Who Is Actually Getting Rich From the $725 Billion AI Boom in 2026?

By the end of 2026, four companies — Microsoft, Google, Amazon, and Meta — will have spent about seven hundred twenty-five billion dollars combined chasing one bet: that artificial intelligence is worth building an entire new layer of the world's infrastructure for. That's up seventy-seven percent from the year before. One chip company just posted a single quarter's profit of nearly forty-three billion dollars because of it. One person's personal fortune is closing in on two hundred billion dollars because of it. And in a small town outside New York City, a company got a seventy-seven million dollar tax break to create exactly one permanent job. Somebody is getting extremely rich off the AI boom. It is very much not everybody. This is where the seven hundred twenty-five billion dollars is actually going — and who's quietly stuck holding the bill.

Here's the version everyone agrees on. Training and running today's AI models takes an enormous amount of computing power — specialized chips, made mostly by one company, Nvidia, packed into massive data centers that need almost as much electricity as a small city. So the four companies that run the biggest AI products — Microsoft's Copilot, Google's Gemini, Amazon's cloud AI services, and Meta's AI tools — are all racing to build more of that computing capacity before their competitors do. They're spending on land, on buildings, on cooling systems, on power contracts, and overwhelmingly, on chips. The bet is simple: whoever controls the most AI computing capacity today captures the most valuable AI customers tomorrow, and eventually charges enough for AI products to pay all of this back, with room to spare. That's the textbook version. It's also, increasingly, not the whole story.

Start with the most straightforward winner: the company selling the shovels. In its fiscal fourth quarter of 2026, Nvidia reported sixty-eight point one billion dollars in revenue — a single-quarter record — with sixty-two point three billion of that from data-center chips alone, up seventy-five percent from a year earlier. Net income for that one quarter: almost forty-three billion dollars. Nvidia doesn't just sell the chips everyone needs — for most of this boom, it's been close to the only company that can supply them at scale, which is a remarkable place to sit while four of the world's richest companies compete to spend more money than each other. The man running Nvidia, Jensen Huang, has seen his own net worth climb toward two hundred billion dollars in 2026, tracking Nvidia's stock, which is up about a quarter this year alone as of early September. When a chip becomes this scarce and this necessary, the company making it doesn't just profit from the AI boom. In a real sense, it sets the terms of the AI boom.

But Nvidia's money doesn't just flow in one direction — and this is where it gets stranger. Nvidia has committed up to one hundred billion dollars toward OpenAI's data-center buildout, plus a separate guarantee worth another hundred five billion for a single campus in Ohio. Oracle has signed on to sell OpenAI roughly three hundred billion dollars of computing capacity over five years — a deal that helped push Oracle's founder, Larry Ellison, into the very top ranks of the world's richest people as the stock jumped on the news. Nvidia has also invested directly in CoreWeave, a cloud company that turns around and buys — you guessed it — Nvidia chips, then resells that computing power to, among others, OpenAI. Follow it closely enough and you get a loop: Nvidia funds companies that spend that money buying Nvidia's own product, whose sales then justify Nvidia's stock price, which funds the next round. None of these are one-time cash payments — they're multi-year commitments and guarantees — but the investor Michael Burry, who famously bet against the housing market in 2008, has publicly flagged this kind of financing as a risk: the revenue looks real on the books, but if AI demand ever slows down, several of these companies could stumble at the exact same time, for the exact same reason.

Then there's a quieter money trail that most viewers never hear about. In late 2025, Meta needed to fund a single massive data-center campus in Louisiana — and instead of simply paying for it, it structured a twenty-seven billion dollar deal with a private-investment firm called Blue Owl, the largest private-credit transaction ever recorded. Under the deal, Meta actually owns only twenty percent of the venture that owns the campus; funds managed by Blue Owl own the other eighty percent, with major investors like PIMCO and BlackRock supplying much of the money. Meta gets to use the entire facility while keeping most of the cost off its own balance sheet, preserving its credit rating and its stock story. The investors get something that looks and pays like a bond, backed by a guarantee from one of the richest companies on earth. It's a clever piece of financial engineering — and it means that if this specific data center becomes obsolete faster than expected, a meaningful chunk of that risk lands not on Mark Zuckerberg's balance sheet, but on the pension funds and insurance portfolios that bought in for a safe, steady return.

So that's who's profiting. Here's who's quietly paying for it — and it's usually not who you'd guess. Start with your electricity bill. The independent watchdog that monitors the mid-Atlantic and Midwest power grid — a region covering fourteen states — found that data-center demand was a primary driver behind twenty-three billion dollars in higher electricity costs for ordinary customers, price increases expected to keep landing on bills through at least the end of this decade. Most of those customers don't have a single data center anywhere near their house. Then there's the tax-break side. In Orangeburg, New York, JPMorgan Chase secured roughly seventy-seven million dollars in tax breaks to expand a local data center — in exchange for creating exactly one permanent job, by far the largest per-job subsidy tracked anywhere in the country. It's not an isolated case. In Texas, state auditors reviewed twenty data-center projects that had promised to keep at least twenty to forty permanent jobs in exchange for tax breaks. Six of them failed to meet even that bar — while Texas's data-center tax exemption alone is projected to cost the state three point three billion dollars in just two years. The buildings get built either way. The bill, one way or another, tends to land on people who never got a vote on any of it.

Which raises the obvious question: is this whole thing built on real demand, or is it a very expensive bet that hasn't been tested yet? Analysts are genuinely split. Some call the bear case 'garbage,' pointing to Nvidia's actual, delivered revenue as proof the demand is real. Others, watching the same numbers by September 2026, were openly describing this as a late-stage AI bubble, drawing uncomfortable comparisons to 1999. Nobody knows which side is right — that's what makes it a genuine bet, not a sure thing. But the structure of the financing matters here: when the money flows in a circle, and the risk gets pushed onto investors who signed up for something safe, the people at the center of the loop can look enormously successful for a long time before anyone finds out whether the underlying demand was ever really there.

None of this means AI itself is fake, or that nobody should invest in it. It means that 'the AI boom is making people rich' and 'the AI boom is good for the average person's finances' are two completely different claims — and right now, the money is flowing much more reliably toward chip makers, cloud landlords, and private-credit investors than toward the towns hosting the buildings or the households paying the power bill. The next time a headline tells you how many billions a company is spending on AI, the better question isn't how much. It's: who exactly gets that money — and who's going to be handed the bill once it's spent.

Next time on Margin: a company everyone trusted — and the one decision at the very top that brought it all down. If you want to understand how money actually moves before the headlines catch up, subscribe — and follow the money.