You’re holding something that didn’t have to exist.
Not in the abstract way. In the literal way. Somebody looked at the world as it was and decided it wasn’t good enough, and bet on a different one. You don’t think about that most days. Why would you. The phone works. The chip is fast.
You can go back and see where this instinct actually starts. Sophie Wilson and Steve Furber were working out of an 18th century turkey barn outside Cambridge, designing a chip because nothing commercially available was fast or efficient enough for what Acorn needed.
They weren’t trying to end up inside every phone on Earth. They were trying to solve what was in front of them.
I think that’s almost always how it starts. Small. Specific. Nobody in the room thinking about forty years from now.
You’d think that kind of certainty only happens once, to the lucky ones. It doesn’t. Gabe Newell had already failed at this. Valve’s first living room console was a mess, handed off to manufacturers who never believed in it the way Valve did. He could have walked away.
He didn’t.
He built the next one himself. Kept the dream close.
It worked. It usually doesn’t.
And then there’s Mike Lazaridis. In a boardroom. Pointing at a keyboard.
I get this. It’s clearly differentiated.
Pointing at a touchscreen.
I don’t get this.
I don’t think he was protecting a number on a spreadsheet. I think he believed it. Right up to the end. I think some part of him still does.
He wasn’t wrong about much. He was wrong about timing, which turns out, in this story, to be almost the same thing as being wrong about everything.
None of them were chasing the money first. You can see that, looking back. That part always came later, for all of them, one way or another.
What you can’t see yet, not from here, is who ends up paying for it.
Here’s what believing costs. Not the believer. Everyone standing near them.
News Corp acquired MySpace’s parent company for $580 million in 2005. Within a few years the staff had been cut down in repeated rounds, eventually to a few hundred people. Nobody who got cut had decided to chase short term monetization over the product. Somebody else decided that. They just worked there.
I think about that distinction a lot. The deciding and the paying are almost never done by the same person.
GameStop owned its own answer to Steam once. A platform called Impulse. Management thought digital distribution was a passing phase, brick and mortar would come roaring back, so they shut it down in 2014. The man who built it, Larry Kuperman, said he thought that job was going to be his forever job.
It wasn’t.
You don’t hear about him in the GameStop story. You hear about the stock. The meme. The short squeeze. Somewhere underneath all of that there was a guy who believed in something his own company decided not to believe in, and then watched it get killed.
There’s a Rambus employee worth mentioning here too, carefully, because it’s not a story to use lightly. He’d made about five million dollars exercising stock options near the peak. Held the rest instead of selling. The stock collapsed. A year and a half later he was gone. I’m not going to turn that into a moral. Some costs don’t need a lesson attached to make them count.
You can find this same shape almost two hundred years earlier if you look. Britain was spending close to 8% of its GDP building railways by 1847. The boom collapsed shortly after, and by 1850 the index of British railway shares had fallen roughly 64% from its peak.
The track is still there. Most of it’s still used today. Somebody got something real out of all that money eventually. It just wasn’t the people who built it, and it wasn’t on the timeline anyone promised.
Somewhere in your portfolio right now, something is running on a timeline nobody promised you either.
I keep noticing how often that’s true. The dream survives. The dreamer doesn’t always get to be there when it pays off.
We just live downstream of the people who were.
Nobody’s voting on this one yet.
NVIDIA already lived through a version of this once. Its first real product, the NV1, bet on a rendering approach almost nobody else adopted. Microsoft went a different direction with DirectX, Sega walked away, and NVIDIA shipped a fraction of what it hoped before there wasn’t much left to do but stop.
It came back from that by admitting it. Not by doubling down. By saying, basically, we got the bet wrong, here’s the version that actually fits where the market is going.
I think people forget that part. They remember the empire. Not the year it almost didn’t have one.
Then came CUDA. Roughly a decade of investment with no clear payoff before the bet on general purpose GPU computing finally landed. No board meeting where someone stood up and said this is working. Just year after year of a platform that lost money and had no real argument for existing, except that Jensen Huang hadn’t stopped believing in it yet.
Now there’s a new bet, the biggest one yet, and nobody including the man making it seems totally sure how it ends.
NVIDIA committed up to $100 billion to OpenAI, with the expectation that OpenAI uses much of that money to fill its data centers with NVIDIA’s own chips. Somebody described it plainly. I’m Nvidia and I want OpenAI to buy more of my chips, so I give them money to do it. It’s not illegal. It’s not even unusual at a small scale. It’s unusual at this scale.
Huang’s answer, when asked about it, was something like we see something very different. I believe he believes that. I don’t think that settles anything.
TSMC didn’t choose any of this the way NVIDIA chose CUDA. TSMC just makes nearly all of NVIDIA’s most advanced chips, and now NVIDIA is most of TSMC. One analyst called it a single point of failure in plain language. TSMC’s own CEO has said, more or less, that he wants to see the demand prove itself real before he builds more capacity around it. That’s a strange thing for the most important manufacturer on earth to have to say out loud.
Underneath all of it, the memory makers tilted almost their entire production toward the kind of chips data centers need. Consumer memory got what was left. Prices went up ninety percent in a single quarter. Apple raised prices on the iPad and the MacBook and said, plainly, we have never seen a component price increase this much, this quickly. Sony raised the PS5 price. Twice in under a year. Microsoft called it a hardware component crisis, in those words.
Valve had been building a new Steam Machine. Their original target was around seven hundred and fifty dollars. It shipped at over a thousand. An engineer there said the memory suppliers tell them, every month, here’s how much you’re allowed to buy, take it or leave it.
Valve didn’t make this bet. Valve just got billed for it.
You might be paying this same bill the next time you replace a laptop and wonder why it costs what it costs.
Somewhere underneath that, residential electricity bills in Virginia have started telling their own story. One resident’s bill jumped from $164.56 to $343.38 over the same period a year apart, despite making no changes to how they used their home. Dominion Energy filings show data centers in the state requesting 70,000 megawatts for future locations, nearly triple the peak the utility has ever needed to power every home in Virginia on the coldest day of winter.
You may not live near a data center. You may not know yet whether you do.
Senator Mark Warner put it plainly: “If corporations are going to run data centers in Virginia, they should cover the cost of them.” The state is still arguing about whether that happens.
Residents don’t get to vote on whether it does.
Masayoshi Son has lived this exact uncertainty before, more than once, with real money, in public. He bought Alibaba early and turned twenty million dollars into something close to sixty billion. He fell, in his own words, in love with WeWork’s founder, and that cost him over eleven billion. He held NVIDIA stock, sold it in 2019, right before the AI boom, and watched what he sold go on to be worth over a hundred and fifty billion. He called it, simply, the fish that got away.
He sold NVIDIA again recently. Put the money into OpenAI instead.
I don’t think he’s learned a lesson here so much as decided the lesson doesn’t apply to him this time either.
Underneath Son, there are people with far less than Son, doing the same thing on a smaller scale. Retail traders piling into chip stocks because the chart’s been going up. A tiny optical company with a million dollars in actual revenue trading at a valuation that implies it shouldn’t exist yet, except that enough people decided to believe it anyway. An economic historian once said retail investors buy because the price went up and they extrapolate that forward, until the price loses all connection to what the thing’s actually worth. He wasn’t talking about AI when he said it. He didn’t need to be.
And above all of it, governments. Not investing exactly. Something closer to refusing to be left out. The United States took a stake in Intel and is backing a five hundred billion dollar infrastructure project. The UK launched its own fund within a week of an American company pulling out of a UK data center, because losing access to the infrastructure felt like losing the future itself. Saudi Arabia committed a hundred billion. The UAE, two hundred billion. South Korea, seventy five, explicitly so it wouldn’t have to depend on either America or China for something this important.
None of that spending has to make a return to be considered worth it. That’s not how a government calculates this kind of bet. It’s closer to how a country prepares for a war it hopes never happens. The cost gets paid either way. Some of it by you, in taxes you didn’t get a line item for.
I keep thinking about 2008. Banks built something nobody fully understood, on leverage nobody fully tracked, until the whole thing came apart and the people who built it were not the people who paid for it. Fifteen years later those same banks are bigger. When one of them failed again in 2023, the government paid for that one too.
Somebody is already writing the argument against doing that again for AI. Before AI has even finished whatever it’s going to do.
I don’t know how this one ends. Nobody does. That’s not me being coy. NVIDIA’s NV1 ended in a year. CUDA’s bet took a decade to prove out. This one might take longer than either, or it might be deciding itself right now while you’re reading this, in a board meeting you’ll never hear about.
What I do know is who’s standing where.
Somebody, right now, is certain about something unproven, the way Wilson and Furber were certain, the way Newell was certain when nobody else believed in the Deck, the way Huang was certain for ten straight years of losses. I don’t think that certainty is the problem. I think it might be the most human thing left in any of this. Wanting something better than what already exists, and being willing to risk real things to get there.
But somewhere downstream of that certainty there’s a resident in Virginia opening a bill that doubled. A retail trader who bought in because everyone else seemed to be buying. A taxpayer in a country that just decided, on their behalf, that this was worth the risk.
They didn’t get a vote. Most people downstream of a big bet never do.
You’re holding something that didn’t have to exist.
Somewhere downstream, something is being decided right now that you didn’t get a vote on either.
We’ve watched this shape form before. We’re watching it form again.
I still don’t know who’s going to be left holding it this time.
We never do, until it’s already theirs.
Colt Avery is a contributing writer at Aktiego. The views expressed in this column are the author’s own and do not represent the editorial position of Aktiego.com.
Sources
The Official History of Arm | Arm Newsroom
How BlackBerry Blew It | The Globe and Mail
News Corporation to Acquire Intermix Media, Inc. | SEC Filing
A Former GameStop Exec Thought Building Its Steam Competitor Would Be His ‘Forever Job’ | PC Gamer
The Collapse of the Railway Mania | University of Minnesota / Odlyzko
The Railway Mania: Not So Great Expectations? | CEPR
Nvidia’s $100 Billion OpenAI Investment Raises Bubble Questions | Fortune
Winter Electric Bill Spikes Spark Scrutiny of Va. Data Centers’ Power Usage | ARLnow
Virginia Senator Pushes Bill to Make Data Centers Cover Costs | The Cool Down
Editorial Disclosure
This article is an original opinion piece examining the pattern of technological conviction throughout computing history and the people who bear the cost when that conviction goes untested, including the current scale of AI infrastructure investment. No financial relationships exist with NVIDIA, OpenAI, TSMC, Valve, GameStop, SoftBank, Arm, or any other company mentioned. Historical and market data are drawn from SEC filings, Arm’s official newsroom, The Globe and Mail, PC Gamer, academic economic research on the Railway Mania, Fortune, and Virginia local news reporting on data center electricity costs. Commentary reflects the author’s own assessment. The information provided on this website is for informational and educational purposes only. Our content is derived strictly from verified online sources to ensure accuracy and objectivity. This analysis does not constitute financial, investment, or professional advice. Readers are encouraged to consult with qualified professionals before making decisions based on this information. For more information, please see our full DISCLAIMER.
Watch on YouTube
