Is the AI Conveyor Belt of Capital About to Stop?
The American economy is little more than a big bet on AI. Morgan Stanley investor Ruchir Sharma recently noted that money poured into AI investments now accounts for about 40% of the United States’ GDP growth in 2025, and AI companies are responsible for 80% of growth in American stocks. So how bad is it that the most recent major deal among AI giants, agreements that have driven up stock prices dramatically, look like a snake eating its own tail?
In recent months, Nvidia announced that it would invest $100 billion into OpenAI, OpenAI announced that it would pay $300 billion to Oracle for computing power, and Oracle announced it would buy $40 billion worth of chips from Nvidia. It doesn’t take a flow chart to get the feeling that these firms are just moving money around between each other. But surely that’s not happening…right?
It’s a little harder to get assurances of that than you might think.
Many of these agreements are, on their face, mutually beneficial. If everything is on the level, while these deals might be circular, they should be moving everything forward. Rishi Jaluria, an analyst at RBC Capital Markets, told Gizmodo that deals like these could result in a “less capacity-constrained world,” which would allow for faster development of models that could produce higher returns on investment.
“The better models we have, the more we can realize a lot of these AI use cases that are on hold just because the technology isn’t powerful enough yet to handle it,” he said. “If that happens, and that can generate real [return on investment] for customers … that results in real cost savings, potentially new revenue generation opportunities, and that creates net benefits from a GDP perspective.”
So as long as we keep having AI breakthroughs and these companies figure out how to monetize their products, everything should be fine. On the off chance that doesn’t happen, though?
“If that doesn’t happen, if there is no real enterprise AI adoption, then it’s all round-tripping,” Jaluria said.
Round-tripping, generally speaking, refers to the unethical and typically illegal practice of making trades or transactions to artificially prop up a particular asset or company, making it look like it’s more valuable and in demand than it actually is. In this case, it would be tech companies that are trying to make it appear like they are more valuable than they actually are by announcing big deals with each other that move the stock price.
So what might suggest whether this money is actually accomplishing anything other than serving as hot air in a rapidly inflating bubble? Jaluria said he’s watching for faster developments of models, advancements in performance, and overall AI adoption. “If this leads to a step function change in the way enterprise is adopting and utilizing AI, that creates a benefit,” he said.
Whether that is happening currently or not is kind of in the eye of the beholder. OpenAI has certainly shown advancements in its technology. The release of its Sora 2 video generation model has unleashed a fresh hell upon the world, used to generate significant amounts of copyright violations and misinformation. But the latest version of the company’s flagship model, GPT-5, underwhelmed and failed to live up to expectations when it was released in August.
Adoption rates of the technology are also a bit of a Rorschach test. The company boasts that 10% of the world is using ChatGPT, and nearly 80% of the business world says that it’s looking into how to utilize the technology. But the early adopters aren’t finding much utility. According to a survey from the Massachusetts Institute of Technology, 95% of companies that have tried to integrate generative AI tools into their operations have produced zero return on investment.
Where these investments are generating a return is in the stock market. Which, frankly, does not quell concerns about these firms simply boosting one another’s bottom line.
Take Oracle, for example. Last month, the cloud provider had a rough quarter by all traditional indicators. It missed on both its revenue and earnings projections, and its net income was flat year-over-year. And yet, the stock price soared. The reason: the company’s plump list of remaining performance obligations—financial agreements that will provide revenue that have not yet been fulfilled. There, the company showed a massive amount of growth, a 359% increase from the year prior, with a projected $455 billion coming in.
That money is not real yet. Nor is the growth the company has promised, claiming that its Oracle Cloud Infrastructure revenue would grow from under $20 billion to nearly $150 billion before the start of the 2030s. But all of it was sufficient for investors to drive up Oracle’s share price enough to slingshot CEO Larry Ellison into the top spot on the world’s richest person list, briefly leapfrogging Elon Musk.
Most of this promised revenue will come from OpenAI, which made a commitment to purchase $300 billion worth of computing power from the company over five years. The clock on that contract doesn’t start until 2027, but assuming it actually happens, it would be one of the largest cloud computing deals in history.
It’s also one of the most unlikely, just based on where the companies involved currently stand. In order to provide the compute that it has promised to OpenAI, Oracle will reportedly need to generate 4.5 gigawatts of power capacity, more than two Hoover Dams’ worth of power. On the other side of the deal, OpenAI will have to pay about $60 billion per year to fit the bill for the agreement. It currently generates about $10 billion in revenue, which, statistically speaking, is less than $60 billion.
You can see a similar circular shape to OpenAI’s recent deal with Nvidia rival AMD, too. The exact details of the agreement weren’t reported, but chipmaker AMD expects to generate tens of billions of dollars over the next half-decade as it sells its AI chips to OpenAI. As part of the agreement, OpenAI gets a swath of shares in AMD, with options to buy up to 10% of the company. Lucky for OpenAI, there’s really no better time to get your hands on some AMD shares than right before it announces a big AI-related deal. The company’s stock price surged by about 35% following the announcement.
With those two most recent deals on the books, OpenAI has agreed to more than $1 trillion worth of computing deals so far this year. That’s a lot for any company to spend, but it’s especially a lot for a still-private company that reports just $10 billion in projected revenue through 2025. Even by its most recent funding rounds, the company as a whole is currently valued at about $500 billion.
Most of those deals have contingencies attached. For instance, Nvidia’s investment in OpenAI isn’t actually $100 billion, but an initial $10 billion for one gigawatt of data center capacity with the potential for $100 billion if 10 gigawatts are ultimately achieved. But the stock prices and valuations certainly seem to treat these deals as if they are set in stone. And OpenAI seems to be operating that way, too. The company claims that it’ll more than 10x its revenue in the next few years, and projects it’ll hit $129 billion annually by 2029.
Conveyor belts of capitalThat type of potentially inflated revenue figure is the kind of thing that makes some people think of the Dot Com bubble of the early 2000s, where we saw companies like Commerce One receive a $21 billion valuation despite barely having any revenue. But Peter Atwater, Adjunct Professor of Economics at William and Mary and President of consulting firm Financial Insyghts, sees a different reflection in the AI bubble: the housing market collapse.
“What we saw at the top of the mortgage market was all of these conveyor belts of capital, money flowing from one party to another party to another party. And what you started to see was that there were multiple points of relationship so that any participant in the system was then dependent on every other conveyor belt in the system working simultaneously to keep the system going,” he told Gizmodo. “In many ways, we’re seeing the same developing web of capital flows across the AI space.”
This creates some obvious problems. The circular deals that, in theory, are wheels moving the whole thing forward all have to keep turning. If any of them stop, the whole thing stops, because they are all so interconnected that no failure is truly isolated.
Atwater said that the types of major, metric-contingent deals that have been dominating headlines in the AI space aren’t all that different from some of what was happening in the mortgage industry back in 2007, where some of the financial commitments required mortgages to meet certain conditions.
“In the frenzy of a bubble, everyone overcommits. The purpose of overcommitting is to stake a claim in what you believe will be an intensely scarce commodity in the future. So you have buyers overcommit and you have sellers agreeing to overprovide as a result,” he explained. “What we find over and over is that commitments are among the first obligations to be cut off once conditions change, once confidence begins to fall.”
Right now, there’s a stomach for those commitments. That isn’t guaranteed to be there in the future if all of these promised returns on investment don’t materialize. Atwater said that the market requires credit markets being willing to continue to extend massive sums of money to cover the agreements made, equity markets that value these transactions at “an extraordinary multiple,” and suppliers capable of delivering the promised products. There’s no guarantee that all of those factors will hold.
The math is already pretty tricky. As tech commentator Ed Zitron has pointed out, major firms like Microsoft, Meta, Tesla, Amazon, and Google have invested about $560 billion in AI infrastructure over the last two years. They’ve brought in a combined $35 billion in AI-related revenue. OpenAI’s commitments are even bigger, with returns that are arguably even smaller.
The company’s development and expansion of its services will rely in no small part on massive data center projects, which will require the same amount of energy to operate as New York City and San Diego combined—energy that currently isn’t even available. And, once again, there is no guarantee that the end product, once all of that energy is spent and data centers are built, will actually generate revenue.
“Ultimately, if you do not have a consumer for the product, there will be no AI space because these companies can’t continue to do this for nothing. Listening to a lot of the calls in the last couple of weeks, there’s a clear open question as to how these companies are going to make money at this,” Atwater said.
For the moment, everyone is seeing green, and hope springs eternal. As long as that is the case, no one will ask where the revenue is coming from. “Right now, the AI sector is operating in a forever mindset. They are acting as if they have a very long period of time under which they can figure this out and make money,” Atwater said. “As long as confidence is high, this entire ecosystem can offer fantasy. When confidence falls, they’re going to be expected to deliver real-term performance in a very short time frame.”
Unfortunately, should that happen, it won’t just be these companies that bear the brunt of the failure. “You have to look at this as a larger ecosystem. To talk about AI today, it means we have to talk about the credit market, we have to talk about the credit market. Wall Street and AI are a single beast,” Atwater said, warning that a very small number of firms currently have a major grasp on the whole of the American economy.
Lots of investors are piling into the AI space, fearful of missing out on a market that seems like it can only go up. But few of them are looking at why those valuations and stock prices keep climbing, showing little curiosity as to what might happen if all of this money is just getting shifted around, artificially inflating the actual value of the companies they are betting on.
“‘Why?’,” Atwater said, “is the last question asked in a bull market.”
gizmodo
