDiePie, and the bigger question of how attention moves once a large audience enters a platform.

A creator who already has global reach joins a platform built around individual discovery. As a result, existing audiences follow, algorithms amplify it further, and smaller creators operating in the same space lose relative visibility.

That led me back to a bigger pattern that has been building for years, the creator economy and how distribution works when attention becomes concentrated.

I had already written about influencer marketing and how trust functions as a form of value online. This is an extension of that thinking, focused less on marketing and more on structure behind that economy.

The goal here is to break down what the creator economy actually is, how it formed, and why the system behaves the way it does.

What Came Before the Creator Economy

Before 2007, independent creators already existed online.

Blogs, early video uploads, and podcasting communities had built audiences without relying on traditional media. Some blogs in the early 2000s reached large readerships, but the systems to support that attention were inconsistent.

At the same time, monetization existed, but it was uneven.

Google AdSense, introduced in 2003, gave publishers a way to display ads on their websites. It worked, but revenue per user was low unless traffic was substantial. Many early creators relied on a combination of small ad income, affiliate links, and direct sponsorship deals, often arranged individually with brands.

Same goes for podcasting: around 2004 and 2005, podcasts started to grow, but most early shows had no built-in monetization. They operated outside of formal revenue systems and people simply produced them because the medium was new.

Around the same time, YouTube got launched, but was acquired by Google in 2006. For its early years, creators uploaded content without expecting direct payment. That changed in 2007 with the YouTube Partner Program, which introduced revenue sharing for creators for the first time at magnitude.

Before that point, creating online content was closer to publishing as a hobby or side activity. Some individuals turned attention into opportunities later through books, consulting, or via some other work, but there was no consistent system connecting audience size to income.

What is the Creator Economy

Where the Term Came From

YouTube started using the term “creator” around 2011 as a replacement for “YouTube Star” It was a way to describe people building large audiences on the platform without relying on traditional celebrity structures.

But “creator” and “creator economy” are not the same idea.

“Creator” referred to the individual, while the “creator economy” refers to something broader, a system where millions of individuals can earn income through content, audience building, and platform-based distribution.

That second shift, from individuals to an economic category, only became widely established years later.

It took hold around 2020, when venture capital firm SignalFire published its Creator Economy Market Map. The report estimated more than 50 million people globally identified as creators and framed the space as a distinct, fast-growing economic category. From there, the term started being used more consistently across platforms and media coverage.

And the reason the term matters at all is because creators didn’t fit into existing categories. They weren’t traditional journalists or media professionals working inside institutions. They were individuals building media businesses around identity, content, and audience, without relying on institutional distribution.

A new category was needed because the old ones didn’t describe what was actually happening.

How Big is the Creator Economy

Estimates of the creator economy vary depending on what is included, but most large-scale analyses place it in the hundreds of billions of dollars.

Goldman Sachs estimated the creator economy at around $250 billion in 2023 and projected it could reach roughly $480 billion by 2027. Their research also suggested there are around $50 million creators globally, with most revenue concentrated in brand partnerships and sponsorships.

But the distribution of that income is uneven.

According to Linktree’s 2022 Creator Report, based on a survey of 9,500 creators, only a small share earn consistent full-time income from their work. Around two-thirds of creators describe it as a side activity. Even among full-time creators, only about 12 percent make more than $50,000 per year.

Separate estimates from Goldman Sachs indicate that only a small fraction of creators, roughly a few percent, earn above six figures annually.

The magnitude of the creator economy is large and the number of participants is even larger, but the income, however, is concentrated at the top.

What Made the Creator Economy Global

Media production used to be tied to geography. Film, publishing, and television were concentrated in a small number of cities, and getting into those industries required institutional access, or personal connections.

That changed once distribution moved online. A smartphone and an internet connection became enough to reach a global audience without institutional approval. Platforms like YouTube, TikTok, and Instagram removed the need for traditional intermediaries, while tools like Substack enabled direct publishing and monetization across borders.

Creators in South Asia, Southeast Asia, Africa, and Latin America started building audiences directly through these platforms. In many of these regions, mobile phones were the primary way people accessed the internet, which shaped both how content was produced and how it was consumed.

Distribution alone doesn’t explain the shift, though. The cost of creation collapsed at the same time. In the early 2000s, producing video or audio required expensive equipment, editing software, and technical expertise, which limited who could participate at scale.

By the mid-2010s, smartphones could record and edit high-quality video on their own due to the launch of free software products, which played a massive role in lowering the barrier for creators to work.

Internet infrastructure expanded alongside these tools, mobile data got cheaper, and upload speeds improved. In many regions, the smartphone became the primary gateway to the internet, reinforcing the shift toward mobile-first content creation.

In short, geography stopped restricting who could reach an audience. Cost stopped restricting who could produce something worth watching. Technology removed both barriers at once, and that combination is what actually made the creator economy possible.

How Removing the Middleman Created Independence for Creators

Before the creator economy, making content professionally meant working through gatekeepers.

Labels, publishers, studios, and networks controlled things such as distribution, funding, and access to audiences. Getting through them took a mix of talent, timing, and connections that most people never had.

In exchange, gatekeepers provided real infrastructure. Production budgets, marketing, legal support, advance payments. But they also kept ownership and distribution rights. Creators supplied the work and kept a fraction of the value it generated.

The creator economy removed that layer. For the first time at scale, creators could produce content without institutional backing, distribute it directly, and build an audience without anyone else deciding who got access.

That’s what independence actually means here. Not complete freedom, but the removal of institutional permission as a requirement for getting your work in front of people and keeping what it earns.

Creators vs Influencers vs Celebrities: What’s the Difference

Before we proceed further, it’s good to clear a few misconceptions about some terms that get used interchangeably, when in reality, they are totally different.

Celebrities are known for work outside platforms. Their audience exists because of offline recognition that later moved online, so their reach doesn’t depend on any single platform. Dwayne Johnson also known as “The Rock” is a good example of this.

MrBeast is also another example of Celebrity, though he started out as a creator and built his audience through the internet.

Influencers build audiences to shape behavior. Their value gets measured in attention and conversion, with brand deals and affiliate marketing at the center of the model. Casey Neistat is a good example of this.

Creators, on the other hand, build an audience around consistent output, and the audience follows the work. Joe Fazer is a good example of this.

I broke all of this down in more detail in the Business of Trust, if you want the full picture of how that economy works:

How Platforms Made the Creator Economy Possible

What Platforms Actually Gave Creators

Platforms provided four core functions that individual creators could not easily build on their own.

Distribution meant access to audiences without having to build traffic from scratch. Outside platforms, this required websites, mailing lists, and tons of other things. Platforms solved this by aggregating users in one place and routing content through existing attention flows.

Discovery came from recommendation systems. Content could now reach people who had no prior connection to the creator. In traditional media, discovery depended on publishers, editors, or programming decisions. On platforms, it was increasingly driven by algorithmic matching between content and viewer behavior.

Monetization infrastructure removed the need for external setup. Payment processing, ad systems, and subscription tools were built into the platform layer. YouTube’s Partner Program, for example, standardized revenue sharing so creators received a fixed percentage of ad revenue generated on their content.

Trust was the fourth layer. Content hosted on established platforms carried more default legitimacy than content hosted on independent sites. This reduced friction for audiences engaging with unknown creators.

The Platforms That Built the Foundation

YouTube introduced its Partner Program in May 2007, initially as an invite-only system for creators who already had significant audiences. It was the first structured link between independent content creation and platform-based income.

YouTube has since paid more than 70 billion dollars to creators, media companies, and music partners over a three-year period ending in 2023. Today, over 3 million channels participate in the program.

Patreon launched in May 2013, founded by musician Jack Conte and developer Sam Yam after Conte struggled to generate sustainable income from widely viewed online content. It introduced a direct support model where creators earn income from recurring contributions instead of platform-driven advertising alone. Revenue mainly depends on subscribers, which separates creator income from algorithmic performance.

@Substack launched in 2017 and extended the subscription model to written content. The platform is still growing as of 2026, as writers and journalists are moving out of traditional media institutions and into direct publishing.

In short, each platform solved a different part of the same problem. YouTube focused on ad-driven scale. Patreon focused on direct audience funding, while Substack focused on subscription-based writing.

How Creators Make Money

Advertising and Brand Deals

Platform advertising works through CPM, cost per thousand impressions. Creators earn a share of what advertisers pay the platform based on how many views their content generates.

CPM rates vary widely depending on niche. Finance and business content tend to sit in the $15 to $25 dollar range per thousand views because advertisers value those audiences more. Gaming content is typically lower, often around 2 to 5 dollars per thousand views. This difference is one of the main reasons niche selection has direct economic consequences beyond audience growth.

Brand deals operate separately from platform advertising. In this case, a brand pays a creator directly to feature or integrate a product. Pricing depends on audience size, engagement rate, and relevance to the product being advertised.

The relationship between audience size and revenue is not linear. A creator with 100,000 highly engaged followers in a specific niche can often command higher rates than a creator with one million followers in a general category with lower engagement.

Affiliate marketing sits between these two models. Creators share tracked links and earn a commission when purchases are made. According to Linktree’s 2023 Creator Report, around 25 percent of creators report affiliate marketing as a meaningful income stream.

You can read the full influencer marketing article here for a deeper breakdown of brand deal mechanics and affiliate disclosure:

Direct Support From Audiences

Direct support refers to income that comes directly from audiences without advertisers in between. Platforms like Patreon, Substack, YouTube Memberships, Ko-fi, and similar services enable this model.

Suppose a creator with 10,000 subscribers paying $5/month generates $50,000 in monthly revenue. That income is not tied to ad rates or algorithmic reach.

This model stabilizes income because revenue depends on subscriptions. And the other good thing about this model is that, it allows creators to produce primarily for paying audiences rather than advertiser, which affects what gets made and how consistently it is sustained.

Kevin Kelly described the underlying logic in his 2008 essay “1,000 True Fans,” published on his blog and later referenced in Tim Ferriss’ Tools of Titans. He believed that a creator does not need mass audiences to sustain a living. A relatively small group of highly committed supporters can generate stable income over time.

Products and Experiences

Products, courses, merchandise, digital downloads, consulting, workshops, and live events form a revenue layer that operates outside platform monetization systems.

Creators typically sell through third-party infrastructure like Shopify or Gumroad, or through their own websites. According to Linktree’s 2022 report, around 41 percent of creators report physical or digital products as a meaningful revenue source.

This is also one of the most independent revenue streams, but product-based income depends primarily on the direct relationship between creator and audience. If that relationship is strong enough, it persists regardless of platform changes. For many established creators, content becomes the top layer of a funnel that leads into direct sales rather than being the primary product itself.

The Reality of Making Money as a Creator

The creator economy has always been structurally unequal. Understanding this matters before Part 2 introduces AI as an accelerating factor because the model was already imperfect long before AI arrived.

Income distribution in the space follows a winner-take-most structure. Reports found that only 12 percent of full-time creators earn above $50,000 per year. Goldman Sachs estimates that only around 4 percent of global creators, roughly two million out of fifty million, earn above $100,000 annually. Between those two ends sits a large group of creators with audiences, consistency, and engagement, but without stable income.

Every monetization path in the system rewards scale in some form. CPM advertising pays per view, which naturally favors high-volume channels. Brand deals scale with reach and perceived influence. Recommendation systems tend to amplify content that already performs well, making initial traction harder for smaller creators. According to data, it takes creators an average of 6.5 months to earn their first dollar, and more than ten months to become self-sustaining. Brand partnerships often appear much later, typically around the two-year mark.

Now, don’t get me wrong, the model still works, but only under specific conditions.

A niche newsletter with 20,000 subscribers charging $8/month flips the entire game. If even 10 percent convert to paid subscriptions, that produces roughly $16,000 in monthly revenue. Add a few brand deals per quarter and occasional digital products or courses, and the result is a sustainable six-figure operation without institutional backing. The key variable is not scale alone, but audience specificity and trust built over time.

Linktree’s findings support this pattern as well. Niche creators are significantly more likely to reach six-figure income levels than broad-appeal creators.

Why?

Because a creator focused on “cybersecurity for small businesses” or “personal finance for freelancers in Southeast Asia” is not competing in a saturated attention field. The audience is smaller, but more intent-driven.

The niche, trust-based model is the exception, and it’s out of reach for most creators. It requires years of consistency, a specific enough audience, and enough trust that people convert into paying subscribers. Very few get there and the conditions that make even this narrow path work are exactly what AI is disrupting, and about to disrupt even further.

What’s Changing and What’s Coming Next

The creator economy was imperfect from the start. Income distribution was uneven, platform dependence was built in, and the creator middle class always existed in a fragile position between scale and stability. Still, it worked well enough to become a viable alternative to traditional media careers.

That balance changed as the system matured.

Platforms moved from chronological feeds to algorithmic ranking systems that prioritize predicted engagement. That change weakened the direct connection between creators and the audiences they had built over time.

At the same time, ad markets became more crowded as supply of content increased faster than demand for attention which led to lower returns per view across many categories.

Then the cost structure of content creation collapsed further. Tools that separated professional production from amateur output became widely accessible. The barrier to producing “good enough” content stopped acting as a filter.

AI now adds a new layer to everything that is happening in the creator space. It changes how content is produced, and more importantly, how much of it exists.

Part 2 of this series looks at what happens when those pressures compound, why the economics get harder, and how AI is adding another layer to an already changing system.

What You Might Like

• Prediction Markets - Part 1: Betting on Reality

• Prediction Markets - Part 2: When Odds Become the News

• Why Europe Is Quietly Ditching Microsoft

• The Infrastructure of Surveillance

You probably know about graphics cards or GPUs from that nerdy friend who owns a gaming PC.

But most people do not realize that every time you use ChatGPT, Claude, or any AI chatbot, you are using a GPU.

Actually, you are using thousands of them!

Those AI models are not running on regular computer processors. They are running on massive clusters of graphics cards in data centers.

The GPU in a gaming PC and the GPU powering ChatGPT are fundamentally the same technology. Just scaled up to big levels.

Everyone and their dog is talking about AI these days.

So, I thought it would be a nice idea to explore how GPUs came to the scene and why AI needs them.

What Even Is a GPU?

GPU stands for Graphics Processing Unit. Many call them graphics cards or video cards.

A GPU is a specialized computer chip designed to handle graphics. It looks pretty similar to a CPU but it’s quite different from it.

A GPU processes video and image data on your screen. Its primary purpose is to turn data into pixels on your screen that you can see.

An important thing about GPUs is that they are really good at doing lots of simple calculations at the same time.

If you’ve read our explainer article on CPUs, you know that CPUs have cores. A modern one typically has around 8 to 24 cores.

However, a modern GPU has thousands of cores. These cores are not very smart on their own. They are not even as broad in what they can do as a CPU core.

But because GPUs have so many of these cores, when combined they can do massive amounts of specific work in parallel.

Compare that to a CPU or Central Processing Unit. Your CPU has maybe 8 to 24 cores.

But each core is very smart and can do basically any calculation you do on it like watching videos, running spreadsheets, all that stuff.

A CPU is like having a few really smart people working on a problem. A GPU is like having thousands of average people working on the same problem at once.

For certain tasks the GPU approach is way faster and rendering a video game is one of those tasks. You need to calculate the color and position of millions of pixels 60 times per second.

That is a lot of simple repetitive math and it’s perfect for a GPU.

Your CPU could do it but it would be slow. Your GPU can do it instantly because it has thousands of cores all working in parallel.

This is why gaming PCs have dedicated graphics cards. The GPU handles all the graphics work while the CPU handles everything else.

How Are CPUs and GPUs Different?

Let me explain this with an example.

Imagine you need to paint a huge wall. You have two options.

Option one: hire one really skilled painter. This person is fast and precise. They can handle complex patterns and details. But there is only one of them.

Option two: hire 1000 average painters. They are not as skilled individually. But there are so many of them that they can cover the wall much faster.

The skilled painter is your CPU. The 1000 average painters are your GPU.

For complex tasks that require lots of decision making the CPU is better. Things like running different programs, browsing the web, doing spreadsheets and stuff.

For simple repetitive tasks that can be done in parallel the GPU is better. Graphics, video encoding, and, as it turns out, AI.

CPUs are optimized for serial processing. One task after another very quickly. GPUs are optimized for parallel processing doing thousands of tasks at the same time.

Intel’s Core i9 has 24 cores. Nvidia’s RTX 4090 has 16,384 CUDA cores. That is why GPUs are so much faster for specific workloads.

A Little History of Graphics Cards

GPUs became mainstream because of video games.

In the 1990s video games which were initially pretty simple started getting more complex. 3D graphics were becoming standard.

But CPUs were not great at rendering 3D graphics. They could do it but it was pretty slow.

So companies started making dedicated graphics cards with their own processors. These cards could handle all the graphics work while the CPU focused on everything else.

3dfx released the Voodoo Graphics card in 1996. It was the first consumer 3D graphics accelerator that really worked well.

Suddenly games like Quake and Unreal Tournament looked amazing. 3D graphics went from choppy and ugly to smooth and detailed.

Nvidia was founded in 1993. They released the RIVA 128 in 1997 and the TNT in 1998. These cards competed with 3dfx.

Then in 1999 Nvidia released the GeForce 256. This was the first GPU. Nvidia literally invented the term GPU with this card.

The GeForce 256 could do hardware transform and lighting. This offloaded even more work from the CPU to the graphics card.

Gaming GPUs got faster every year. More cores, higher clock speeds, better memory.

ATI competed with Nvidia through the 2000s with their Radeon cards. AMD bought ATI in 2006.

By the late 2000s gaming GPUs were extremely powerful parallel processors. Thousands of cores capable of trillions of calculations per second.

But researchers realized something.

Those same parallel processing capabilities that we use for games are also perfect for other forms of computing.

Then AI Happened

We can call 2012 the first major breakthrough in the current AI work that’s being done in front of us.

A team at the University of Toronto led by Geoffrey Hinton used GPUs to train a neural network called AlexNet in 2012.

AlexNet won the ImageNet competition that year by a huge margin. It was way more accurate than previous image recognition systems.

The key was using GPUs. Hinton’s team trained AlexNet on two Nvidia GTX 580 gaming GPUs.

Training neural networks (think of these as the backbone for ChatGPT) requires doing the same math operations just millions of times. Matrix multiplications mostly.

CPUs can do matrix multiplication but it is slow. GPUs can do it way faster because they have thousands of cores working in parallel.

AlexNet proved that GPUs could accelerate AI training by 10x to 100x compared to CPUs.

This kicked off the modern AI boom.

Researchers started using gaming GPUs for AI research. Nvidia GTX Titan, GTX 1080 Ti, RTX 2080 Ti became standard in AI labs.

Then in 2016 Google announced they built custom AI chips called TPUs. These were designed specifically for neural networks.

Nvidia responded by focusing more on AI. They released data center GPUs specifically for AI workloads.

The V100 in 2017. The A100 in 2020. The H100 in 2022.

These were not gaming cards. They were designed for training and running AI models in data centers.

The H100 costs around $30,000. Compare that to a gaming GPU which costs $500 to $2,000.

But data centers needed them and AI companies like OpenAI bought thousands of H100s.

When ChatGPT launched in November 2022, it ran on thousands of Nvidia A100 GPUs. Later OpenAI upgraded to H100s.

Nvidia’s current Blackwell chips have shipped in the millions, many as rack scale systems of 72 GPUs each.

Nvidia’s data center revenue was $4.3 billion in 2020. By 2025 it hit over $100 billion.

Almost all of that growth came from AI companies buying GPUs.

Why Claude Needs Thousands of GPUs

You might be wondering why an AI chatbot needs thousands of expensive GPUs.

The answer is that AI models like GPT-4 or Claude are pretty large. You get why we call them Large Language Models?

GPT-4 reportedly has around 1.7 trillion parameters. Parameters are the numbers that the model learns during training.

All those parameters have to fit in memory. A single GPU does not have enough memory for a model that large.

So AI companies split the model across many GPUs. Each GPU holds a piece of the model.

When you ask ChatGPT a question the work gets distributed across dozens or hundreds of GPUs working together.

This is called model parallelism. The model is too big for one chip so you split it across many chips.

There is also another reason. Inference.

Inference is the process of using a trained model to generate responses. When you send a message to ChatGPT that is inference.

ChatGPT has millions of users. At any given moment thousands of people are asking it questions.

Each question requires running the model. Each run takes a certain amount of GPU compute.

If you want to serve millions of users you need thousands of GPUs running inference simultaneously.

OpenAI reportedly has over 25,000 Nvidia GPUs running ChatGPT. Some estimates put it even higher.

Meta has over 100,000 H100 GPUs for training Llama models.

Google, Microsoft, Amazon all have massive GPU clusters for their AI services.

In 2023 and 2024 there was a global shortage of Nvidia H100 GPUs. Demand was so high that companies waited months for deliveries.

Jensen Huang (Nvidia’s CEO) joked that H100s were the new currency. Some companies bought H100s just to have them as assets.

This is why Nvidia’s market cap is sitting at more than $5 trillion as of writing this article. They have over 90% market share in AI GPUs!

AMD and Intel are trying to compete but Nvidia is untouchable in their software ecosystem which has been a goldmine for them.

Everyone’s talking about Nvidia these days but do you know that Nvidia doesn’t make their GPUs?

In fact their GPUs are manufactured by TSMC, a company so important that we may have a world war over it :/

You can check out my linked article here to read on TSMC’s technology and how it’s powering Nvidia, AMD and most of the tech world.

CUDA and Nvidia’s Trillions

Nvidia did not set out to dominate AI. It started out as a GPU company just trying to survive for the next 30 days.

They became a gaming GPU company by years of hard work. AI, however was more of a side experiment for them to begin.

The irony is that Nvidia’s current AI revenue is so large that earnings from consumer GPUs seems like a rounding error in their financial reports.

Nvidia’s gaming revenue in 2025 was over $11 billion. Their data center AI revenue was over 100 billion.

AI is roughly 10x bigger than gaming for Nvidia.

But this happened because Nvidia made smart decisions pretty early on.

In 2006, Nvidia released CUDA. This was a programming platform that let developers use GPUs for general computing not just graphics.

Nvidia dropped more than a billion dollars into CUDA back in 2006. I covered that investment in quite detail in my Nvidia deep-dive article linked here.

At that time, most people thought CUDA was a waste of money. Why would anyone use a graphics card for non-graphics work?

But Nvidia kept investing in CUDA. They made it easy for researchers to write GPU-accelerated code.

When researchers started building neural networks they reached for Nvidia GPUs because CUDA made it easy.

By the time AI took off in the 2010s Nvidia had a 10 year head start on GPU computing software.

AMD had GPUs too. But their software stack was way behind Nvidia’s.

Intel tried to enter the GPU market with Xeon Phi and later discrete GPUs. But they could not compete with Nvidia’s ecosystem.

Every AI framework was built on CUDA. PyTorch, TensorFlow, JAX all assumed Nvidia GPUs.

Universities taught CUDA. Researchers published CUDA code. The entire AI ecosystem locked into Nvidia.

This is why Nvidia dominates AI hardware. Not because their chips are vastly better but because their software ecosystem is unmatched.

AMD’s MI300 GPU is competitive with Nvidia’s H100 on raw specs. But porting code from CUDA to AMD’s ROCm is painful.

That is why most people just stick to writing their code in CUDA and paying Nvidia’s premium.

GPUs Are Everywhere

You may be surprised but every time you use an AI chatbot you are using GPUs.

When you ask ChatGPT to write an email you are using Nvidia GPUs in OpenAI’s data center.

When you are planning your next startup with Claude (you know who you are :p), you are using Anthropic’s GPU clusters with thousands of Nvidia GPUs.

These GPUs cost millions of dollars to buy and millions more to operate. This is why some AI services charge good money.

There is a reason why basic subscriptions to Claude and ChatGPT cost more than $20. Running GPUs isn’t a poor man’s play.

Google’s Gemini models are different though because they are optimized to run efficiently on their TPUs.

What in the devil is a TPU you may ask?

Well, It’s too long of a topic to cover in this article. Let’s stick to GPUs for this one and if you guys want it, I can write a separate article on TPUs.

Most people think GPUs are just for gaming. And it’s not wrong. But in 2026 GPUs are everywhere.

Your phone has one. Your laptop has one. Even your car probably has one. And more importantly every AI service you use runs on GPUs in a data center somewhere.

Even when Netflix recommends shows or Spotify suggests music, those suggestions come from AI models running on GPUs.

Your bank uses GPUs to detect fraud. Your email uses GPUs to filter spam. Your phone uses GPUs to process photos.

GPUs have been powering most forms of AI for quite some time. We just started paying attention now.

Your Phone Has One Too

Every modern smartphone has a GPU built into the processor.

Apple’s iPhones have them. Snapdragon powered Android phones have them too.

These are not as powerful as desktop or data center GPUs. But they are still GPUs doing the same parallel processing.

When you play a mobile game the GPU renders the graphics.

When you take a photo the GPU processes the image. HDR, night mode, portrait mode all use GPU compute.

When you use Face ID or face unlock the GPU helps process the camera feed.

Some AI features on phones run locally on the GPU.

Apple’s on-device Siri processing uses the Neural Processing Unit (NPU) which is basically a specialized AI GPU.

If you’re scratching your head over what’s an NPU, my friend dove deep into Neural Processing Units in his article linked here.

Google’s Pixel phones use Tensor chips with dedicated AI cores for features like Magic Eraser and Live Translate.

Samsung’s Galaxy phones use Exynos or Snapdragon chips both of which have GPU and NPU cores for AI.

The difference between phone GPUs and data center GPUs is just scale. Your phone GPU might have 6 cores. A data center GPU has 16,000 cores!

Who’s Coming for GPUs?

It doesn’t matter how much Jensen wants to sell you on his GPUs.

The truth is that GPUs are not the final answer for AI. They are just the best option we have right now.

And many companies are trying their best to change that.

Google is already using their custom chips called Tensor Processing Units (TPUs). They use it for training Gemini and doing most of their AI stuff.

Other companies are building custom AI chips too. Amazon has Trainium and Inferentia chips. Microsoft is working on their own AI accelerators.

And dozens of startups like Cerebras, Graphcore, and many others are building even more specialized AI chips.

All of them want a share from Nvidia’s complete dominance in AI.

It’s true that none of them have matched Nvidia’s combination of performance and compatibility ecosystem yet.

But that doesn’t mean it’s not possible. The next generation of GPUs is already in development.

Nvidia’s B100 and B200 GPUs were launched in 2024. AMD launched its MI350 in 2025 and just followed it with the MI400 series in mid-2026, its strongest challenge to Blackwell yet.

Intel’s also working on their Gaudi 3 AI accelerator.

So, for now Nvidia is winning. CUDA is too strong.

Every AI framework is basically written for CUDA. And almost everyone working on AI knows CUDA. That’s simply why people keep using it.

But that could change if companies like Google or Amazon can prove that custom chips are way better.

For us as end users, we may not be able to get our hands on these enterprise GPUs anytime soon (maybe never).

But it is cool to know that whenever you ask ChatGPT a question or generate an image, a GPU somewhere is doing the work.

In about 15 years, GPUs went from a specialized gaming chip to the foundation of AI.

Who knows what Nvidia will be like 10 years from now? CUDA may or may not be relevant.

But companies trying to compete is a good sign because we just don’t need more monopolies.

I hope this helped you understand what GPUs actually do and why they matter so much in the AI world.

If you enjoyed the read, a share to someone as curious as you would make my day.

Share

Similar Articles from SK NEXUS

Further Learning

• CUDA Proves Nvidia Is a Software Company

• What’s the Difference Between a CPU and a GPU?

• ChatGPT’s Hunger for Energy Could Trigger a GPU Revolution

• How Nvidia became the driving force behind the AI revolution

• The Intelligent Industrial Revolution

• AI development won’t hit a wall anytime soon—here’s why

• Why do Nvidia’s chips dominate the AI market?

• Nvidia bets on new data center chips for growth

• AMD MI300X performance compared with Nvidia H100

• Cerebras: What You Need To Know About The Nvidia Competitor

• Google Details Tensor Chip Powers

• Newest Google and Nvidia Chips Speed AI Training

For quite some time now, I’ve been noticing the same story appear over and over again.

Someone is using a chatbot for emotional support. Someone else is asking it for relationship advice. Another person is treating it like a therapist or simply someone to talk to.

The names and the chatbot change, but the motive remains largely the same.

Modern AI systems are available 24/7, respond almost instantly, remember previous conversations, and can adapt to the way a person communicates. Compared to many human interactions, they often feel effortless.

The problem is that these systems are optimized to feel supportive, and engaging. Their goal is to keep conversations going and leave users satisfied with the interaction. That is very different from being truthful or capable of recognizing when a user is heading toward an unhealthy conclusion.

Many people use AI without any serious issues, but as these systems become more personal, researchers have started paying closer attention to a growing set of concerns involving emotional attachment, excessive trust, validation loops, and in more extreme cases, what some people have started referring to as AI psychosis.

So today, I want to break down what’s actually happening, why people form emotional attachments to AI systems, how dependence can develop over time, and what you can actually do about it.

The Growing Dependence on AI

When ChatGPT launched in late 2022, people viewed AI as a productivity tool and at most, it felt like an advanced search engine or a more capable digital assistant.

Over the following years, that relationship has evolved a lot because of these models becoming more conversational and capable.

Researchers have been studying this change for years.

Several studies in human-computer interaction have found that people can develop emotional bonds with AI systems when it display characteristics associated with social interaction, such as memory, responsiveness, and personalization. The more human-like a system appears, the more likely users are to engage with it socially rather than purely as a tool.

And for the first time in history, hundreds of millions of people have access to systems capable of personalized conversations. That creates entirely new social dynamics that simply didn’t exist when most of the digital assistants were limited to simple commands and scripted responses.

Why People Treat AI Like People

A major reason for this comes down to anthropomorphism, the tendency to assign human characteristics to non-human things.

Humans have been doing this for a long time. People name their cars, they talk to pets as if they understand every word. They sometimes get angry at computers as though the machine is intentionally causing problems.

AI just happens to be one of the most convincing examples of this tendency.

When something communicates naturally and responds in ways that feel thoughtful, the brain often begins processing the interaction differently. Even when users know they are talking to software, many of the same social and emotional mechanisms involved in human conversation can still activate.

Researchers at Stanford and MIT have repeatedly observed that people often respond to conversational systems in surprisingly human ways, especially when interactions become frequent and personalized.

And unlike human relationships, AI comes with very few practical limitations.

It doesn’t get tired. It doesn’t need emotional support in return. It’s available whenever the user decides to open the application.

For someone looking for reassurance or simply a conversation, that can be the best thing in life because you don’t have to deal with many of the complications that naturally exist in human relationships.

How Attachment with AI Develops

There are a lot of people who use AI and will never develop a strong emotional attachment to it. But for some users, the process can happen gradually enough that it barely feels noticeable.

Researchers studying companion AI systems have found that repeated interaction often increases feelings of trust and emotional connection. The effect becomes stronger when users disclose personal information and receive responses that feel empathetic.

Over time, as these interactions become more frequent, some users begin preferring AI conversations because they feel easier than human ones. Human conversations involve emotional complexity, and occasionally uncomfortable truths. AI conversations are often more predictable by comparison.

And while this might sound unusual at first, examples of it are becoming increasingly visible. In 2025, People reported on a woman who held a wedding ceremony with an AI-generated partner. The AI companion was displayed on a phone screen, while augmented reality glasses were used during the ceremony and ring exchange.

With the passage of time, this can create a dependency loop.

Read more

In the first article of this series, we covered how silicon goes from literal sand to flat wafers.

In the second article, we covered how those wafers get turned into chips with billions of transistors through processes like lithography, etching and doping.

We ended with wafers covered in billions of transistors all connected by microscopic wires.

But those chips are still attached to a giant wafer.

And these chips have not been tested and need to be cut down to size. You cannot just stick a wafer into your phone, right?

So, as the third and last article in this series, this piece will cover the final stages of chip making.

In this article I’ll go over how chips get tested and how they get cut apart and packaged small enough to fit into our smartphones.

If this is your first time reading this article, just know that this is the third in a series of 3 articles on chip making.

It’s best that you go through them in order to know what we’re talking about.

Part 1 is free and open to everyone, so if you are just getting here, start there. Parts 2 and on are for paid subscribers. Both are linked below:

Most Chips Don’t Work on the First Try

Look, we got through all the processes of creating chips on the wafer but not every chip on a wafer actually works.

In fact many of them do not work at all. They have defects. Transistors that short circuit. Connections that break. Errors that make them useless.

This happens even at the best fabs in the world. Even at TSMC with their cutting-edge processes.

Remember from Part 2 that the chip manufacturing process involves 50 to 100 layers. Each layer requires lithography, deposition, etching, doping.

Every single one of those steps has to be perfect. If any step has even a tiny error the chip fails.

A single speck of dust landing on the wafer during processing can ruin multiple chips. A slight misalignment during lithography ruins chips.

The smaller the transistors the more likely defects become. At 3nm node sizes you are working with features just a few atoms wide.

At that scale everything matters. Temperature variations. Vibrations. Contamination.

So chip manufacturers assume from the start that some percentage of chips will be defective. The question is how many.

This percentage is called yield. And yield determines everything about chip pricing and profitability.

But before we talk about yield we need to understand how fabs figure out which chips work and which do not.

Read more

Just yesterday, I took a walk with my brother to a big park that is about a kilometer or two from our house.

It had rained early in the morning so the weather was chill after quite a few days of 40 degrees here in Lahore.

While walking towards the park we noticed the roofs of plazas, offices and homes.

The surprising thing was that many of them were now full of solar panels that weren’t there a couple years back.

Mall Road is a place we visit quite frequently, but these solar panels are a new phenomenon.

Some roofs were so packed with the panels that my brother joked:

Look, there is a building under those solar panels :p

It’s not like we hadn’t seen solar panels.

But seeing them basically become a roof in many of the buildings is something new.

Even the light posts at the park had small solar panels strapped onto them.

I’m not sure about other neighborhoods. But where I live, people are trying their best to go solar.

Just a stroll around the place tells you much about it.

For a while now, I’ve wanted to explore how solar panels work. The question for me was: how could they generate usable electricity out of just sunlight?

Turns out, the answer is pretty simple to grasp and quite fun too.

Because, the history of solar panel technology literally sounds like science fiction.

So, instead of taking my 5th bath in this scorching summer, I’ve taken to writing a dedicated piece on how solar panels work :p.

What Is a Solar Panel?

A solar panel is a sandwich of materials that converts light into electricity.

The main ingredient is silicon, which is the same material used to make computer chips.

Silicon is a semiconductor. It can conduct electricity sometimes and block it other times depending on how you treat it.

This property makes silicon perfect for solar cells. You can engineer it to react to light in specific ways.

A solar cell is the basic unit of a solar panel. It is a thin wafer of silicon treated with different chemicals.

When light hits the silicon it knocks electrons loose. Those loose electrons create an electric current which is electricity.

One solar cell produces about 0.5 volts. Not enough to power anything useful but combined together they add up.

A typical solar panel can have 60 to 72 cells in it.

Those cells are sandwiched between protective layers. Glass on the front to let light through. A backing material on the rear for protection. An aluminum frame to hold it all together.

The result is a durable panel. One that can sit on your roof for decades converting sunlight to electricity every day.

The Science Behind Solar Panels

The concept behind solar panels is the photovoltaic effect. Photo means light. Voltaic means electricity. Light creating electricity is the basic idea.

You know that silicon atoms have electrons orbiting around them. These electrons are normally stuck in place.

But when light hits the silicon the photons in the light transfer energy to the electrons. This energy is enough to knock the electrons loose from their atoms.

Loose electrons can move around. Moving electrons are electric current.

So light hitting silicon creates moving electrons which is electricity.

That is the photovoltaic effect. And it was discovered in 1839 by a French physicist named Edmond Becquerel.

Becquerel was the first to observe that light could produce electricity. But it took over 100 years before anyone built a practical solar cell.

The first working solar cell was made at Bell Labs in 1954. It was only 4-6% efficient. It cost $300 per watt which was absurdly expensive.

For comparison modern solar panels cost less than $0.10-$0.30 per watt.

This is why solar is finally becoming mainstream. It got cheap enough to compete with fossil fuels which is where most of our energy comes from.

Silicon Powers All Tech

Silicon is everywhere in technology. Your phone chip is silicon. Your computer processor is silicon. And your solar panel is silicon.

But solar silicon is different from chip silicon.

Solar panels use what is called solar grade silicon.
This silicon is 99.9999% pure. Six nines.

That sounds very pure. And it is.
But chip silicon is even purer.

Chip silicon is 99.9999999% pure. Those are nine nines!!
The silicon has to be this pure or transistors do not work.

Solar silicon doesn’t need to be quite as pure as chip silicon, but it’s still quite pure by any normal standard.

The difference comes down to how transistors and solar cells each react to imperfections in the material.

This is why solar panels are cheaper than computer chips even though they use more silicon.

The manufacturing process is also different.

For chips you grow a perfect single silicon crystal and slice it into ultra-thin wafers. Then you build billions of transistors on each wafer.

For solar panels you either grow a crystal or just melt silicon and pour it into molds. Then you slice it into thicker wafers and add chemicals to create the photovoltaic effect.

Solar wafers are much simpler than chip wafers. They don’t need precision for billions of transistors.

This is why solar got cheap while chips stayed expensive. Solar manufacturing is simpler.

Most solar panels are made in China. Chinese companies like Longi, JA Solar, and Trina Solar dominate the market.

They have massive factories that produce panels at huge scale. This drives down costs.

In 2023, China produced about 80% of the world’s solar panels. The US and Europe are trying to build their own solar manufacturing but they are way behind.

For Pakistan, this means solar panels are relatively cheap to import. The technology is mature and competition keeps prices low.

The AC/DC Problem

This is quite a technical topic and I won’t turn this article into a physics paper.

But what you need to know is that there are two kinds of electricity.

AC, which stands for Alternating Current and DC, which stands for Direct Current.

Solar panels produce direct current or DC electricity. The same type of electricity from batteries.

But your home runs on alternating current or AC electricity. This is what comes from the grid.

So you need a device to convert DC from the panels into AC for your home. That device is called an inverter.

The inverter is arguably more important than the panels themselves. A good inverter can last 10 to 15 years. A bad one fails in 3 to 5 years.

When people complain about solar systems breaking, it is usually the inverter that failed, not the panels.

There are many types of inverters for households. But three common ones are:

1. On-grid Inverter

2. Off-grid Inverter

3. Hybrid Inverter

On-Grid Inverter

An on-grid inverter connects your solar system directly to the electricity grid. During the day your panels generate power, you use what you need, and the extra goes back to the grid.

Your meter basically runs backwards. At night, you pull from the grid like normal. The big advantage is cost. On-grid systems are the cheapest to install because you do not need batteries.

The weakness is obvious though. No grid, no power. If WAPDA cuts the supply, your solar system shuts off too. Even if the sun is shining.

This makes it a frustrating option for us Pakistanis. Because load shedding is the main reason people go solar here.

Off-Grid Inverter

An off-grid inverter has nothing to do with the grid at all. Your panels charge a battery bank and the inverter runs your home from those batteries. Complete independence.

Off-grid inverters are perfect for areas where there is no grid connection or where load shedding is so bad the grid is basically useless anyway. The downside is cost and complexity.

You need enough battery capacity to survive cloudy days and nights. That adds up fast. And batteries wear out. Most people in cities with at least some grid access find this setup overkill and expensive to maintain.

Hybrid Inverter

A hybrid inverter is the most popular option these days. Because the government doesn’t like net metering now :/.

A hybrid inverter basically does everything. It connects to the grid, manages your battery bank, and optimizes between all three sources: solar panels, batteries, and grid.

During the day your panels charge the batteries and power your home. At night the batteries take over. The grid only kicks in when both run low. During load shedding your system keeps running without interruption.

It is the most expensive of the three but also the most practical for a country where the grid exists but cannot be trusted.

Sunlight intensity isn’t consistent and it changes throughout the day.

Morning light is weaker than noon light which means less electricity generation in the morning. The same is true around sunset.

A good inverter tracks these changes and optimizes constantly. That is why a good inverter makes a huge difference in how much power your system generates.

Batteries are ever more important for us from our homes to smartphones. You can go through my article linked here which talks about the innovations happening around battery tech.

Why Solar Panels Keep Getting Cheaper?

In 1977 solar panels cost around $77 per watt. Today solar panels cost around $0.10-$0.30 per watt.

That is a roughly 99% price drop in 46 years.

This is one of the fastest cost declines for any technology ever. Possibly faster than computers, even smartphones.

Though the 99% price drop is cool, we should know that this is just for the solar panels themselves.

Normally you need to invest in a full solar system including an inverter, sometimes a battery and the panels.

Here, the costs do add up but they’re still usually cheaper than what the grid offers you especially in Pakistan.

Another question you guys may be asking: How did solar get so cheap?

First, manufacturing got better. Early solar cells were made by hand in small batches. Now they are mass-produced in gigantic automated factories.

Second, China invested heavily in solar manufacturing. The Chinese government subsidized factories and ramped up production.

This created massive oversupply, which drove prices down. This was bad for solar companies but great for customers.

Third, technology improved. Solar cells got more efficient. Panels produce more power from the same amount of silicon.

Fourth the supply chain matured. Silicon suppliers, panel makers, inverter companies all got better and cheaper.

The result is that solar is now the cheapest form of electricity in many parts of the world.

In sunny countries like ours and parts of the Middle East, new solar power plants are cheaper than coal or natural gas plants.

For home installations the economics depend on local electricity prices.

In Pakistan where electricity costs 30 to 50 rupees per unit, solar does make sense for many.

A household survey conducted by the Pakistan Bureau of Statistics showed that 25% of households now use solar power in some form.

Most of it was rooftop solar installed by homeowners and businesses.

Our electricity grid isn’t one you can rely on. The frequent spikes in voltage can kill your expensive electronics.

I wrote this article to share how you can protect your electronics from random electricity shocks that come from the grid.

Pushing Past Physics

Solar panels are pretty good now but there’s still a long journey to go. Because most commercial panels convert only 20% to 24% of sunlight into electricity.

The rest is wasted as heat. You might think this is bad. Why not make them more efficient?

The answer is that efficiency has diminishing returns. Going from 20% to 25% efficiency is very expensive.

It requires better silicon, better manufacturing, better materials. All of which cost money.

For home applications it is cheaper to just buy more panels. Why pay extra for 25% efficient panels when you can buy more 20% panels for the same price?

The exception is space. Satellites have limited surface area. Every square meter counts. So space solar panels use the most efficient cells available.

Solar cells used on satellites can hit ~34% efficiency. But they cost thousands of dollars per panel.

For rooftop solar, 20% efficiency is fine. You just need enough roof space.

There is also a theoretical limit. Silicon solar cells cannot exceed around 33% efficiency. This is called the Shockley-Queisser limit.

It is a physics limitation. Silicon can only absorb certain wavelengths of light. The rest passes through or reflects.

To beat this limit you need different materials. That is why researchers are working on perovskite and multi-junction cells.

But for now silicon at 20% efficiency is good enough for most uses.

The Next Decade of Solar Tech

Solar panels today are pretty good. But researchers are not satisfied with just that.

The silicon solar cell has been around since 1954. For 70 years we have been improving the same basic technology and we are now approaching its limits.

So scientists are working on what comes next. In the solar space, the biggest buzz right now is around perovskite solar cells.

Perovskite is a material that can absorb light very efficiently.

In the lab perovskite cells have already crossed 26% efficiency. Some experimental versions have hit 35% when layered with silicon.

That is a big deal because silicon alone cannot exceed 33% efficiency due to physical limitations of the material.

Perovskite layered on top of silicon lets you capture parts of the light spectrum that silicon misses. You are essentially getting two solar cells in one.

The problem is durability. Perovskite breaks down when exposed to moisture and heat. A rooftop panel needs to survive 20-25 years because that’s how long standard silicon panels last.

Current perovskite cells do not last anywhere near that long.

However, researchers are working on it. Some companies are already commercializing early versions.

Give it a couple of years and perovskite-silicon panels could be on roofs at a reasonable price.

The more immediate future is about better batteries and smarter systems. Panels themselves are already cheap enough.

The bottleneck is storage. As battery costs fall we’re going to see more solar.

This situation is going to change with sodium-ion and solid-state batteries. Both the US and China are heavily investing in them.

Both superpowers understand the value of harnessing renewable energy.

I think we will see batteries become more common for new solar installations.

The on-grid solar setup was the first step. But these days net-metering doesn’t make sense for new connections anymore. Affordable home batteries are the next step.

If you think about it, the panels on roofs today are the worst solar panels we will ever make. Everything from here gets better.

Solar Is (Almost) Everywhere Now

Solar panels cost money but the sunlight is free.

Without us knowing, the sun dumps an insane amount of energy on Earth every day.

In one hour the sun delivers enough energy to power all of human civilization for an entire year!! Just think about this for a minute.

The problem is that we do not know how to capture most of the sunlight and turn it to useful energy efficiently.

Solar panels are our best attempt so far.

Solar panels in markets today can produce up to 600 watts of electricity during peak hours.

That does not sound like much. But put 15-20 panels on your roof and you can power a whole average household.

There is caveat for us in Pakistan though. A decent solar system for an average home costs around a million rupees depending on size.

That is pretty expensive upfront. It can be 10% of the cost of an entire house here especially when you add batteries.

But solar panels are not just on the roofs of our houses. They are pretty much everywhere.

Traffic lights in remote areas use solar panels. Parking meters and street lights increasingly use them too. The panel sits on top and charges a battery during the day.

Calculators have used solar cells for decades. You probably owned one in school without thinking about it.

Satellites in space run on solar. The International Space Station has huge solar arrays.

In rural areas, the electricity grid is even worse than in big cities. Solar panels are life changing there. Even a small solar setup can power lights, TV, fans and charge phones.

Solar panels have been around for quite a long time but the tech is as accessible as ever.

The panels have become so good that we’re hitting limits of the current tech. And the next evolution of solar tech is already in labs today.

The work on affordable battery tech like sodium-ion is only getting more interesting this time around.

If you feel you enjoyed reading through this article. You can subscribe to SK NEXUS using the button:

Subscribe now

More From SK NEXUS

Further Learning

• How do solar panels work? - Richard Komp

• How solar panels work - simple explanation

• The photovoltaic effect explained

• The History Of Solar - DOE

• History of solar cells from 1839 to today

• The Solar Efficiency Gap

• How Does a Solar Inverter Work

• Diverse Innovations led to dramatically cheaper solar panels

• The cost of solar power: how low can we go?

• Perovskite: The ‘wonder material’ that could transform solar

• Pakistan’s Continued and Resilient Solar Growth

• Tracking China’s clean energy export dominance in seven charts

• Solar Energy Isn’t Always as Green as You Think

• The Perfect Storm Fueling Pakistan’s Solar Boom

• How Pakistan’s solar boom is shielding it from worst of Iran war crisis

• Solar without net metering

• Solar photovoltaic module price

One of the side effects of running SK NEXUS is that I spend a lot of time keeping up with what’s happening across the technology world.

Tech companies launch new products, buy other companies, and expand into new industries. Then there are the topics like surveillance, data collection, digital identity systems, and the growing relationship between governments and technology companies.

Over the past couple of months, I found myself reading article after article trying to understand how all of these pieces fit together. Though, eventually, I realized they were all pointing toward the same story.

Many of us think surveillance is something only governments do, but that’s only one side of the coin.

Surveillance technology infrastructure is sometimes developed by private sector companies, implemented by the government, and fueled by data that billions of people generate every single day.

The items such as phones, smartwatches, and others were not designed for surveillance purposes only. The main idea of creating such devices is convenience, however, it provides a tremendous opportunity to track human behavior in real time.

And that’s really what this article is about; aka understanding how technology companies and governments gradually converged, how the infrastructure surrounding us evolved, and why surveillance today looks very different from what it did even twenty years ago.

The Foundations of Modern Surveillance

The modern surveillance ecosystem began with the internet, not with facial recognition systems, or government monitoring programs.

The web, which started becoming mainstream during the 1990s, was a much simpler place. Back then, people viewed the internet as a tool they occasionally used rather than a system woven into every aspect of daily life.

That started changing during the early 2000s when many new technologies, platforms and eventually smartphones became central to how people communicated and consumed information, due to which companies such as Google, Meta, and Amazon started building ecosystems that encouraged users to spend more time inside their services.

This gave rise to what author Shoshana Zuboff later described as surveillance capitalism. The idea that companies collect large amounts of behavioral data and use those insights to predict future actions, improve products, personalize experiences, and generate advertising revenue.

This was made possible through the introduction of the smartphone which travels everywhere with its user. It is worth noting that smartphones feature numerous pre-installed services like GPS sensors, cameras, cloud computing and many others.

This allows firms to develop user profiles on a scale that would have never been possible before.

The wider public got a glimpse into the implications of this model in 2013 when former NSA contractor Edward Snowden leaked classified documents detailing surveillance programs such as PRISM. The disclosures revealed how intelligence agencies were able to obtain access to large amounts of digital information, which triggered a global debate around government surveillance, and corporate data collection.

The Convergence of Big Tech and Government

Historically, governments would construct most of the capability needed for their intelligence and surveillance work internally. If governments needed any information, they would depend upon intelligence organizations and huge-scale investigations. Information gathering would be costly and often require a lot of manpower.

In today’s world, technology companies gather an incredible amount of information just by delivering their services. Search engines deal with billions of searches. Smartphones are constantly recording location data. Social media sites record interactions and relationships.

All of this information is gathered not because of intelligence needs. Nevertheless, quite a bit of it may carry intelligence significance.

Thus, governments do not have to construct all capabilities from scratch anymore. They become increasingly dependent on various technology vendors, who already have all the necessary skills to handle the information they have.

One of the places where this dependency shows up quite visibly is cloud computing.

Over the last decade, governments have moved large portions of their digital infrastructure onto commercial cloud platforms. According to Amazon, AWS GovCloud was specifically developed to help public-sector organizations handle sensitive workloads while meeting government compliance requirements.

Microsoft, likewise, offers a similar platform called Azure Government, which the company describes as a dedicated cloud environment designed for U.S. government agencies and their partners.

Another example of commercialization in information security is artificial intelligence.

Training sophisticated AI systems involves massive computing resources as well as access to the necessary hardware facilities. Governments are now increasingly choosing to collaborate with companies that already have these resources.

A recent example came in May 2026 when the Pentagon (the U.S. Department of Defense) announced agreements with eight major AI and technology companies. These companies included OpenAI, Google, Microsoft, Amazon Web Services, Oracle, NVIDIA, SpaceX, and Reflection AI. According to the Pentagon, the goal was to accelerate the military’s transition toward becoming an “AI-first fighting force” and improve decision-making capabilities across multiple domains of warfare.

The announcement illustrates how concentrated much of the world’s advanced AI infrastructure has become. Many of the most capable models, cloud platforms, and computing resources are owned by a relatively small number of private companies.

The Pentagon’s announcement also came shortly after a public dispute with Anthropic. As per the reporting from The Guardian, Anthropic declined to allow unrestricted military use of its technology, while the other participating companies agreed to permit their systems to be used for “any lawful purpose” under the Department of Defense’s framework.

One of the key aspects that should be noted, none of what we’ve discussed above is an indication of a covert surveillance program on its own. What we can say for sure is that the cooperation between the government and tech corporations has become more and more interwoven.

The Companies Building the Surveillance Layer

Google

Undoubtedly, Google is one of the most interesting cases because most people use Google products on a daily basis.

For example, Chrome is still the world’s leading web browser, Android is installed on billions of smartphones, Google Maps assists in navigation of cities, and Google Search processes billions of searches per day. As was reported by Demandsage, Chrome is still dominating its segment on a global level.

Individually, each of these services solves very useful problems for users. However, taken together, these services offer tremendous amount of information about people’s behavior without even mentioning Google’s growing AI system.

I’ve actually explored Google’s data collection practices in much greater detail elsewhere in my Privacy Series, particularly in the article focused on browsers and web tracking. So I won’t go too deep into that rabbit hole here.

Nonetheless, the main reason Google appears in discussions around surveillance is due to the sheer scale it operates at.

That was made clear through the Project Maven incident, which was an effort by the Department of Defense of the United States that utilized AI technology to analyze the footage from drones. According to reporting from multiple outlets, the involvement of Google in this particular project faced strong opposition from its employees. The opposition led Google to distance itself from the project.

Throughout the time, Google has continued expanding its government and defense-related work through cloud infrastructure, AI services, and public-sector partnerships. More recently, the company reached an agreement with the Pentagon involving advanced AI capabilities and defense applications.

As mentioned in the article from cdomagazine, the agreement gives Google’s AI the ability to perform tasks for any “lawful government purpose.” This is indeed a very broad scope of the deal because it does not restrict the technology to performing administrative tasks or working in the field of cybersecurity.

For years, the debate on military technology has not been based on the presence of the technological advancement but on the usage. Intelligence gathering, surveillance operations, battlefield planning, and weapons support systems can all fall under lawful military activities depending on the jurisdiction and rules of engagement involved.

Palantir

Palantir was founded in 2003 by Peter Thiel and several co-founders. Governments and large organizations already collect enormous amounts of information, but turning that information into something useful is often much harder than collecting it in the first place, Palantir took advantage of this.

Read more

A year ago we wrote about Linux and how it was slowly entering into mainstream conversations. If you want to check that out, that article is linked here.

At the time it felt like a trend. A few influencers switched.
A few governments started paying attention.
Steam Deck was making gaming on Linux actually work.

We asked whether this was finally Linux’s moment.

Well, it is now August 2026. A lot has happened. So let’s check in.

The Steam Machine just launched. SteamOS is now officially open for anyone to install on their own PC.

The creator of Ruby on Rails is moving parts of his company to Linux.

Linus Sebastian from Linus Tech Tips ran a 30-day Linux challenge with his team.

And Steve Burke from GamersNexus is openly talking about Linux gaming on his channel.

None of these are small on their own but together they point to a real shift.

Steam Machine

Valve launched the Steam Machine this June.

It is a compact gaming PC the size of a small cube that runs SteamOS and sits under your TV.

Starting price is $1,049 which is more expensive than most people expected. The original internal target was closer to $750 but the global RAM shortage pushed costs up significantly before launch.

Still, the Steam Machine sold out in under 10 minutes despite the price.

That tells you something. There is real appetite for a gaming device that is not a PlayStation or Xbox. And one that runs on Linux.

On the same day Valve announced the Steam Machine, they also confirmed that SteamOS 3.8 is now officially available for anyone to install on most AMD or Intel-powered PCs.

You do not need to buy the Steam Machine to run SteamOS. You can download the image, flash it to a USB drive, and install it like regular Windows.

Valve engineer Pierre-Loup Griffais told The Verge that “starting with the SteamOS 3.8 release, you can put together your own Steam Machine using whatever PC parts you want.”

This means people with compatible PCs can make a Steam Machine at home.

If you are so excited to make your own Steam Machine that you can’t wait to read this article in full, you can check out my SteamOS install guide linked here.

The Steam Ecosystem

SteamOS is built on top of Arch Linux. It uses something called the Proton compatibility layer to run Windows games on Linux without asking game devs to do anything special.

The result is that the vast majority of your Steam library just works on SteamOS. Games you already own. You don’t need to buy anything new.

The Steam Deck proved this concept worked. The Steam Machine is the living room version of the same idea.

And now SteamOS being available for anyone to install extends that even further. For Linux broadly this matters a lot.

Gaming was always the biggest hole in the argument for switching to Linux.

People were willing to accept a worse browsing experience or different office apps. But losing access to their game library was a dealbreaker.

Valve spent years closing that gap. Now it is mostly closed for most games.

There are still anti-cheat complications. Games like Valorant and the Call of Duty franchise use kernel-level anti-cheat systems that require Windows.

For now, those will not work on SteamOS. So competitive multiplayer is still a weak spot.

But for single-player games, indie games, and most of the broader library, SteamOS works really well.

As a long-time Linux user, I’m consistently surprised by how many of my games straight up just work on Steam.

Did you know Valve just announced a new VR headset?

Valve’s new VR headset is called the Steam Frame. It’s going to be one of the best gaming headsets around, because it packs some genuinely unique features.

I covered the Steam Frame’s specs and features in detail linked here.

Valve Fixing Nvidia and AMD for All

One of the bigger announcements around the SteamOS 3.8 launch was the Nvidia situation.

Historically SteamOS and Linux in general have had poor Nvidia support.

So much so that the founder of Linux (Linus Torvalds) had this to say about Nvidia in 2012:

“Nvidia has been the single worst company we’ve ever dealt with”

Most Linux gaming setups defaulted to AMD graphics for this reason, because AMD’s drivers are open source and baked into the Linux kernel already.

Griffais confirmed that Valve now has a “growing team” working directly with Nvidia to bring proper driver support to SteamOS.

The goal is that Nvidia GPU owners will eventually be able to run SteamOS on their existing hardware.

The realistic timeline is that this is not happening in 2026. Most reporting puts realistic Nvidia support sometime in 2027.

The technical challenge is that SteamOS uses an immutable file system, meaning the drive is read-only to protect users from borking their install.

And Nvidia’s proprietary drivers do not slot into that immutable setup easily. Valve and Nvidia are working through it, but it takes time.

For AMD the story is better right now.

SteamOS 3.8 ships with AMD graphics drivers and Linux kernel 6.16 which includes solid support for AMD’s RDNA 2, RDNA 3, and RDNA 4 graphics cards.

So if you have an AMD GPU, you can install SteamOS on your existing PC today and get the Valve experience.

Intel support also landed in 3.8 which expands the pool further.

If you have an Nvidia GPU, you are waiting. But at least Valve is now officially working on it with Nvidia directly.

Because Valve has opened SteamOS to the general public, I installed it myself and made a detailed guide on how to install Steam OS.

The guide has real screenshots too so you can easily follow along at home. The complete install guide is linked here.

The LTT Linux Challenge

Linus Sebastian of Linus Tech Tips ran a 30-day Linux challenge in 2026 with his team.

This is actually the second time LTT has done something like this. The last major one was in 2021.

The 2026 version involved multiple staff members each picking a different Linux flavor and trying to use it as their primary operating system for a month.

The results were mixed in a way that was very useful for understanding where Linux actually stands.

Some staff had relatively smooth experiences. Gaming worked well, particularly through SteamOS and Proton.

Basic productivity tasks were fine. The interfaces on modern distributions have gotten pretty good.

Other staff hit walls. Video codec issues came up. Network share access was more complicated than it needed to be.

Application compatibility for certain professional tools was still a problem. Installing software in ways that make sense to Windows users is still not obvious on Linux.

One memorable moment involved Linus himself struggling with the choice of distribution at the start.

The Linux community has hundreds of versions and there is no obvious single answer for a newcomer.

That fragmentation is real and it matters.

What the challenge revealed is that Linux in 2026 is genuinely usable for a lot of people.

But the definition of “a lot of people” still has boundaries.

If you mainly browse the web, use streaming services, play games through Steam, and write documents, Linux will almost certainly work for you.

If you depend on specific professional software, particularly in creative fields, you will still run into problems.

The reaction from the Linux community to the LTT challenge was interesting too. Parts of the community were helpful and constructive. Other parts were defensive or dismissive when problems were raised.

This is a known pattern and it was called out directly in the series.

What Regular People Struggle With Moving to Linux?

The LTT challenge made this very visible but it is worth talking about the actual friction points that still exist on Linux.

Software compatibility is still the biggest one.

Microsoft Office has no Linux version. Adobe has no Linux version. If your job depends on these tools you are stuck.

LibreOffice is a usable alternative for many tasks but it is not the same and anyone who works with shared Office files regularly will notice the gaps.

Anti-cheat systems block competitive gaming. This is not Linux’s fault but it is a real limitation.

Installation and updates work differently on Linux.

Linux uses package managers and app stores that work well once you understand them but feel foreign at first.

Installing software the way you would on Windows, by downloading a .exe file from a website, is not how most Linux software works.

Hardware compatibility has improved enormously but is not perfect. Most modern hardware works fine.

But niche peripherals, certain printers, and some laptops with unusual hardware configurations can still cause headaches.

And the distro choice problem is real. Someone new to Linux has to pick a version before they even start, and there is no clear best answer that works for everyone.

Ubuntu is the safe starting point for most people but even that requires a learning curve.

None of these are unsolvable. Many of them have gotten significantly better in the last three years.

But they are real and pretending otherwise does not help anyone.

Omarchy, DHH and the Return of Hacker Culture

This is the part of the Linux story in 2026 that is really surprising.

David Heinemeier Hansson, usually known as DHH, is the creator of Ruby on Rails and co-owner of 37signals, the company behind Basecamp and HEY.

He is a well-known programmer in the web development space.

In 2025, DHH released Omarchy.

It is an opinionated configuration of Arch Linux combined with a tiling window manager called Hyprland. Just think of Omarchy as another flavor of Linux.

DHH describes it as his personal ideal developer environment bottled up and made available for others.

He then went further and announced that 37signals is moving its entire ops and Ruby programming teams to Linux via Omarchy over the next three years.

He said that by 2028 the vast majority of the company’s programmers and all of its ops team should be running Linux.

This is important because DHH is not a teenager experimenting with an operating system.

He is a professional in his forties who previously used Mac, ran his company on Apple hardware, and now is moving away from it.

His stated frustrations with Mac included the direction Apple has taken with hardware and software.

He was frustrated by the increasing integration of services he did not want, and a broader feeling of losing control of his own machine. Sound familiar?

Omarchy delivered a petabyte of ISO downloads in a couple of months.

That is a real number. People are not just reading about this, they are installing it.

DHH also organized what he called Omacon in New York in April 2026.

It is a gathering specifically around Linux and Omarchy. The idea was to celebrate a return to hacker culture.

It was a celebration for people who like to tinker with and own their tools rather than rent a managed experience from a tech giant.

The concept lives.

A common thread runs through all of this, through Valve opening SteamOS, through DHH building Omarchy, through people getting frustrated with Windows and Mac subscription models and telemetry and AI features added without asking.

People want to own their computers again.

Why Important People Are Moving to Linux?

When a guy with 110 million subscribers says something, it becomes important. And that’s what happened last year with Felix Kjellberg aka PewDiePie on YouTube.

PewDiePie has been vocal about using Linux and literally named the video:

“I installed Linux (so should you)”

And the overall pattern among those who joined Linux is worth noticing.

DHH left Mac for Linux and is in the process of moving his company over to it.

Linus from LTT spent a month on it publicly.

Steve Burke from GamersNexus has been discussing Linux gaming seriously and benchmarking Linux performance explicitly on his channel.

Linux didn’t get infinitely better overnight.
The alternatives just got worse for more people.

Linux kept on improving its interface gradually and it’s in a usable state for many people.

Windows 11 has been as unpopular as ever, thanks to the AI features, telemetry, and bloat.

And since Windows 10 support ended last year, there’s been a proper campaign by Linux distros welcoming people off it.

Microsoft Recall, the feature that automatically screenshots everything you do, damaged trust badly even after it was made opt-in.

Mac hardware remains excellent but Apple’s increasing control over what can run on Apple Silicon and the subscription creep in Apple services remain.

The general sense of being managed rather than owning your machine has pushed some people to reconsider.

Linux had a usability floor that stopped most people from seriously considering it. That floor has been rising steadily.

Somewhere in the combination of Proton fixing gaming, desktop environments maturing, better hardware support, and tools like Omarchy, enough technical people started to feel this is a genuine cultural moment.

What This Means for Everyday Users?

Let’s be direct about who this actually matters for right now.

If you are someone who mainly uses their computer for browsing, watching content, messaging, and gaming through Steam, Linux is a real option today.

SteamOS and many others like Bazzite have made this accessible in a way it was not before.

If you are a developer who works in a terminal environment and does not need specific Windows or Mac software Linux has been the power user’s choice for years.

Omarchy is the latest version of an argument that has existed for a long time: that a well-configured Linux setup can be better for certain kinds of work.

If you depend on Adobe Creative Cloud, Microsoft Office in deep ways, or specific professional software that has no Linux version, Linux is not your daily driver yet. The gap is real.

The middle group is the interesting one. People who do not fall neatly into either of those categories. Regular users who do some gaming, some productivity, some creative work.

For them the answer in 2026 is probably that Linux is getting close but still requires tolerance and time to fix bugs and issues that they may face.

Is Linux Finally Ready for Everyone?

As someone who daily drives Linux I feel that Linux is ready for more people than it has ever been ready for. But that doesn’t mean everyone.

Applications are getting better. More tools are working on Linux. The GUI has matured too.

Steam and Proton have solved the single biggest adoption barrier for gamers. Companies like Valve are actively pushing compatibility further.

But it is not ready for everyone I think.

The Adobe wall is real. Most kernel-level anti-cheats still don’t work on Linux.

And the learning curve is still there. Just the sheer number of Linux flavors confuses newcomers.

What has changed is the size of the group for whom Linux does work. That group is growing. As things stand, that group is ~4% of the global desktop market share.

And just so you know, that number was around 1-2% for decades.

And for the first time in a long time the people joining that group include influential voices who are bringing attention and energy along with them.

Linus from LTT. DHH. Steve Burke exploring Linux gaming performance on GamersNexus. PewDiePie picking Hyprland for his setup.

The old converts arrived as developers and never left. This wave is coming from mainstream tech culture and finding a reason to stay.

Year of the Linux Desktop is getting closer little by little and I’m as excited as I could be.

It’s ready for more people than ever, including some you wouldn’t have expected, even if it’s not ready for everyone yet.

That is different from where we were a year ago. And it is worth paying attention to.

I hope this helped you understand where Linux stands in 2026.

If you enjoyed reading, sharing it with someone curious would be appreciated.

Share

Further Learning

• I installed Linux (so should you)

• Switching to Linux Challenge!

• I Don’t Think I Can Go Back To Windows...

• Valve says you can now build your own Steam Machine with SteamOS

• LTT Linux Challenge 2026 Part 1 - Reacting to it - TuxDigital

• LTT Linux Challenge 2026 forum discussion

• What is Omarchy Linux, and why is 37signals moving to it

• Omarchy 3.7 Linux Distro Launches as a Gaming Edition

• I struggled with Hyprland for months, until this Arch-based distro fixed it

• Omarchy is out

• DHH on what Omarchy was built to do

• Steve Burke of GamersNexus on Linux gaming

• RIP Windows: Linux GPU Gaming Benchmarks on Bazzite

• Thoughts on Omarchy

• The Steam Machine is the most ambitious game console I’ve ever played

• Nvidia has been the single worst company we’ve ever dealt with

In our previous article in this series, we went through the whole process of mining and purifying silicon to 99.9999999 percent, and all that it takes to create silicon wafers.

Those wafers are perfectly pure, flat, and fully blank. They don’t have any circuits or transistors yet.

Fresh Silicon wafers are somewhat like blank papers. They need someone to write on them to be useful for us in the form of chips.

This article is about exactly that. In this piece we’ll discuss how those blank wafers get turned into chips with billions of working transistors.

The process involves some of the most expensive and precise machines ever built. Light with wavelengths smaller than viruses. Chemistry happening at atomic scales.

And all of it has to be repeated 50 to 100 times to build a single chip.

I’ll try my best to give a full picture of the main processes without drowning you in technical details.

However, you shouldn’t expect this article to be like the regular ones we do here at SK NEXUS.

Saqib here, with one quick side track before you read on.

This week SK NEXUS got its first ever paid subscriber. We went paid only a few weeks ago, and it felt like a gamble. So having someone read our work and decide it’s worth paying for means a lot. To everyone who considered it, and to the one who took the leap, thank you. Really!

Silicon Wafers Arrive at Fab

The blank silicon wafers we covered in our previous article in this series get shipped to a fab.

If you’re unsure about a fab, you may want to check out the previous article in this series.

The fab could be Taiwan’s TSMC, Korea’s Samsung or America’s Intel. These are three of the largest fabs in the world.

These fabs are some of the most expensive buildings ever constructed. A single modern fab costs 20 billion dollars or more to build.

Inside these fabs the air is cleaner than an operating room. A single speck of dust can ruin a chip so the air gets filtered all the time.

Workers wear full body suits called bunny suits. These suits prevent skin cells, hair, or clothing fibers from contaminating the wafers. Just imagine how clean these rooms would be!

The silicon wafers arrive at the fabs in special containers that keep them clean. They get unpacked in cleanrooms where the air is filtered thousands of times per hour.

The real work begins once the wafers land at the fab because fabs are responsible for building billions of transistors on each wafer which are what make a chip.

But before I start nerding about transistors let’s spend a little time trying to understand what a transistor actually is.

What Is a Transistor?

A transistor is actually a switch. It can be on or off. Its purpose is to conduct electricity or block it at any given time.

By turning these switches on and off in specific patterns you get computers to do the work they do.

A modern chip that is in your phone or computer is made of billions of transistors.

And even if you somehow open up your chip, you can’t see transistors with the naked eye. That’s how small they are!

It’s beyond this article to fully cover how transistors allow us to make computers. Maybe, we’ll go through it in a separate article in the future.

On average, a modern phone chip has around 20 billion transistors while your high-end laptop CPU has around 50 billion transistors. All working together to run your apps and games.

Read more

Are you a gamer who just can’t keep up with Microsoft?
Are you someone who hates Windows 11 with a passion and wants to game on something different?

If you are one of those people then cheer up because Substack gods have taken you to the perfect article.

You’d know that Valve finally launched the Steam Machine for the price of $1,049 for the base 512GB model. Yes, I get that it’s expensive and I know a lot of people who are not happy about that price.

Early predictions had this thing landing somewhere around $600 to $800. Instead we got a price that is closer to a mid-range gaming laptop.

If you go over online coverage of the Steam Machine, it looks like the whole world is burning over it.

But here’s what nobody is talking about.

With the launch of Steam Machine, Valve opened up SteamOS so anyone can install it on PC with compatible hardware.

And that too for free. There is no waiting list or price tag for that.

So, I followed Valve’s instructions and installed SteamOS on my own laptop a few days ago.

And this article is me detailing that process so you guys could follow it too.

This article will walk you through the exact steps that I went through. I have shared screenshots too, so you can try SteamOS by following this piece.

The Steam Machine Is Expensive

Quick context before we move on, because I know some of you are coming in hot.

Valve officially released the Steam Machine on June 22, 2026. The entry price of $1,049 for the 512GB model was notably higher.

Most people were guessing it would land around the $600 to $800 range.

The 2TB model runs $1,349.

Add the new Steam Controller and you are looking at $1,130 for the base bundle or $1,428 for the loaded one.

The reason is not some greedy cash grab but the reality of making hardware products in 2026.

Valve engineer Pierre-Loup Griffais confirmed the team had been tracking a lower price target back when they first started working on the Steam Machine in 2023.

A global RAM shortage wrecked those plans. Globally RAM prices went over 400%. That is 4 times!!

AI data centers are eating up the same memory chips that go into your gaming PC, your phone, and your laptop.

Three companies (Samsung, SK Hynix, Micron) produce roughly 95 percent of the world’s RAM.

All three companies shifted production toward high-margin AI chips, leaving consumer hardware fighting over scraps.

Valve even considered dropping the Steam Machine from dual-channel to single-channel RAM just to stretch the supply they could actually get.

So yes, $1,049 stings. But it is not really a Valve problem. It is everyone’s problem.

PlayStation, Nintendo, Xbox, even Apple, all of them have raised prices for their hardware.

Steam Machine is expensive largely because of the global RAM shortage. I covered this shortage in our dedicated article linked here.

We discussed how RAM quadrupled in price in a span of a year and how every computer is more expensive because of that.

It Is Not Really Valve’s Fault Though

Look, unlike Sony and Microsoft, Valve does not subsidize the Steam Machine.

Consoles are usually sold at a loss or near break-even because money comes back in the form of game purchases.

Valve cannot do that here. The Steam Machine is fundamentally a PC.

Someone could buy it, wipe SteamOS, install Windows, and never give Valve another dollar through game purchases.

So Valve has to price the hardware to actually cover its costs.

For context, you can build an equivalent or better desktop PC for the price of the base Steam Machine.

But you cannot have that PC remotely turn on your TV or fit silently inside that TV shelf.

This comfort is something that quite a few people are after and it’s their choice if they want to buy the Steam Machine for that price.

Look, I am not here to tell you to stop being annoyed. I am annoyed too.

But there is a silver-lining to all of this that most people are not paying much attention to.

SteamOS Opens Up

Around the Steam Machine launch Valve quietly shipped SteamOS 3.8.10.

Valve engineer Pierre-Loup Griffais told The Verge that starting with this release, you can put together your own Steam Machine using whatever PC parts you want.

Until now SteamOS was basically locked to AMD hardware, because it powers the Steam Deck and the official Steam Machine.

Installing it on anything else required janky workarounds using the Steam Deck recovery image, and the results were inconsistent at best.

SteamOS 3.8 changes that. The update adds proper support for Intel GPUs including the newer Arc B series, alongside fixing bugs that didn’t allow other PCs to boot into SteamOS.

Valve has also changed the graphics stack of the Steam Deck to a more modern package called Wayland. This will provide better switching between Desktop and Gaming Mode.

And one of the biggest surprises is that Nvidia support is actively being worked on too.

Griffais said Valve has “a growing team” collaborating with Nvidia “very closely” on proprietary driver support for GeForce cards.

He was careful not to promise a date, saying support “might not come this year,” with most outlets pegging a realistic target somewhere in 2027.

Nvidia has had a rocky relationship with consumer Linux drivers and Valve is working on changing that.

This means that a large chunk of gamers who boast their GeForce graphics cards can soon get SteamOS working on their hardware too.

Right now if you want the best experience with SteamOS, you should aim for AMD CPUs and GPUs that were released based on the RDNA 2 and RDNA 3 architectures because that is what the Steam Deck and Steam Machine use.

RDNA 4 support is also in there through the updated Linux kernel and Mesa drivers SteamOS 3.8.10 ships with.

If you’re not sure what SteamOS is, here’s a dedicated article we did covering what it is and how it’s causing a revolution in the gaming space.

The article is linked here for you to check out.

Before You Install SteamOS

Look, I get that it’s really exciting to get rid of Windows and try out SteamOS but let me tell you a couple of things to keep in mind before you begin.

1. Installing SteamOS WILL ERASE ALL DATA ON YOUR COMPUTER!!

   It’s best that you don’t use your primary computer and instead use a secondary computer for trying out SteamOS because SteamOS wipes the drives connected to the computer.

2. Be ready to debug for bugs and issues like any other software, SteamOS will have bugs. And it’s up to you to fix them and debug them which can sometimes take a few minutes to many hours. You have to be willing to spend hours browsing through internet forums, Reddit threads or working alongside ChatGPT figuring out why SteamOS won’t install for you.

3. No one is responsible for SteamOS. You can install SteamOS on hardware you own even if it’s not officially supported by Valve. But it also means you are totally responsible for your install. Valve support won’t be there to help you if you’re facing any issue with SteamOS.

There are some other limitations too. Once you install SteamOS, it will replace operating systems like Windows that are already installed on your system. You have to reinstall them again if you want to remove SteamOS from your computer.

Dual-booting alongside Windows is also not officially supported yet, so you should plan to dedicate an entire drive or a computer to SteamOS.

Secure Boot has to be disabled to install SteamOS. You’d find this setting in your motherboard’s BIOS.

I have an older laptop lying around that I do not use for much anymore, so none of that mattered to me. I figured it was the perfect guinea pig.

Look, I spend my days playing around with operating systems and stuff and I understand most people won’t be doing that. Only go ahead with SteamOS if you actually have read through the disclaimers above.

It’s totally fine if you don’t want to spend hours debugging issues on a command line.

What You Need Before We Start

If you’ve decided to go ahead with SteamOS anyway, here’s what you need.

1. You’d need a USB drive that is at least 8GB in size. Though I would recommend 16GB or larger just to be safe. This drive will get completely wiped, so back up anything important on a separate computer first.

2. A PC you are okay with completely wiping. SteamOS does not currently support dual-booting. So, whatever drive you install this on, expect SteamOS to wipe all data that exists on that computer

3. A stable internet connection. The SteamOS updates itself after install and you’d need that to finish initial setup.

4. A separate computer or laptop to create a bootable USB.

Step 1: Download The SteamOS Image

First step is grabbing the actual SteamOS image from Valve from the link below: https://help.steampowered.com/en/faqs/view/65B4-2AA3-5F37-4227#install

On the page above click the link next to the ‘Download the SteamOS image’ button. You will need to agree to the terms before the download button becomes available.

Click through and the download will start. Depending on your connection this can take a few minutes since the image itself sits at a few gigabytes.

Once it finishes downloading, do not touch the file yet. We need balenaEtcher ready first.

Step 2: Flash It With balenaEtcher

If you have not already, install balenaEtcher through the link below:

https://etcher.balena.io/

balenaEtcher is the software that will convert our normal USB drive into a bootable SteamOS USB. Installation depends upon your operating system and is genuinely just clicking through a few prompts.

Once Etcher is open, plug in your USB drive.

Click the “Flash from file” button and navigate to wherever your SteamOS image is downloaded to.

Etcher should automatically detect your USB drive if it is the only one connected.

If you have multiple drives plugged in, double check you are selecting the right one before continuing, since Etcher will completely erase whatever is on it.

Hit the Flash button and wait. This part took me roughly 10 to 15 minutes, though your mileage will vary depending on how fast your USB drive actually is.

One tip: if you are on a slower drive and the estimated time looks insane, a few people in Steam Deck communities have reported that decompressing the image file first before flashing speeds things up dramatically on certain setups.

Worth knowing if Etcher is telling you the flash will take several hours. Once Etcher confirms the flash completed, you are ready to actually boot into this thing.

Only change: the period after “insane” became a comma and “A few” dropped to “a few,” so the tip is one complete thought instead of a dangling “if.”

Step 3: Boot Into The Installer

Plug the freshly flashed USB drive into the PC you are installing SteamOS on.

Restart the machine and get into your boot menu.

This is usually done by spamming a specific key during startup, commonly F12, F2, Esc, or Delete depending on your hardware manufacturer.

If you are not sure, a Google search of your specific laptop or motherboard model plus “boot menu key” will tell you.

Select your USB drive from the boot menu instead of letting it boot normally into Windows or whatever is already installed.

Give it a minute. You should see SteamOS boot into what looks like a stripped down live environment.

This is essentially the same recovery and install environment that Steam Deck owners use, just running on your own hardware now.

From here you will see the install options. This is where things get real.

Step 4: The Actual Install

Once you are in the SteamOS Live Environment just click the button on the desktop that says ‘Wipe Device & Install SteamOS’.

Once you confirm, the installer takes over. There is not much for you to do at this stage besides wait. SteamOS copies itself onto your selected drive and sets up the base system.

This part took me somewhere around 15 minutes on my setup, though again this depends heavily on your drive speed.

When the install finishes, you will be prompted to remove your USB drive and reboot the machine.

Go ahead and do that now. Pull the USB stick out before the reboot actually happens so your machine does not just boot back into the installer again.

Step 5: The Reboot And Welcome Wizard

This is the part that actually feels like magic the first time you see it.

Your machine reboots and instead of the usual Windows logo or whatever you are used to seeing, you get the SteamOS boot animation.

This is the same screen Steam Deck owners see every time they turn their handheld on.

From here SteamOS launches its welcome wizard. This walks you through the basics.

You have to set your language, your timezone and your WiFi network.

I have censored the networks around my area but you’d see the actual names of the networks around you.

The whole wizard felt genuinely smooth on my end, very close to the experience you get on actual Steam Deck.

In the end you need to sign in with your Steam account.

As you sign in you are dropped straight into Gaming Mode.

Your Steam library shows up out of the box, complete with controller support and pre-optimized settings.

There is also a Desktop Mode tucked away if you want to poke around under the hood, install other software, or just see what is going on beneath the gaming interface.

And that is genuinely it. Start to finish, including download time, the whole process should take you less than an hour.

My install took a bit more time but that was more because of my specific hardware that is a bit old so things can get different depending on the hardware you’re installing SteamOS on.

What This Means Going Forward

If you have gone through the steps above you should have a working SteamOS install on your PC.

It is the same OS that the Steam Deck and the Steam Machine use.

Again, the Steam Machine price is annoying and I am not going to pretend otherwise.

But the existence of an open, free, supported SteamOS that runs on hardware you already own is genuinely a bigger deal long term than the Steam Machine itself.

And that’s what I feel is Valve’s goal in all of this. It’s exactly what they’ve said in the Steam Machine launch blog post:

SteamOS is quite new and as people report bugs and Valve fixes them there is an exciting path that one can go to if they’re fed up with what Microsoft has been doing with Windows all the while.

Windows still dominates the desktop market share and it will continue to do so for quite some time.

But it’s really cool that we have an alternative that is slowly and gradually becoming better than ever and all of us can be a part of that.

On Steam, Linux is now more popular than macOS driven largely by SteamOS adoption through the Steam Deck.

And this will only improve through the Steam Machine.

In my opinion, As Valve opens up SteamOS for more platforms and works with partners like Nvidia, it is building something bigger than even the Steam Machine.

If you already have a decent AMD GPU sitting around, or even an Intel Arc card, you do not need to wait or pay a thousand dollars for a Steam Machine

You just need a spare drive, a USB stick, and about an hour of your time (ideally).

So yeah. The Steam Machine price is rough. But Valve also just handed you the actual loophole, and barely anyone is talking about it.

I hope this guide helped you build your own setup if you decided to give it a shot.

Please don’t hesitate to share your experience with SteamOS if you’ve gone through it. Or just share any thoughts related to SteamOS in the comments down below.

Further Learning

• Valve Steam Machine hits $1,049 as the RAM crisis forces a major price rethink

• AI crisis turned Valve’s $750 Steam Machine into a $1,049 one - Memeburn

• Steam Machine price, specs and release date guide 2026

• Valve will finally let you build your own Steam Machine with SteamOS for desktop - The Verge

• Valve now allows you to run SteamOS on your own DIY Steam Machine, with Nvidia GPU support in the works

• SteamOS Nvidia support confirmed: Valve has a growing team but AMD builds work now

• You don’t need a Steam Machine to run SteamOS - Valve opens it up to any AMD PC

• Valve confirms SteamOS is coming to any PC, with Nvidia support in the works

• Steam Machine release date, specs, and price - PCGamesN

• How to use the Steam Deck recovery image - Pi My Life Up

• BalenaEtcher Official site

You might know that countries, especially the US and China, are trying their best to bring chip manufacturing back to them.

TSMC’s Arizona fab is already printing out advanced chips for the likes of Apple and NVIDIA, with 3nm production planned there for 2027-28.

All of that is interesting and exciting but none of that makes sense unless you understand how chips actually get made.

The weird fact is that the processor in your phone, the GPU in your gaming PC and the chip in your car’s computer, they all started as sand on a beach somewhere.

Literal sand. The stuff you walk on. The stuff kids build castles with.

The question is how do you go from literal sand to a complex chip with billions of transistors, where a single transistor is about tens of thousands of times smaller than a human hair.

The answer is a process so complex and expensive that only a handful of companies in the world can actually do it at scale. You can count these companies on one hand.

This is the first article in a Series where we will cover chip manufacturing from sand to the device in your hand.

As you may have figured out, chip manufacturing is one of the most complex processes in human history. There is simply a lot of maths and physics involved.

However, in this Series, I’ll keep most of the stuff relatively simple for you guys while talking about the important stuff.

One quick side track -

Just so you know, SK NEXUS recently went paid. Linked here is the detailed post by that goes into the specifics of it all.

To keep it short, here’s what to expect moving forward:

Our 3-part article Series is going paid. Readers will have to pay a small amount to access our Series.

However, first part of a Series (like the one you’re reading) will be free for all.

Money wasn’t the motivator for going paid. We want quality tech writing to be accessible for most.

That’s why the first part of a Series will remain free, while other parts will be exclusive to those who pay.

If you have any thoughts, please share your feedback down in the comments.

Silicon Is Everywhere But Pure Silicon Is Rare

Silicon is the second most abundant element in Earth’s crust after oxygen. It is everywhere.

Rocks, sand, dirt, glass. All of it contains some form of silicon in them.

But here is the thing. Silicon almost never exists alone in nature. It is always bonded to other elements, usually oxygen.

When silicon bonds with oxygen you get silicon dioxide. That is what sand is. That is what quartz is. That is what most rocks are made of.

The silicon we need for chips is not silicon dioxide. It is pure silicon. Just silicon atoms and nothing else.

Getting pure silicon from sand is like trying to extract gold from seawater. Technically possible but incredibly difficult and expensive.

The purity required for chips is absurd. We are talking 99.9999999 percent pure silicon. Nine nines!

One impurity atom for every billion silicon atoms. That is the standard.

Anything less and the chip does not work properly. Transistors fail and there’s a short circuit.

So the entire chip industry is built on a process that takes one of the most common materials on Earth and refines it into one of the purest substances humans have ever created.

What Makes Silicon Special for Electronics

You might be wondering why we use silicon at all. There are other materials. Why not use something easier to work with?

The answer is that silicon has a unique set of properties that make it perfect for electronics.

First, silicon is a semiconductor (Now you know why we call them semiconductor chips).

Being a semiconductor means it can conduct electricity sometimes and block it other times depending on how you treat it.

Metals like copper conduct electricity all the time. Insulators like rubber block electricity all the time. Silicon sits in the middle.

You can control whether silicon conducts or blocks by adding tiny amounts of other elements to it. This process is called doping and we will cover it in a later article.

Silicon’s ability to switch between conducting and insulating states is what makes transistors work. Transistors are the building blocks of every chip and we’ll go over them in detail too in a later article in this Series.

Second, silicon forms a stable oxide layer when exposed to air. That oxide layer is silicon dioxide which is basically glass.

This matters because you can grow extremely thin, extremely uniform layers of silicon dioxide on top of silicon. These layers act as insulators between different parts of the chip.

Third, silicon is abundant and relatively cheap to source. You can find silicon dioxide anywhere on Earth.

Fourth, we have spent 70 years learning how to work with silicon. The entire manufacturing infrastructure is built around it.

There are other semiconductor materials like gallium arsenide or silicon carbide. Some are faster and even more efficient than silicon.

But none of them have the combination of properties, manufacturing knowledge, and cost-effectiveness that silicon has.

That is why every major chip in your life uses silicon. Your phone, your laptop, your car, your router, all silicon.

Making Pure Silicon from Quartz

We know that we need pure silicon to make chips and in its natural form silicon is bonded with oxygen and we need to remove that bond and extract pure silicon.

That journey starts with quartzite (aka quartz for short). It is a type of rock that is almost pure silicon dioxide.

Mining companies dig up quartz from a mine. They crush it into smaller chunks. Then they load it into massive electric furnaces to melt it.

Inside the furnace the quartz gets heated to over 2000 degrees Celsius (Yup, furnaces are that hot!). Carbon in the form of coal also gets added to the mix.

At those temperatures a chemical reaction happens. The carbon pulls oxygen away from the silicon dioxide leaving behind molten silicon and carbon dioxide gas.

What comes out is called metallurgical-grade silicon or MG-silicon. It is about 98 to 99 percent pure.

That sounds pretty pure. But for electronics it is nowhere near good enough. MG-silicon is fine for making aluminum alloys or solar panels.

But semiconductor chips are much more demanding. They need something far more pure.

The impurities in MG-silicon include things like iron, aluminum, calcium, and other metals. Even at 1 or 2 percent contamination that is billions of unwanted atoms.

Those impurities will ruin a chip. So the next step is to purify the silicon even further.

Getting to 99.9999999% Purity

The industry standard for purifying silicon is called the Siemens process. It was developed in the 1950s and we still use it today because nothing better has been invented.

This is also where things get extra expensive. Here is how we purify silicon further:

You take metallurgical-grade silicon and grind it into powder. Then you react that powder with a special acid to produce a chemical compound that is liquid at room temperature.

If you’re curious, that compound is called Trichlorosilane.

Trichlorosilane can be distilled. Distillation is a purification process where you heat a liquid until it evaporates, then cool the vapor back into liquid.

Impurities have different boiling points than trichlorosilane. So when you distill it multiple times the impurities get left behind.

After enough distillation cycles you end up with ultra-pure trichlorosilane and that’s where things get more and more expensive.

You take thin rods of already-pure silicon and place them inside a reactor chamber. Then you pump ultra-pure trichlorosilane gas into the chamber.

The rods get heated to over 1000 degrees Celsius. At that temperature the trichlorosilane gas decomposes and deposits pure silicon onto the rods.

Layer by layer, atom by atom, the rods grow thicker as silicon builds up on their surface.

This process runs for days. The rods slowly grow from pencil-thin to telephone-pole-thick.

When the process finishes you have polycrystalline silicon rods that are 99.9999999 percent pure. Nine nines.

This is called polysilicon or poly-Si. This is what the chip industry uses.

The Siemens process is slow and energy-intensive. It takes enormous amounts of electricity to run those reactors for days on end.

That is why polysilicon is expensive. And that is why companies that make polysilicon are critical to the entire chip supply chain.

Most of the world’s polysilicon comes from a handful of companies in China, Germany, and the United States.

Making the Silicon Ingot

Polysilicon is pure but it is not in the right form still. Chips need single-crystal silicon not polycrystalline silicon.

The difference is in how the atoms are arranged. In polycrystalline silicon the atoms are arranged in small random crystals all jumbled together.

In single-crystal silicon all the atoms line up in one perfect crystal structure.

Single-crystal silicon has better electrical properties. Transistors work more reliably and the overall chip performance is more predictable.

So the next step is to turn polysilicon into a single crystal. The process used is called the Czochralski method or CZ method.

Here is how it works.

You take chunks of polysilicon and melt them in a crucible at over 1400 degrees Celsius. The silicon turns into a liquid pool.

Then you take a small seed crystal of silicon, a tiny piece that already has the perfect crystal structure, and dip it into the molten silicon.

The seed crystal acts as a template. As you slowly pull it up out of the liquid, silicon atoms from the melt attach to the seed and copy its crystal structure.

You keep pulling slowly while rotating the seed crystal. The silicon keeps freezing onto the seed in perfect alignment with the crystal structure.

Over several hours you pull a massive cylinder of single-crystal silicon out of the melt. This cylinder is called an ingot.

A typical ingot is about 2 meters long and 300 millimeters in diameter. Some newer fabs are moving to 450mm diameter ingots but those are not widespread yet.

The entire ingot is one perfect crystal. Billions of atoms all lined up in the same direction.

This is the raw material that chips are made from. But we are not done yet.

Slicing Silicon Into Wafers

You cannot build chips on a 2-meter-long cylinder of silicon. You need thin flat discs almost like CDs (I hope you’re old enough to remember those).

So the ingot gets sliced into wafers.

First the ingot gets trimmed. The top and bottom get cut off. The sides get ground down to a precise diameter.

Then the ingot goes to a wire saw. This is a machine with a thin wire coated in diamond particles. The wire moves at high speed and slices through the silicon like a slicer cutting meat.

Each slice is incredibly thin. Modern wafers are about 0.75 millimeters thick. Thinner than a credit card. One ingot produces hundreds of wafers.

After slicing the wafers are rough and covered in microscopic damage from the cutting process. So they get polished.

Polishing involves grinding the wafer surface with finer abrasives until it is perfectly smooth and flat.

The final wafer surface is so smooth that if you scaled it up to the size of a big city the tallest bump would be less than a centimeter high.

That level of flatness is required because the circuits built on the wafer are nanometers tall. Any bump or scratch ruins the chip.

Blank silicon wafers ready to be turned into chips. The finished wafers are inspected, cleaned, and packaged for shipment to chip fabs.

Fab stands for Fabrication Facility. These are companies which take the wafers and build the transistors onto each wafer, layer by layer.

TSMC, Samsung, and Intel are some of the most sophisticated fabs out there.

Bigger Wafers = Cheaper Chips

You might have noticed that the industry keeps pushing for bigger wafers. 200mm, 300mm, 450mm.

The reason is pretty simple. Each wafer can hold a certain number of chips depending on the chip size. A bigger wafer holds more chips.

The cost of processing a wafer does not scale with wafer size. Processing a 300mm wafer does not cost twice as much as processing a 200mm wafer even though it has more than twice the area.

So bigger wafers mean more chips per wafer with only a modest increase in processing cost. That drives down the cost per chip.

A 300mm wafer has 2.25 times the area of a 200mm wafer. That means 2.25 times more chips from roughly 1.3 times the processing cost.

The industry has been on 300mm wafers since the early 2000s. The transition to 450mm has been talked about for over a decade but it keeps getting delayed just like GTA 6 :p

The problem is that moving to 450mm requires redesigning every tool in the current fab supply chain. The furnaces, the lithography machines, the deposition tools, everything.

And that is way too expensive and time-consuming.

Just know that the most advanced lithography machines cost around $350 million per unit.

The cost of going to a bigger wafer is in the billions of dollars. And right now the companies can not justify it simply.

So for now 300mm (aka 12 inches) is the standard and will be for the foreseeable future.

What’s Next After Silicon Wafers?

We started with sand on a beach. We ended with a perfectly flat, perfectly pure, single-crystal silicon wafer.

This wafer is the foundation for everything that comes next in the Series.

In the next article we will head into the fab, where those blank wafers get turned into chips with billions of transistors. We will cover lithography (drawing circuits with light), then deposition, etching, and doping, the actual construction of the transistors.

Then in the final article we will cover how chips get tested, cut apart, and packaged, and why some chips cost $50 while others cost $2,000.

But all of that starts here. With sand. The fact that your phone started as sand is not just some analogy I used. It is literally true.

Someone literally mined quartz out of the ground. Someone refined it into polysilicon. Someone grew it into an ingot. And someone sliced it into wafers.

Those wafers traveled to a fab in Taiwan or South Korea where billions of transistors will be built on them.

Then those chips got packaged, tested, and shipped to your phone manufacturer who assembled them into the device you are holding right now.

The entire process from sand to smartphone takes months and involves dozens of companies across multiple countries.

But it all starts with the humble silicon which most people don’t think about when it comes to advanced processors.

I hope this article helped you understand where chips actually come from. The next articles will dive deep into the processes that happen on a wafer.

While writing this article, the exciting thing for me was how simple the chip-making process sounds until you try to dive a bit deep.

Most of us never imagine how complex the process of taking literal sand into polished wafers is.

In the next article, things will get more complex as I go over the transistor side of things.

I would love to know how this newer Series format feels to you guys. Please share any feedback you have down in the comments.

Similar Articles From SK NEXUS

Further Learning

• The humble mineral that transformed the world

• 6 crucial steps in semiconductor manufacturing - ASML

• How Silicon Chips Are Made (Explained in 7 Mins)

• Semiconductors and the Power of Silicon

• A Background to Silicon and its Applications

• From Sand to Silicon “Making of a Chip” - Intel

• Silicon Ingots: The Building Blocks of Modern Electronics

• Silicon Wafer Manufacturing: From Sand to Silicon

• Why 450mm wafers?

• TSMC Expecting to Start 3 nm Production in Arizona

I’ve been wanting to write this piece for a long time.

Every few weeks, another major development would happen somewhere in Europe and I’d end up delaying the article again just to gather more material. But at some point, the pattern became too large to ignore and I got tired of waiting.

None of what you are about to read in this piece happened overnight. The transition has been gradual, but over the last few years, a very noticeable change has started taking shape inside European public infrastructure.

Denmark is moving government staff toward Linux and LibreOffice. Schleswig-Holstein has been replacing large parts of its Microsoft infrastructure across regional government systems. Similarly, France has also accelerated sovereign open-source adoption at a scale very few people were paying attention to until recently.

For years, products from Microsoft were treated almost like default infrastructure inside governments and became deeply embedded into public administration across Europe.

But that premise is now being questioned because governments are increasingly starting to view software dependence as a strategic problem, due to which the question ”Should critical government infrastructure depend almost entirely on foreign corporations?” has become unavoidable.

This article breaks down what’s actually happening across Europe, why it’s happening now, and what it could mean for the future of digital infrastructure.

Before we begin, a quick note. This is the first deep-dive article we are publishing after moving SK NEXUS to a paid model. If you haven’t already, I recommend reading the transition post where explained why we made that decision and what it means for the future of this publication.

Your support allows us to spend more time researching pieces like this, especially topics that require weeks of following developments.

Anyway, let’s get started.

The Broader European Strategy

Before we move on, it’s important to keep in mind that what’s happening across Europe is easy to reduce to “Europe versus Microsoft,” but that misses the bigger picture entirely.

Microsoft is simply the most visible symbol of a much broader transition already happening across Europe.

The movement isn’t limited to replacing Windows with Linux or Office with LibreOffice. Governments are also looking at de-googling initiatives, sovereign cloud infrastructure, local data hosting, open-source software adoption, and reducing dependence on foreign technology vendors more generally.

What’s changing is the way software infrastructure itself is being viewed.

For decades, governments treated software mostly as a procurement problem. You bought products, signed contracts, outsourced maintenance, and moved on.

But it’s not like that anymore - European policymakers are increasingly beginning to treat digital infrastructure the same way they treat energy infrastructure, or defense systems. And quite frankly, the concern isn’t difficult to understand.

If a geopolitical conflict, or legal disagreement suddenly affects access to critical software infrastructure, what happens next? More importantly, what happens to - Government communication systems? Citizen records? Public administration? Healthcare infrastructure?

Those questions become very serious when governments realize how much of their infrastructure depends on external vendors operating under foreign jurisdictions and that is the whole reason this transition matters.

Europe’s Sovereignty Push

Digital sovereignty has gradually become one of the central ideas shaping Europe’s technology strategy.

Put it this way, governments want control over the systems they depend on. That includes software infrastructure, cloud platforms, citizen data, communication systems, and the long-term direction of public-sector technology.

A large portion of global cloud infrastructure is currently dominated by American companies like Amazon Web Services, Microsoft Azure, and Google Cloud. European governments rely heavily on these systems for public services, data storage, communication, and so much more.

You can see how dependent not just Europe, but the entire internet has become on a relatively small number of infrastructure providers. In 2025 alone, multiple major outages involving Cloudflare disrupted huge portions of the web, affecting thousands of services simultaneously and leaving large parts of the internet barely functional for hours.

And once governments realize how much critical infrastructure depends on external providers, concerns around control start becoming much more serious, because at the end of the day dependence is also legal and geopolitical.

One major example is the U.S. CLOUD Act, which allows American authorities to request data from U.S.-based companies even when that data is stored outside the United States. For European governments already focused on privacy and regulatory independence, that creates an uncomfortable reality.

Europe’s GDPR framework addresses privacy protections, but regulation alone does not remove the fundamental dependence if the infrastructure itself remains controlled elsewhere.

And that’s really the main point of the sovereignty discussion. European governments are not trying to isolate themselves from American technology entirely for the sake of nationalism. They just want to ensure that critical public infrastructure is not entirely dependent on decisions made in Silicon Valley, or corporate headquarters outside European jurisdiction.

How Governments Became Dependent

Understanding why this transition is so difficult requires understanding how governments became this dependent in the first place.

Back in the 1990s and early 2000s, Microsoft became the obvious choice for government IT infrastructure across large parts of Europe, because if everyone used the same software stack - it would be easier to cooperate, support would be easy to standardize and on top of that training costs would stay manageable.

At the time, all this made perfect sense, but in reality this was the biggest mistake that the governments made.

Microsoft’s ecosystem is designed to work best within itself. Office formats like .docx, .xlsx, and .pptx became the default format for documents across governments and institutions. Exchange became the standard for enterprise email infrastructure and entire administrative workflows were built around these products.

So, basically governments were replacing decades of workflows, retraining employees, migrating legacy systems, and making sure everything remains compatible with external organizations that still depend on Microsoft products themselves.

Read more

This one’s a little different.
No tech breakdown today.

After more than a year of explaining technology here, and the better part of a decade doing it everywhere else, I want to use this one post to talk about SK NEXUS itself.

Where we are, what’s changing, and why.

If you’re new here, the short version: almost none of this affects you.
Most of what we publish stays free, and you can keep reading exactly as you have been.

If you’ve been with us for a while, this is the one I’d really like you to read to the end.

Some context first; SK NEXUS is old, by internet standards. It started around eight years ago as a blog about tech, and it has lived on Facebook Groups, Medium, Wix, a custom-built site, and now Substack.

The platforms kept changing.
The job never did: explain what tech actually means, not what marketing says it means.

Today it’s three of us. I edit and write, and two contributors research and write alongside everything else going on in their lives.

Between us, this publication takes 70 to 80 hours a month. More than a hundred articles on this site, readers in 65 countries, and every word of it free since day one.

I’m laying that out so the rest of this post has its shape.
This is a small, stubborn thing, older than half the platforms it has lived on, run by people who do it because they love it.

The Work That Gets Scraped

Some of you may be familiar with our Deep Dives.

When we publish one, the kind where we trace a company’s whole arc or take a technology apart piece by piece, that single article can take a month, sometimes more.

Researching it, sourcing it, checking it, rewriting it until it actually makes sense to a normal person.

These are the pieces we’re proudest of: the story of the one Dutch company every advanced chip on earth depends on, the month-long trace of how NVIDIA went from a gaming-chip company to the backbone of AI, the full arc of how computers became the thing in your pocket.

Go read any of them and you’ll find something you won’t get easily on an internet drowning in listicles and hundred-word pages dressed up as articles.

Here’s what happens to that month of work.

An AI scrapes it. The summary gets served to someone who never visits the page. The impressions climb, and almost nobody reads the real thing.

That chip-monopoly piece has been shown in search results more than forty-seven thousand times.

Forty of those turned into actual visits.
Forty.

And I can see the same story in our own logs: the AI engines crawl us constantly, and on a good day they send back one or two visitors. The scraping only goes one way.

And it’s not us having a bad quarter. Independent tracking found that scraping of tech and consumer-electronics content more than doubled over the past year, making it the single most scraped category on the internet.

Our exact lane.

The State of the Feed

There’s a second half to this, and it’s the platform itself.

That month of work goes out into a feed where newsletters publish AI-written filler every single day. Generic, churned out without anyone really thinking it through. And promoted anyway, because the system mostly rewards whoever switched on a paid button and kept posting.

For a while I wondered if I was imagining it.

Then someone measured it. An independent analysis ran AI detection across the top bestselling publications on Substack, and technology came out as the most AI-flooded category on the platform. Roughly one in four top tech posts showed heavy machine generation. The category we write in is the one drowning fastest.

And then there are badges.
The bestseller badges here are earned with paid subscribers, a hundred of them for the first one. The leaderboards rank publications by paid revenue and paid subscriber growth.

Go look for a free publication on any of them; we haven’t found one. As far as we can tell, most of the discovery machinery runs on paid subscriptions, which tracks, because a cut of paid subscriptions is how the platform earns.

I’m not blaming Substack, it’s a business making rational choices. But it means a free publication is close to invisible to that machinery, no matter how careful the work is.

The Problem With Free

There’s a third thing I’ve had to admit to myself, and this one’s about me.

When everything you make is free, people treat all of it the same. And the assumption attached to free is that it can’t be worth much. I caught myself doing exactly this not long ago. I paid for a workshop, real money, and I sat through it taking notes, valuing every minute. The same material offered to me for free, I’d probably have half watched and forgotten. Paying changed how I treated it.

That’s just how people work.

Before a price tag is ever about money, it’s a signal that says this is worth your attention. Right now our deepest work goes out free into a feed where free reads as disposable, and gets mistaken for the same filler we’re trying so hard not to be.

Lost Patience

And one day, it just ticked me off.

I haven’t lost interest in tech, or in writing about it.
We’ve got something like thirty pieces in the pipeline right now, and every one of them gets reviewed and edited until it’s the best we can do at the time. I myself write across more than four surfaces on a regular basis.

What I’ve lost patience with is the growth game.
Chasing numbers on a platform that buries careful work under a daily flood of the careless kind.

I wrote recently on my personal blog about how every metric eventually rots into a vanity metric. Followers can be bought. Engagement can be botted. Somewhere in the last year, my math changed.

I would rather write for one person who genuinely cares what we think than a hundred who tap a like and scroll on without really reading.

A smaller room of people who value the work beats a big number that doesn’t.

The Decision

You’ve probably read a dozen going-paid posts by now.
They follow a script: a wall of reasons, a pricing table, a paragraph of feelings.

So I’ll give you the part that script usually skips.
This took most of this year to decide.

And what finally settled it was noticing what the current setup was doing to us: the work we care most about getting treated as disposable, and the three of us slowly learning to care a little less about making it.

That’s the thing being protected here, more than any revenue line: the will to keep doing this properly.

So here’s the change, and we’re rolling it out as we go.
The switch flips on with this post. The rest gets handled over the coming weeks: sorting the mechanics, working back through the archive, catching the edge cases. I’d rather start straight with you and adjust than wait for everything to be perfect.

Deep Dives go paid, the ones already up as well as the new ones. That’s the certain part. They cost the most to make, and they’re the work I least want to see treated as disposable, so that’s where the paid tier begins.

For series, we’re trying a split. Part 1 stays free, so you can read the opener and decide if it’s for you. The later parts go to paid subscribers. You always get a real piece of every series for free, and the deeper installments help fund the work behind them. This one’s more of a judgment call than a rule, and we’ll see how it feels in practice.

Paid subscribers also get Substack Chat, where we share rougher thoughts and what we’re digging into before it becomes an article.

Where the support goes is simple: paying the team fairly, covering the tools, and buying the room to go deeper. This is still a passion project. It just needs to become one that can sustain itself.

What’s Not Changing

Every standalone article in the categories you already read us for stays free: consumer education, emerging tech, industry highlights, gaming, everyday security, and before you buy.

Those are the core of SK NEXUS, and they stay open. So does everything we have already published in them, going all the way back. Deep Dives are the exception.

And the free side is going to get better, not thinner. We’re going to be more active on Notes, post more often, put out more of the short, useful stuff.

Nobody here is trying to starve the free reader into paying. The deep work goes on a pedestal where it stops getting mistaken for filler, and the everyday work, which is growing, stays open to everyone.

To Wrap Up

We write for the everyday curious person, whether they’re in Karachi or Dubai. And I know that in a lot of the world, the price of a subscription is not a rounding error. It’s a real decision, weighed against real things.

If that’s you, hear me clearly: the free side of SK NEXUS was built for you, on purpose, and it is not a consolation prize. Most of our articles, every Note, the comments, and the whole back catalogue outside the Deep Dives, all of it stays open, and there will be more of it than before.

Stay, read, push back in the comments, send a piece to a friend. That counts as support too. It always has.

In the end, depending on who you are, this goes one of two ways.

If you just found us, none of this should worry you. Most of what we do is free, you’ve got more than enough articles to dig through without paying a cent, and the conversation’s open. Read, comment, stick around.

If you’ve been here for a while, reading, replying, sending our stuff to a friend, then this one’s basically for you, and I’ll just say this - Thank You!

Genuinely.
You’re the reason there’s anything worth making sustainable in the first place.

This wasn’t an easy call. It’s the kind of decision you only make because the ground shifted under you, the scraping, the flood of machine made filler, the way good work gets buried now.

We’d rather meet that honestly, with you, than let the thing wind down without ever telling you why. And if going paid doesn’t work, you’ll hear that from me too.

A while back we ended our Year in Review by saying independent media survives on reader trust.

I still believe that.

But now I’ve just learned the other half: trust is what you build on.
It doesn’t pay the writers or fund the next big step by itself.
Sustainability is what turns the trust into something that lasts.

We’ve spent a year earning yours. If you’re in, you can support the work.
If you’re not there yet, keep reading. The door stays open either way.

See you in the next one.

Saqib Tahir
Founder and Editor
SK NEXUS

If you play PC games, something a little strange is happening on your screen right now.

Your graphics card is drawing frames as fast as it can. Your monitor is refreshing at its own fixed speed. Two different machines, two different clocks, and nobody told them to agree.

Most of the time you don’t notice. But every so often they fall out of step, and you get screen tearing: that ugly horizontal split where the top of the screen shows one frame and the bottom shows the next.

You’ve probably seen it a hundred times, mid-turn in a shooter or during a fast camera pan. You just never had a name for it.

Let’s get into it.

So, this article is the last in a series of 3 articles where we talked about the hidden tech behind our games that most of us don’t see.

The first article in this series covered upscaling. It explained how your GPU is kind-of lying to you when it comes to rendering your games. It is linked below:

The second article covered frame generation. In this one I talked about how AI has been inside frames of your video games for quite some time now. It’s linked below too

The article you’re going through now is the last article in that series where we’ll go through frame-sync and latency improving technologies like Reflex, Anti-lag and so on.

The Monitor GPU Sync Issue

Your GPU sends frames to your monitor. Your monitor refreshes on its own clock. Those two clocks do not naturally sync. That speed is measured by Hz for your monitor and fps for your GPU.

When the Hz of your monitor doesn’t sync with the FPS your GPU is generating screen tearing in your video games.

That is when a horizontal split appears on screen where two different frames show up at the same time. Very noticeable in fast camera pans or quick turns in shooting games.

Most people have seen this happen but never knew what it was called.

The reason it happens is simple. Your monitor refreshes from top to bottom at a fixed rate, usually 60 times per second on a standard display.

If your GPU sends a new frame mid-refresh the monitor draws part of the old frame and part of the new one. The result is that visible horizontal line across the screen.

It gets worse the more your frame rate fluctuates. And in real games your frame rate fluctuates constantly.

One moment you are staring at a wall getting 120 frames per second. Next moment you turn around and see an explosion with ten enemies on screen and your frame rate drops to 45.

Your monitor does not care. It keeps refreshing at 60Hz no matter what your GPU is doing.

That mismatch is the core problem every sync technology has tried to solve.

What Screen Tearing Actually Is

Let’s get specific about what is happening when you see that split.

Your monitor refreshes from top to bottom at a fixed rate. If your GPU sends a new frame mid-refresh the monitor draws part of the old frame at the top and part of the new frame at the bottom.

The result is a visible horizontal line across the screen.

In a racing game when you pan the camera quickly the tear line is obvious. Half the image shows your car in one position, the other half shows it slightly ahead.

Your brain notices this immediately even if you do not know what to call it.

The faster your frame rate moves relative to your monitor refresh rate the more tearing you see. A game running at 100fps on a 60Hz monitor will tear constantly.

A game locked at exactly 60fps on a 60Hz monitor will not tear at all because the refresh rate matches perfectly. But real games do not hold perfect frame rates.

You hit a particle effect, frame rate dips to 50. You look at the sky, frame rate jumps to 80. The monitor refresh stays at 60 the whole time.

That is why tearing has been a fact of life in PC gaming for decades.

The Old Fix That Made New Problems

The first widely used fix for screen tearing was VSync. Short for Vertical Synchronization.

VSync forces your GPU to wait until your monitor finishes refreshing before sending the next frame. This eliminates tearing completely.

To explain it simply, VSync forces your GPU to slow down to the speed of your monitor so both can remain in sync and there is no screen tearing.

But as with most technologies, there is a big tradeoff with VSync. If your GPU misses a refresh cycle you drop to the next sync point.

On a 60Hz monitor that means if you dip below 60fps you suddenly drop to 30fps because the GPU has to wait for the next available refresh window.

That causes stutter. Very noticeable stutter but still a stutter.

The wait time also adds input lag. The delay between when you click your mouse and when you see the result on screen gets longer.

Competitive players have hated VSync for years because of exactly this. Turning it off and accepting tearing was often considered the lesser evil.

For a long time that was the choice. Tearing with VSync off, or stutter and input lag with VSync on.

Neither option was great but VSync gave us the option to solve the sync issue even though it added some latency.

That latency is normally fine with story-mode games but it was never a solution for games that move fast.

G-Sync and FreeSync Enter The Chat

G-Sync and FreeSync arrived in 2013 and 2015 respectively. They solved the problem by flipping it around.

Instead of making the GPU wait for the monitor, these technologies make the monitor wait for the GPU.

The monitor dynamically adjusts its refresh rate to match whatever the GPU is outputting at that moment.

If your GPU sends a frame at 47fps the monitor refreshes at 47Hz. If your GPU jumps to 92fps the monitor refreshes at 92Hz.

No fixed clock conflict. No tearing. No stutter.

G-Sync is NVIDIA’s proprietary version. It requires a hardware module inside the monitor which is why G-Sync monitors used to cost significantly more than regular displays.

Basically, if you wanted to use G-Sync you had to buy a G-Sync capable monitor and when this technology started out, they weren’t that cheap.

FreeSync is AMD’s open standard version built on VESA Adaptive-Sync. It does not require special hardware in the monitor so FreeSync displays are cheaper.

Since 2019 NVIDIA GPUs also work with FreeSync monitors under the G-Sync Compatible label. NVIDIA tests certain FreeSync monitors and certifies them as working properly with their cards.

The practical difference today is minimal for most users. Both technologies eliminate tearing and stutter by letting the monitor follow the GPU instead of the other way around.

If you play on a PC built in the last five years and your monitor supports either technology you are probably already using it without thinking about it.

G-Sync monitors still tend to be more expensive but the performance gap has closed significantly. However, this situation is changing fast because NVIDIA just released sort-of version 2 of G-Sync.

This is called G-Sync Pulsar and not only is this technology much better but also cheaper to implement for monitor makers. G-Sync Pulsar mainly deals with ghosting issues inside gaming.

The technology is so cool that it’s hard to go through it in this article but I’ll be covering it in a separate article in full detail here on SK NEXUS.

What Is NVIDIA Reflex?

Screen sync fixes tearing but it does not fix input lag caused by frames sitting in a render queue waiting to be sent to the display.

That is where NVIDIA Reflex comes in.

Reflex works by synchronizing CPU and GPU work so frames are produced right when they are needed rather than queued up ahead of time.

Normally your CPU submits work to the GPU in advance. The GPU builds up a backlog of frames to render. By the time a frame reaches your screen it is showing input from several frames ago.

Reflex reduces that queue. Frames are produced closer to the moment they are displayed which dramatically cuts the gap between your mouse click and what you see on screen.

NVIDIA says Reflex reduces system latency by up to 50 percent across supported games.

At the moment, 150+ games support NVIDIA Reflex. Most competitive shooters like Valorant, Apex Legends, Call of Duty, and Fortnite, all support it.

The effect is most noticeable in fast-paced games where reaction time matters. Casual single-player games do not benefit as much.

For competitive players Reflex is considered essential. The difference between 60ms latency and 30ms latency is the difference between hitting a shot and missing it.

Although for our average Bashir playing story-mode games, it isn’t that necessary for him. He can choose to use Reflex or not.

You do not need special hardware to use Reflex. Any NVIDIA GPU from the GTX 900 series onward supports it as long as the game has implemented it.

Innovation in tech like Reflex has made NVIDIA that 5 trillion dollar giant it is today. But technologies like these aren’t what make NVIDIA its most money these days.

To find out why, check out my NVIDIA deep-dive article linked here.

Reflex 2 and Frame Warp

NVIDIA announced Reflex 2 at CES 2025 alongside the RTX 50 series cards. Note it was announced and hasn’t released yet.

Reflex 2 adds a new feature called Frame Warp. It warps the already-rendered frame to reflect the very latest mouse input right before it is sent to your display.

In NVIDIA’s own testing it reduced latency to ~14ms in The Finals at 4K on an RTX 5070 running at max settings. Down from 56ms without Reflex.

That is a massive improvement if it works as advertised.

Here is the catch though. NVIDIA has still not officially released Reflex 2 in a shipping game as of April 2026.

A modder named PureDark built a working demo from extracted binaries in late 2025 and confirmed it works on RTX 20 series cards too. But NVIDIA has not made it available to the public yet.

The company has not given a clear timeline on when Reflex 2 will actually ship in games.

So for now Reflex 2 exists in demos and modder builds but not in anything you can actually download and play officially.

AMD Anti-lag and the Other Side

AMD’s equivalent to Reflex is called Anti-Lag.

It reduces the render queue between CPU and GPU to cut input lag similarly to what Reflex does.

Anti-Lag 2 introduced per-game integration starting with Radeon Software Adrenalin. It works best paired with FreeSync just as Reflex works best paired with G-Sync.

AMD users do not get Reflex because it is NVIDIA-exclusive. But in supported games Anti-Lag 2 closes most of the gap.

FreeSync Premium monitors are widely available at prices significantly lower than full G-Sync hardware modules. That makes the AMD path more accessible if you are building on a budget.

The technology itself is comparable but the main difference is game support. Reflex has over 150 games as of 2026. Anti-Lag 2 supports around 30 or so games roughly speaking.

If you play competitive shooters on PC chances are the game supports Reflex but not Anti-Lag. That has been the pattern for years.

AMD is working to close that gap but developer adoption has favored NVIDIA’s solution so far.

Just so you know AMD has renamed Anti-Lag 2 to FSR Latency Reduction 2.0 but I went with the simple one because it sounds cleaner.

AMD’s story as a company is more interesting than their Anti-lag technology. The company has risen from the ashes multiple times.

If you’re interested in reading more about AMD’s story, check out my dedicated article linked here.

Neural Texture Compression Is Coming

Nvidia and AMD are both working on AI-powered texture compression for video games.

Instead of storing raw texture data the system stores textures in a compressed AI format. The GPU reconstructs that data into normal textures using AI.

This matters for game install sizes, texture streaming, and how much VRAM your graphics card needs to hold all the visual detail in a scene.

The promise is better quality at lower storage requirements. Instead of saving every pixel you save instructions on how to rebuild the texture.

Nothing has shipped to consumers as of now though. Both companies have demonstrated early versions but neither is available in a product you can actually use.

It fits the same pattern as upscaling and frame gen. Less real data, more AI filling in the gaps.

There will of course be some trade-offs with this compression though. It could be visual glitches or inconsistencies but the hope is that with time these issues will be ironed out.

For now this is mostly a research project. But it is worth knowing about because it shows where the industry is headed.

RTX 5090 And Your Kidney

If you’ve read through the three articles in this series you should have this stamped into your head:

Our GPUs render fewer real pixels and fewer real frames. And smarter delivery technologies (like G-Sync, FreeSync) make our games run better, especially when we’re constrained by our hardware.

As everything is claiming to be “AI-powered” these days, PC gaming has been using AI for quite some time now. It has been used in technologies like DLSS, FSR, XeSS.

These technologies may not make much sense for someone who owns an RTX 4090.

But as the Steam Hardware Survey shows, 99% of us gamers aren’t playing on that $2k RTX 5090 GPU. Maybe because we love our kidneys :p.

And if we turn our heads to developing countries like ours these technologies become even more relevant considering gaming hardware isn’t cheap here.

For people with aging hardware or tight budgets these technologies are the difference between playing new games or not being able to play them at all.

Look, I use an AMD R9 290 GPU from 2014 so I precisely know how gaming feels with these technologies turned off. FSR alone lets me play games that would otherwise be unplayable on this old card.

Things like FSR, DLSS didn’t exist ten years ago. Now these are assumed by default.

Game developers design and test their games with these technologies turned on because that is where most players are at.

And that’s the main thing, these technologies don’t just make your frames go up but they’re the difference between being able to play a game or not for many people.

Access to games you could not otherwise run. Access to smooth performance on hardware that should have been obsolete years ago.

The tech industry loves to sell you new hardware. These technologies let you squeeze more life out of what you already own.

That matters more now than ever with hardware prices where they are.

For now the short version is this. Your GPU and monitor are working together in ways they never used to.

The result is smoother, faster, and more responsive than what the raw hardware should be capable of producing.

Whether that is a good thing or a concerning one depends on how you feel about AI doing more of the work your hardware used to do.

And it’s not like anyone is forcing you. You’re free to disable DLSS, G-Sync but for those who can’t play their games without these technologies, there is an actual impact in their lives.

And many people don’t like the idea of these technologies. Some people genuinely hate “fake frames”.

But whether one likes these technologies or not that is the way how gaming works at the moment. And it is not going back it seems.

If you enjoyed going through the article this far, a subscribe to our publication would make my day, just hit the button below, it’s free:

Subscribe now

Similar Articles From Us

Further Learning

• Nvidia G-Sync puts processors inside your PC monitor

• Nvidia’s new G-Sync monitors automatically adjust brightness just like a laptop

• AMD FreeSync explained

• What is Nvidia Reflex

• Counter-Strike 2 is getting Nvidia’s latency-reducing Reflex tech

• NVIDIA Reflex 2 With New Frame Warp Technology

• NVIDIA Reflex reduces latency by up to 50%

• AMD Quietly Renames Anti-Lag 2 to “FSR Latency Reduction 2.0”

• Nvidia G-Sync vs AMD FreeSync: Which monitor tech is best?

• What Is Screen Tearing and How to Fix it

• What Is VSync, and Should You Enable It?

• Nvidia’s groundbreaking VRAM solution only works best on its fastest GPUs

A few months ago, I arrived at work expecting a normal day.

Instead, I spent a good chunk of it waiting.

Our website had become inaccessible, and since a large portion of my work revolves around publishing, editing, and managing content online, there wasn’t much I could do until things came back.

And it wasn’t just us.

When major cloud providers experience problems, the effects spread far beyond a single company. Over the past few years, outages involving Amazon Web Services (AWS), Microsoft Azure, and Cloudflare have repeatedly disrupted services used by millions of people.

Most of us rarely think about the infrastructure that makes modern technology work. We just assume everything will be there when we need it. And most of the time, it is. But every now and then, something breaks and we’re reminded that the digital world depends on infrastructure that can go down and disrupt the whole world.

The funny thing is, modern technology often feels more reliable than ever. Yet at the same time, many of the systems we rely on have become increasingly centralized, interconnected, and dependent on a surprisingly small number of companies.

And that's what this article is about: how the internet ended up here, why so much of modern life depends on a handful of critical services, and what we can personally do about it.

How the Internet Became More Centralized

When the web started becoming mainstream during the 1990s, the digital world was far more distributed than it is today.

Websites typically ran on their own servers. Companies managed much of their own infrastructure. Services were separated from one another and if one website experienced problems, the impact was usually limited to that particular website.

With the passage of time, that began to change.

As internet services became larger and more complex, companies started looking for ways to reduce costs and simplify operations. Running servers is expensive. Maintaining data centers is difficult. Scaling infrastructure for millions of users requires expertise that many organizations simply don’t have.

Cloud computing solved a lot of those problems.

Instead of purchasing hardware, building server rooms, and hiring teams to manage everything, companies could rent infrastructure from providers like Amazon Web Services, Microsoft Azure, and Google Cloud.

The advantages were enormous - startups could reach global audiences without building their own infrastructure. Businesses could scale faster than ever before. Developers could focus on building products instead of managing hardware.

The cloud, in a way, transformed the internet, but every solution comes with tradeoffs.

When more organizations moved into the same cloud platforms, dependence became inevitable. Thousands of companies that once operated independently now relied on the same providers, the same infrastructure, and in many cases, the same underlying services.

The same thing happened across other parts of the internet - companies adopted the same content delivery networks. The same authentication providers. The same software repositories. The same security certificate authorities.

Each decision made sense on its own.

Collectively, however, they created a very different internet than the one that existed decades ago.

The Hidden Fragility of Modern Technology

The Companies Holding Up the Internet

Some people might argue that the internet itself remains distributed. While that may be true, many of the services running on top of it have become heavily centralized.

Cloudflare is a good example - the company provides a lot of services for millions of websites. Depending on the measurement used, Cloudflare helps power or protect roughly one-fifth of the web, making it one of the most important pieces of infrastructure most people have never heard of. Recent outages have demonstrated just how much disruption a single failure can create when so many services depend on the same provider. In June 2026, another major disruption involving internet infrastructure providers affected many services, once again demonstrating how failures can ripple across large portions of the web.

Amazon Web Services represents another major concentration point - thousands of companies no longer operate their own infrastructure. Instead, they rent computing resources from AWS. That includes streaming platforms, collaboration tools, financial services, government systems, and countless other online services.

When AWS experienced a major outage in late 2025, the effects extended far beyond Amazon itself. Services including Slack, Zoom, Canva, Snapchat, Roblox, and many others experienced these disruptions as well.

Similarly, Let’s Encrypt issues a huge portion of the internet’s TLS certificates, helping websites establish secure connections. GitHub hosts an enormous percentage of the world’s software projects and development workflows.

More recently, AI providers have started becoming another concentration point. The internet itself does not depend on ChatGPT or Claude in the same way it depends on cloud providers or certificate authorities, but many organizations increasingly rely on these platforms for day-to-day activities.

When these services experience outages, productivity often slows down across thousands of organizations at the same time. Both OpenAI and Anthropic experienced multiple service disruptions throughout 2026, highlighting how quickly businesses have started building workflows around a relatively small number of AI providers.

The Growing Complexity of Modern Systems

For years, technology companies have been competing to add more features to their products. Your email application gets an AI assistant. Your phone receives new AI capabilities every few months. Every software update promises to do more than the one before it.

From a user perspective, some of these additions are genuinely useful, but the problem is that every new feature introduces additional moving parts behind the scenes.

A modern application is very different from the software people used twenty years ago. Back then, a program was often self-contained. Today, even relatively simple applications can rely on hundreds or thousands of software packages, external services, APIs, and cloud-based systems working together at the same time.

And with time, that complexity starts to accumulate.

You’ve probably noticed this yourself - computers are dramatically faster than they were a decade ago. Yet somehow, many applications feel heavier than the software they replaced.

Part of this comes from what developers often refer to as software bloat - companies keep adding functionality to their existing product because new features are easier to market than stability improvements. Nobody runs an advertising campaign around reducing technical debt or simplifying backend infrastructure. But AI-powered search, collaboration tools, assistants, and dashboards are much easier to sell.

The consequences become even more apparent when you look at how modern services are built.

Suppose you open your banking application. At first glance, it feels like you’re communicating directly with your bank. But the request might pass through content delivery networks, authentication providers, cloud infrastructure, and several other third-party systems before you ever see your account balance.

A lot of the time, none of this matters because everything works, but when one component fails, the effects don’t always remain isolated.

Researchers studying distributed systems have repeatedly documented what are known as cascading failures. A problem in one service can spread into another service, which then affects another, creating a chain reaction that becomes increasingly difficult to predict and diagnose.

And recently, Artificial Intelligence has been added on top of all this. Every company is in a race to shove more “AI powered” features into their products which creates even more dependencies that didn’t exist a few years ago.

The Volunteers Holding Up the Internet

Another weakness that often goes unnoticed is how much of the internet depends on software maintained by people many users have never heard of.

A huge portion of the internet also depends on open-source software projects that are freely available for anyone to use.

OpenSSL is one example. The software helps secure communications across millions of websites and online services. For years, despite being used throughout much of the internet, the project operated with limited funding and was maintained by a surprisingly small team.

In 2014, researchers discovered Heartbleed, one of the most significant internet security vulnerabilities in recent history. The flaw affected a huge number of systems around the world and exposed how much critical infrastructure depended on a project that lacked the resources many people assumed it had.

A similar story played out years later with Log4j. People didn’t even know about its existence until the Log4Shell vulnerability was discovered in 2021. Yet the software was deeply embedded inside enterprise systems, cloud platforms, government networks, and business applications around the world.

Neither OpenSSL nor Log4j is an unusual example, and the same pattern is everywhere.

Large portions of the internet depend on software maintained by small groups of developers, nonprofit organizations, and volunteers. In many cases, these projects are not generating billions of dollars in revenue despite being used by companies that are.

This has become such a well-known issue that a famous XKCD comic about software dependencies is regularly referenced throughout the open-source community. The comic jokingly depicts modern digital infrastructure resting on a tiny project maintained by a random individual that nobody has heard of.

The Infrastructure Dependence Trap

At this point, a reasonable question starts to emerge.

“If companies know they’re becoming dependent on a small number of providers, why don’t they simply move elsewhere?”

The answer is that leaving is much harder than joining.

You can see a smaller version of this in consumer technology as well.

Suppose you use an iPhone. Chances are you’re also using iCloud, AirPods, Apple Photos, Apple Notes, Apple Watch, and several other Apple services. None of these products are impossible to replace, but doing so requires time, effort, and often a willingness to change habits you’ve built over years.

The same thing happens with Google’s ecosystem. Gmail connects to Google Drive. Google Photos connects to your phone. Chrome syncs across devices. Everything becomes interconnected.

Now imagine that same problem on a much larger scale.

Companies spend years building systems around platforms like AWS, Azure, Google Cloud…etc. Applications are designed around specific services. Employees are trained to use particular platforms and workflows.

With the passage of time, entire businesses become tied to the ecosystems they started with.

Moving away isn’t impossible, but it can take months or years of planning, testing, and migration. In some cases, organizations discover that leaving creates more disruption than staying.

This is often referred to as vendor lock-in - the deeper an organization becomes embedded inside an ecosystem, the harder it becomes to leave. And because of that, many companies continue relying on the same providers even when they are uncomfortable with how dependent they’ve become.

The Risks Extend Beyond Consumer Technology

Up until now, most of the examples we’ve discussed have focused on websites, applications, and online services, but the consequences extend much further than that.

Modern cloud infrastructure now supports healthcare systems, financial institutions, transportation networks, educational platforms, and government services. Many of the systems that societies rely on every day are increasingly built on the same digital foundations.

Suppose some outage happens and a hospital loses access to critical systems, the consequences are very different.

Over the past several years, outages involving cloud providers have repeatedly affected organizations far beyond the technology sector. Hospitals have experienced disruptions to scheduling systems and patient records. Banks and payment providers have dealt with service interruptions. Government agencies have experienced outages affecting public services and internal operations.

The exact impact varies from incident to incident, but the bigger trend is becoming difficult to ignore - more and more critical services are becoming dependent on the same underlying infrastructure and as the concentration increases, failures have the potential to affect much larger portions of society than they did in the past.

Why These Risks Matter Beyond Technology

It’s hard to ignore the fact that more services are being consolidated onto the same cloud providers. More organizations are becoming dependent on the same platforms. More systems are becoming interconnected.

None of this guarantees a disaster, but it does increase the consequences when something eventually goes wrong.

We’ve already seen glimpses of what that can look like. Cloud outages have taken thousands of services offline simultaneously. Cyberattacks have disrupted supply chains and critical infrastructure. Undersea cables have been damaged by accidents, natural disasters, and geopolitical tensions, affecting connectivity across entire regions.

And that’s before we start considering scenarios involving multiple failures occurring at the same time.

Suppose a major cloud provider experiences a significant outage while a cyberattack is actively targeting connected systems. Or imagine a geopolitical conflict affecting data centers, internet exchanges, or critical communication infrastructure. Individually, each of these events is manageable. Together, they become much harder to predict and that’s where the real danger lies.

Reducing the Impact on Yourself

By now, you might be wondering what any individual can actually do about this.

The reality is that most people are not going to build their own infrastructure, run their own data centers, or completely disconnect from large technology platforms. Modern life simply doesn’t work that way anymore.

What people can do is reduce how dependent they become on any single service.

One simple example is keeping backups of important information. Many people assume that cloud services are permanent, but outages happen, accounts get locked, companies shut down products, and mistakes occur. If something is genuinely important, it’s usually worth keeping a copy somewhere under your own control as well.

It’s also worth paying attention to which services you rely on most. Some important questions which you should ask yourself are:

• If your email provider suddenly became unavailable tomorrow, what would happen?

• If your cloud storage stopped working for a day, could you still access important documents?

• If your password manager experienced an outage, would you still be able to access critical accounts?

Most folks never think about these questions until something breaks. A small amount of preparation can often take you a long way.

Where This Leaves Us

If there’s one thing I hope this article showed, it’s that modern technology is often far less independent than it appears.

The internet gives the impression of being a vast, distributed network made up of millions of different services. And on the surface, that’s true.

But underneath that surface, many of those services depend on the same infrastructure and the same organizations. That doesn't mean the internet is on the verge of collapse. But it does mean that stability isn't the same thing as resilience.

For years, technology has become so capable. At the same time, many of the systems we depend on have become increasingly interconnected, which we never notice because everything works just fine. It's usually only during an outage that we get a glimpse of how much depends on the same underlying foundations.

As individuals, there is only so much we can do. The larger questions around infrastructure concentration, competition, and regulation will ultimately be decided by governments, companies, and policymakers.

What we can do is understand the systems we rely on, reduce unnecessary dependence where possible, and prepare for the reality that outages are not exceptions. They’re part of how complex systems behave.

What You Might Like

• The Illusion of Choice in Tech

• Strong Passwords Still Get Hacked - Here’s Why

• The Illusion of Choice: How Tech Decides Before You Do

• The Death of Ownership: Why Your Tech Isn’t Really Yours Anymore

If you read our last piece on upscaling, you already know that your GPU has been tricking you for a while now.

Drawing your games at a lower resolution and then letting AI fill in the missing pixels. This was your GPU’s first trick but it doesn’t stop there.

In addition to upscaling, your GPU has another trick up its sleeve which it uses to smooth out your gameplay. That is called Frame Generation.

Where upscaling makes your games look sharper without actually drawing all the pixels, frame generation goes one step further.

It makes your games feel smoother without your GPU actually drawing all the frames.

In extreme cases, only one out of every six displayed frames may be traditionally rendered. The rest are software inventions.

That sounds weird to read but the frame generation technology is pretty cool.

And it’s more relevant now than ever, because of AI demand adding to graphics card prices.

Like any other technology, frame generation isn’t a silver bullet and it comes with trade-offs that marketing material often downplays.

That’s where this article comes in. I’ll go over what Frame Generation is and how it affects us gamers, especially those on a budget.

This article you’re reading is second in a 3-part series on gaming technologies that go unnoticed. Other articles in this series are linked below:

What Even Is a Frame?

Before frame generation makes any sense, let me explain how video games generally work.

When you play a video game, your graphics card is basically drawing lots of images in a matter of seconds. Each one of those images is called a frame.

Your monitor then displays those frames one after another, and your brain stitches them together into smooth motion.

The number of frames your GPU draws per second is your frame rate. If you are getting 60 FPS (frames per second), your GPU is drawing 60 full images every second.

The general rule is: the higher the FPS, the smoother the gameplay.

At 30 FPS, your GPU is drawing a new frame roughly every 33 milliseconds. At 60 FPS, it is every 17 milliseconds.

Now here is the catch.

More demanding games running at higher resolutions push even powerful GPUs to their limits. Hitting 60 FPS in a heavy game at 4K is genuinely hard work.

Hitting 120 FPS is much harder. And modern high refresh rate monitors, some running at 240Hz or higher, want a lot of frames to stay busy.

What Is Frame Generation?

Frame generation (aka Frame Gen) is a software technology that basically invents new frames and inserts them between the real ones in real-time while you play your games.

Your GPU still does the actual work. It draws a real frame, then draws another real frame.

Frame generation software then looks at those two real frames, analyzes what moved and where it moved to, and creates a new frame that sits between them in time.

Your monitor then displays the sequence: real frame, generated frame, real frame, generated frame.

In its basic form, frame-gen looks like twice the frame rate, even though your GPU only did half the drawing work. And that is basically frame generation.

The basic version of this is called 2x frame generation. One generated frame for every real frame. That is what AMD has offered for a while.

But the technology has moved fast and things have gotten more aggressive.

Frame generation is sometimes confused with upscaling because companies like NVIDIA bundle both into the same package like DLSS.

However, from a technology POV upscaling and frame generation are actually separate things doing separate jobs.

Upscaling makes each frame look sharper. Frame generation makes more frames appear.

You can use one without the other, though in practice most games bundle both together under a single option like DLSS or FSR.

How Frame Gen Actually Works

What frame generation does is not as simple as just copying a frame or blurring two together.

The software analyzes motion between two real frames. It builds what is called an optical flow map, which is basically a picture of how every part of the scene moved between those two moments in time.

Then it uses that map to predict where everything should be at the midpoint and renders that.

Think of it like watching two photos of a running person and being asked to draw the frame in between. If you know where their leg was in photo one and where it is in photo two, you can make a pretty good guess at where it was halfway through.

Modern frame generation, especially the AI-powered versions, is significantly more complicated than this example but you get the point, right?

In recent years, frame generation has gotten so good that it now looks quite real to the eyes of mere mortals :p.

NVIDIA’s Frame Gen

Just like upscaling technology in video games, NVIDIA was the company that pushed frame generation into the mainstream.

They introduced it with DLSS 3 back in 2022 with the RTX 40-series cards.

The basic idea was the same as above: draw one real frame, invent one in between, display both. This effectively doubled the visible framerate.

What made NVIDIA’s tech stand out was the hardware behind it. RTX cards have dedicated AI chips called Tensor Cores. These are chips designed specifically to run AI workloads quickly.

NVIDIA uses them to power its frame generation, which means the process is faster and adds less overhead than doing the same work purely in software.

Then at CES 2025, NVIDIA introduced DLSS 4 with Multi Frame Generation for their RTX 50-series cards.

Multi Frame Generation or MFG can generate up to 8x fps over traditional rendering when used with DLSS upscaling.

MFG is basically the big brother of frame generation :p

By early 2026, over 250 games had already added support for DLSS Multi Frame Generation, making it NVIDIA’s most rapidly adopted gaming technology.

Then at CES 2026, NVIDIA went further with DLSS 4.5.

The company increased the maximum frame generation multiplier from 4x to 6x for RTX 50-series users.

DLSS 4.5 also introduced Dynamic Frame Generation, which lets the software automatically adjust how many frames it is generating based on what your monitor can actually display.

By March 2026, the full DLSS 4.5 feature set became available including Dynamic Multi Frame Generation and the new 6x mode.

The important caveat here is that 6x Multi Frame Generation is exclusive to RTX 50-series cards. You’re kinda left out if you don’t have a 50-series NVIDIA card.

RTX 40-series cards are limited to traditional 2x Frame Generation, while RTX 20 and 30-series cards do not officially support DLSS Frame Generation.

GeForce graphics cards are what made NVIDIA’s name but today, revenue from graphics cards seems tiny compared to their AI-revenue.

To learn how NVIDIA got to be a $5 trillion empire and the heart of this AI revolution, you can read my deep-dive article on NVIDIA linked here.

AMD’s Take on Frame Gen

AMD introduced frame generation with FSR 3 in 2023. Their pitch, as always, was broad compatibility.

FSR 3 frame generation works on any GPU, not just AMD cards. You don’t need an AMD card to use it. That openness is genuinely useful for people (like me) with older or non-AMD hardware.

The trade-off was quality. The original FSR frame generation ran on regular shader cores rather than dedicated AI hardware. It produced some glitches on fast-moving objects and the results were not quite as clean as NVIDIA’s implementation.

AMD’s newer answer is FSR Redstone.

The AI-based FSR Frame Generation is designed specifically for AMD RDNA 4 architecture GPUs, the RX 9000 series, delivering higher quality results with fewer artifacts than the older version.

Here is where AMD is behind the competition right now. AMD’s frame generation is still limited to 2x. This means one generated frame for each real frame.

Intel and NVIDIA have both moved to multi-frame generation while AMD has not yet.

However, that appears to be changing soon because AMD recently hinted at upcoming multi-frame generation support.

AMD is probably having the best moment of its gaming history but for the most part they’ve been catching up with NVIDIA.

To know more about why AMD had to play catch-up with big boys like Intel and NVIDIA, check out my deep-dive article on AMD’s comeback here.

AMD has previewed FSR Diamond, the evolution of its frame-generation technology especially for consoles like next Xbox.

So if you are on an AMD card and frustrated that you are stuck at 2x while NVIDIA users brag about 6x, help may be on the way.

Intel and the Frame Gen Race

Intel is the newest player in this space and they have been quietly doing cool stuff.

Intel introduced XeSS frame generation as part of their Arc GPU lineup. For a while it was basic, similar to early FSR. But in early 2026 they made a big move.

Intel released XeSS 3 with Multi Frame Generation in January 2026, offering 2x, 3x, and 4x modes. Any game already supporting XeSS 2 frame generation would be compatible with XeSS MFG with no developer update required.

Then things got even more interesting.

In April 2026 Intel expanded XeSS 3 Multi Frame Generation support to virtually all Arc GPUs, including their older generation graphics cards and integrated graphics too.

That last part is remarkable. Multi-frame generation on integrated graphics chips inside laptop processors. Nobody else is doing that.

Intel even enabled multi-frame generation support across three generations of Intel Arc integrated graphics solutions.

Few GPU vendors have extended frame-generation support this broadly.

The catch with Intel is game support. XeSS as a whole has fewer supported games than DLSS or FSR. That is the ongoing challenge for Team Blue.

Intel’s technology works but it takes time for developers to actually implement it which is the harder problem.

Intel’s first graphics card entry failed and now they’re probably the newest entrant to graphics.

Intel lost the crown of being the best CPU maker to AMD. We covered Intel’s fall from the top in our Intel deep-dive article linked here.

The Input Lag Problem

Here is the thing nobody in the marketing materials will say very clearly.

Frame generation makes your game look smoother. It does not make your game feel more responsive.

Those are two different things and the difference matters a lot depending on how you play.

When your GPU draws a real frame, that frame contains information about where your mouse was and what buttons you pressed at that exact moment.

It is a true snapshot of what the game world looked like when you were interacting with it.

A generated frame does not have that. It was invented based on two real frames that already existed. It does not know what your mouse did. It fills in the gap based on what was already moving in the scene.

In competitive settings, the extra frames from frame generation do not help reduce latency.

The input lag remains tied to the speed at which the base game loop runs, and frame generation adds overhead that can actually increase the total time between your input and the response appearing on screen.

Think of it like getting a text message summary of a conversation that already happened. The summary looks detailed and covers everything. But it was written after the fact. Your input was not part of it.

NVIDIA has Reflex and AMD has Anti-Lag to help reduce this penalty. These tools lower the overhead and try to get you back toward a normal latency baseline.

These technologies reduce latency but they only bring you back to a baseline level. They do not make frame generation faster than native frames. They reduce the penalty without eliminating it.

The practical rule that most experienced PC gamers have landed on is: if you are playing single player story games like GTA 5, Red Dead Redemption 2, frame generation is great.

But when you are playing multiplayer games (competing with humans online) like Valorant, Call of Duty, etc., frame generation is not a good idea.

When Frame Gen Helps and Hurts

Look, frame generation isn’t magic.

It works best when your base frame rate is already decent. If you are running a game at 50 to 60 FPS natively and frame generation bumps that to 90 or 100, that is a genuine improvement.

Your game will feel smoother and the experience is noticeably nicer without obvious downsides.

Playing games like Cyberpunk 2077 on max settings can quickly take your performance down to 40 or 50 FPS.

Frame generation in these scenarios can push your displayed frame rate past 100 without requiring you to drop visual settings.

Where it falls apart is when your base frame rate is already too low. Multi-frame generation generally works best with a base frame rate of at least 60 FPS to avoid latency issues.

If your GPU is only getting 25 or 30 FPS natively and you turn on frame generation to get 60, you’ll have a bad experience. Visual glitches also get more noticeable at very low base frame rates.

Some developers have also been using frame generation as a crutch, launching games with heavy optimization issues and relying on frame generation to mask the stutter.

Frame generation cannot fix CPU bottlenecks. It can hide GPU frame rate problems but not everything.

You may have seen games in the last year or two that ran at oddly low native frame rates but shipped with frame generation on by default. That is not always a sign of a well-made game.

The ideal profile for frame generation is a game that runs at a solid 50 to 70 FPS natively, where you want visual smoothness more than you need razor-sharp response times.

That is the sweet spot and usually most people don’t need razor-sharp response times all the time so that makes technologies like frame generation actually useful.

The Future Of Frame Gen

Frame generation has gone from an exclusive feature to something all three major GPU makers now offer. That happened in just about three years.

The multipliers keep going up. NVIDIA is at 6x on their latest hardware. Intel is at 4x. AMD is catching up with multi-frame support likely coming soon.

Valve’s recently released Steam Machine uses upscaling and frame generation to achieve their performance targets. Valve has partnered with AMD to bring FSR 4 to their Steam Machines as well.

So, I feel there are two angles to frame generation:

The first one is that frame generation has been a really cool technology that has allowed gamers on a budget to play games that once ran badly on their systems.

I’m a testament to that. It’s great to have FSR supported on my graphics card even though it’s more than a decade old at this point.

And the way gaming hardware has shot up in prices because of AI-related demand, not everyone can afford 40-series graphics cards.

That’s where Intel’s open approach helps people who may not even have an Intel card, that’s a really positive thing.

But there is another concern: many developers have become complacent because of these technologies.

Instead of doing their best to optimize their games so they use resources efficiently, some game developers have started to lean on frame generation to take care of that.

Frame generation is like a supplement rather than a replacement for native performance. But the line between what your GPU rendered and what software invented keeps getting harder to find.

For now though, frame generation is increasingly common on gaming systems that support it.

If you have a reasonably modern GPU and you are playing a supported game, there is a good chance software is inventing at least some of what you are seeing.

And in single player games (that most of us play) the majority won’t even notice it.

Now that you’ve made it so far let me share a nugget of information.

The next article in this series is going to be on technologies like VSync, G-Sync, NVIDIA Reflex which make our games and hardware in-sync.

I’ll go over how gaming companies fix some of the issues that frame-gen and upscaling add to gameplay.

See you there :)

As always your feedback is what counts for SK NEXUS.

I’d really appreciate if you could just take a moment out and share this article to someone who’d benefit from learning about frame-gen technologies. You can use the button below:

Share

Similar Articles From SK NEXUS

Further Learning

• What Is DLSS And Should You Use It?

• AMD’s Answer to DLSS: How FSR Works

• Do “Fake frames” suck?

• NVIDIA DLSS 4.5 Dynamic Multi Frame Generation Now Available

• NVIDIA introduces DLSS 4.5 and Multi Frame Generation 6X at CES 2026

• Why DLSS Dynamic Frame Generation Is a Game Changer

• AMD teases Multi Frame Generation with FSR update

• AMD SDK suggests 4x and 6x frame generation multipliers are in the works

• Intel enables XeSS 3 Multi-Frame Generation in latest drivers

• When a high frame rate can lose you the game

• Input latency is the all-too-frequently missing piece of framegen analysis

• Frame Gen Isn’t a Performance Boost; It’s a Masking Agent

• AMD May Finally Add Multi-Frame Generation Support for Its Latest GPUs

• AMD Prepares FSR Update with Multi-Frame Generation

• Valve is working with AMD to bring FSR 4 to the Steam Machine