The AI Bubble: How Circular Payments Keep the S&P 500 Afloat

The AI Loop: Nvidia and Oracle Are Financing Their Own Customer

Microsoft puts $13 billion on OpenAI's table and keeps 20% of its revenue. Months later, that money shows up in Microsoft's accounts as profit for its cloud division. The thread that kicked off the debate compares it to handing someone €1,300 and recording that you earned €1,300 because they handed it back to you. An accounting joke that, according to this analysis, would be the foundation on which much of the value of the world's biggest tech companies rests.

The pattern repeats piece by piece. Nvidia receives $40 billion in orders from Oracle, adds another $60 billion and delivers them to OpenAI, which in turn uses them to buy Nvidia chips. The money goes full circle and returns to its starting point, leaving in its wake booked revenue, rising shares and growth that exists only on paper.

The Financial Loop of AMD, Nvidia and OpenAI

AMD handed over a warrant worth 160 million shares in exchange for a 10% stake in OpenAI, and the company committed to buying its GPUs. The detail that breaks the scheme is simple: the buyer has no money. OpenAI closed 2024 with losses of $8.5 billion. Its own CEO admitted it won't be in the black until 2029.

Why construct such an elaborate scheme? Because, according to this analysis, the ultimate goal is to inflate the stock. A listed company finances itself with shares or debt, and debt must be repaid no matter what. If the value rises, shareholders applaud. If it rises on revenue that has paid itself, the day someone checks where the money comes from, the applause stops.

The $550 Billion OpenAI Says It Will Pay

The accumulated commitments total $550 billion for the Stargate project, $22 billion for Coreweave and at least $350 billion for Broadcom, plus the $50 billion it says it has already paid. All with annual revenue of $100 billion, according to figures the industry uses, and after a negative 2024. The CEO has already set profitability for 2029, five years out.

Oracle, the Link Creaking Loudest

Oracle signed a contract under which it will receive $300 billion over five years. A timeline that would require OpenAI to generate far more revenue than it does today. Its credit risk indicators have hit historic highs, and its bonds maturing in 2056 yield more than 8%. It is one downgrade away from the edge of junk status: if that happens, some $120 billion in bonds would automatically drop out of debt indices.

The company issued $25 billion in bonds at 6.85% to build the data centers its client needs, with Santander among the underwriters. And its cash is burning: it invests twice what it generates. For its New Mexico project, the so-called Project Jupiter, it even sent a force majeure notice because local authorities twice denied a permit for a 17-mile pipeline that was to feed its fuel cells.

Shadow Banking 2.0: $300 Billion Off Balance Sheet

According to information gathered in the debate, big tech companies keep an exposure of $300 billion linked to AI off their balance sheets through guarantees. For the thread's author, the practice is not new: it would be the same mechanism that fueled the parallel credit propping up the AI bubble and that, in his view, blew up in 2007. The difference he points to is that now the guarantor and the beneficiary share an owner, a customer and a shareholder. When the product being packaged is your own revenue, risk is not diversified: it is concentrated.

What Happens If OpenAI Collapses?

The Magnificent Seven of the S&P 500 account for more than 30% of the market. If those companies correct, they drag down every index fund, generalist ETF and pension plan that tracks the market. The data point that is hardest to dispute: if you strip the tech companies out of the S&P 500, the rest of the index has not grown for two years. The market's rise is not the market's rise. It is the rise of a handful of companies. Volatility is already making itself felt: Microsoft fell as much as 12% at the open of a single session.

Michael Burry, one of the names associated with the 2008 housing bubble, had nearly 50% of his portfolio in bearish options on Nvidia between the first and second quarters. In the third, he closed that position and held ASML and Meta. His public warning, after two and a half years without publishing, points in the same direction: the only winning move is not to play.

Cryptocurrencies and Speculative Assets

The debate extends to assets that produce nothing on their own. For part of the analysis, past crypto rallies are explained less by fundamentals than by a recruitment scheme in which each new entrant supports the previous one. Those who bought in with small regular contributions may discover there is no one left behind to buy them out. The most repeated conclusion: crypto's fundamentals are worth zero.

The Bubble Inside the Bubble

As one debate participant noted, there is a second layer almost no one looks at. AI tools have spent months mass-producing code, reports, minutes, presentations and spreadsheets. Much of that new documentation is built on earlier documentation already generated by AI itself, with no one reviewing it in depth. The result, according to this analysis, is a pile of synthetic material feeding on itself. The problem will not be the volume, but finding someone capable of auditing it.

Is This a Dot-Com-Style Bubble?

The comparison comes up again and again in the thread. The parallel holds in one respect: a real, transformative technology whose value is wildly overstated long before a business model exists to justify it. The difference raised by those taking part in the debate lies in who pays for the party and in the fact that the players now are some of the largest companies on the planet, with banks acting as underwriters.

What Falls and What Holds Up

The distinction repeated most is not between price and value, but between companies. A broad selloff would drag down the share prices of sound companies that will keep selling the same things — food, clothing, basic services — but whose prices catch the panic. For anyone with cash and homework done, that would be the moment to buy quality, not to run for the exits. Companies with stable profits and a dominant position in their market are the ones that hold up best.

While some calculate the crash, others celebrate. A data analytics firm reported a fourth quarter with $1.4 billion in revenue, 70% more than the previous year. The two narratives coexist without friction in the same index.

And the data point that throws you off does not come from a promise about the future, but from a registered prospectus: one of the big AI firms filed for its IPO with $4.6 billion in revenue in 2025, $8 billion in operating losses and a plan to spend $518 billion on cloud and infrastructure over the coming years. The numbers, on the same page, do not speak to each other.