Article Directory
The AI Revenue Machine Is Eating Itself
There’s a persistent hum in Silicon Valley right now, a low-frequency vibration of anxiety beneath the roar of generative AI hype. It’s a feeling reflected in headlines like A tangled web of deals stokes AI bubble fears in Silicon Valley. It’s the kind of feeling you get before an earthquake—the animals are nervous, but the tourists are still taking pictures. Everyone from Jamie Dimon to the Bank of England is talking about an AI bubble. Even Sam Altman, the CEO at the epicenter of this tremor, admits that "many parts of AI... are kind of bubbly right now."
That’s a remarkably candid, if convenient, admission. It frames the issue as one of over-enthusiasm, of a few "silly start-ups" getting "crazy money." It’s the standard, cyclical narrative of tech booms. But my analysis of the available data suggests something different is at play. This isn't just a valuation problem driven by market sentiment.
The issue is more structural. The financial arrangements propping up the biggest names in AI have become so self-referential, so deeply intertwined, that the entire ecosystem is starting to look like a closed loop. We’re not just debating if the price is right; we're questioning if the revenue is even real.
The Corporate Ouroboros
The term being whispered in skeptical circles is "circular financing," a sanitized phrase for a troubling mechanism. In its simplest form, it’s when a company invests in its own customers, who then use that capital to buy the investor’s products. It’s a corporate ouroboros—the snake eating its own tail to fuel its own growth. On the balance sheet, it looks like a miracle. Revenue skyrockets. Demand appears infinite. But it’s an illusion of prosperity generated within a sealed system.
Let’s map this out with the players involved. Nvidia, the undisputed king of AI chips, is now the most valuable publicly traded company in the world. Its revenue growth is astronomical. A significant portion of that revenue comes from a handful of major clients, including OpenAI. Here's where it gets interesting: Nvidia is also a major investor in many of its key customers. It holds a stake in CoreWeave, an infrastructure provider that, in turn, supplies OpenAI. Nvidia also expanded its direct investment in OpenAI in a deal reportedly valued at $100 billion.
So, Nvidia gives money to Company A. Company A then uses that money to buy Nvidia’s chips. Nvidia books the revenue. Its stock soars. Is that revenue a true signal of organic market demand, or is it an echo of Nvidia’s own balance sheet?
This isn't an isolated case. Microsoft is heavily invested in OpenAI and is also its biggest cloud provider. Oracle has a $300 billion deal with OpenAI, which is helping fund the colossal Stargate data center project in Texas (a project valued at about $500 billion—to be more exact, a half-trillion dollars). OpenAI, a private company that has never turned a profit, is now turning around and making massive purchases from chipmaker AMD, potentially becoming one of its largest shareholders.
I've looked at hundreds of these filings and deal structures in my career, and this particular web of cross-investments is unusual in its scale and audacity. It creates a feedback loop where investment capital is immediately recycled back as revenue, inflating the top-line numbers of the industry's most critical players. Nvidia’s Jensen Huang defends the practice, stating on CNBC that OpenAI isn't required to buy his chips with his investment. That may be technically true, but it misses the point entirely. The incentives are so powerfully aligned that any other outcome is improbable. When your primary supplier is also a major shareholder, the pressure to "keep the family happy" is immense.
Building Monuments on Phantom Demand
This financial engineering has tangible, real-world consequences. At the Computer History Museum, a place usually dedicated to celebrating past glories, early AI entrepreneur Jerry Kaplan stood before a packed audience and delivered a stark warning. The room was silent as he described living through four previous tech bubbles. This one, he argued, is different because of the sheer magnitude of capital involved. "When [the bubble] breaks, it's going to be really bad," he said, "and not just for people in AI. It's going to drag down the rest of the economy."
His most potent warning was about the physical artifacts this boom will leave behind. Companies are building "enormous data centres in remote places like deserts" based on this seemingly insatiable demand for AI development. OpenAI’s Stargate project, for instance, aims to build a 10-gigawatt complex (a staggering amount of power) by the end of this year. We are pouring concrete and steel based on financial models that may be fundamentally flawed. What happens if the circular financing stops and the "demand" evaporates? We're left with, as Kaplan puts it, "a new man-made ecological disaster," rusting away with no one left to hold accountable.
This model is now going international. OpenAI is actively courting countries like Canada, pitching its ability to build out sovereign AI infrastructure, a move that has prompted questions like, One of the world's biggest AI companies wants a deal with Canada. Is sovereignty the trade-off? On its face, it’s an attractive offer. But it’s an offer built on the same financial logic. The concern for Canada isn't just about data sovereignty under the U.S. CLOUD Act, which gives American authorities access to data held by U.S. firms regardless of location. The deeper question is whether Canada would be buying into a system whose economic underpinnings are suspect.
The old Nortel case keeps coming up in private conversations for a reason. The Canadian telecom giant famously used vendor financing to lend money to its customers so they could buy its equipment. It worked beautifully, until it didn't. The collapse was catastrophic. The parallels aren't perfect, but the core mechanism—creating your own demand—is eerily familiar. This raises the most critical question that no one seems to be able to answer: if you strip out all the vendor-financed and circular deals, what is the actual, organic, third-party demand for AI at its current price point? We simply don't know. And that is the definition of a bubble.
The Numbers Are Telling the Wrong Story
My core thesis is this: the current debate over the AI bubble is focused on the wrong variable. We are arguing about valuation multiples when we should be auditing the revenue itself. The intricate web of cross-investments between chipmakers, cloud providers, and AI labs has created a distorted reality where capital flows in a circle, creating the appearance of explosive, organic growth. It’s a beautiful machine, but it may be running on its own exhaust. The real risk isn't that a few AI startups are overvalued; it's that the foundational metrics we use to measure the entire industry's health might be an accounting fiction. The numbers aren't lying, but they are telling a story that has very little to do with the real world.