The AI Trade’s Circularity Problem Just Got a Warning Label From the Bank for International Settlements

When the Bank for International Settlements, the institution that functions as a central bank to the world’s central banks, devotes a section of its flagship annual report to artificial intelligence, that is worth more attention than another earnings-season sound bite about GPU demand. That is what happened on June 28, when the BIS released its Annual Economic Report 2026 and named an AI capital spending bust, the unwinding of circular financing arrangements, and fragile sovereign debt as three of the developments most likely to shake the global financial system. The report compared today’s buildout to canal mania in the 1830s and the dot-com crash of the early 2000s, and it warned that a sharp repricing of AI-related stocks could hit consumer spending harder than past corrections because so much household wealth is now tied up in a handful of AI-linked names. Coming from an institution whose entire job is spotting systemic risk before it shows up on a trading screen, that is not a throwaway line.

None of this means the AI buildout is fake or that the technology is not real. It clearly is. But the financial plumbing behind the spending has gotten complicated enough that even the people who worry about financial plumbing for a living are starting to ask pointed questions. So it is worth walking through what is actually happening, why Oracle has become the poster child for the anxiety, and what the next few weeks of earnings reports need to show if the bulls are going to keep winning the argument.

What “Circular” Actually Means

The AI ecosystem now runs on a small set of companies that are simultaneously suppliers, customers, and investors to one another. Nvidia sells chips to hyperscalers and also takes equity stakes in AI labs. Those labs then sign multibillion-dollar compute contracts with the same hyperscalers, whose capital expenditure in turn shows up as revenue for Nvidia. Microsoft, Amazon, Google, Meta, OpenAI, and Anthropic all touch multiple points in that loop at once. The BIS describes this as circular financing: money moves in a closed system where each participant’s spending becomes another participant’s revenue, which makes growth numbers look organic even when a meaningful share of the demand signal originates inside the loop rather than from end customers.

There is nothing inherently improper about this. Vendor financing and strategic equity stakes are old tools. The concern is what the BIS flagged directly: every major player is making a similar, enormous bet at roughly the same time, on the belief that only a few winners will end up dominating the market. That dynamic, the report argues, is exactly the kind of setup that produces collective overcommitment, where firms plow money into projects with genuinely uncertain returns because the fear of being left behind outweighs the fear of overbuilding.

Oracle Is the Warning Arriving Early

If you want to see what happens when the market stops taking the circular story on faith, look at Oracle. The stock fell roughly 25% during the first half of 2026, and it just logged its worst week since the dot-com bust of 2001, according to CNBC, as investors focused on the size of the bet the company has made on AI infrastructure. Oracle is carrying more than $122 billion in long-term debt and has said it plans to raise around $40 billion more through a combination of debt and equity to keep building data centers, on top of nearly $250 billion in long-term data center lease commitments running 15 to 20 years, according to reporting from the Motley Fool.

The number that really got Wall Street’s attention is Oracle’s remaining performance obligations, essentially contracted future revenue that has not been delivered yet. That figure hit about $638 billion, up 363% from a year earlier. Sounds fantastic on the surface. The catch is timing: Oracle expects to convert only around 12% of that backlog into revenue over the next 12 months, with another 34% over the following two years, which means less than half of that giant number becomes actual revenue within three years even in a reasonably favorable scenario. Layer on top of that a roughly $300 billion multiyear deal to supply data center capacity to OpenAI, a single customer whose own path to profitability is still being written, and you can see why investors are asking whether Oracle’s growth depends on one counterparty’s ability to keep paying its bills.

Oracle is not the whole AI trade. But it is the clearest example so far of what the BIS is worried about: debt piled onto long-dated infrastructure bets, backed by contracts with customers whose own finances are not fully independent of the same ecosystem.

The Bull Case Has Not Disappeared

It would be a mistake to read all this as proof the AI boom is a mirage. J.P. Morgan Asset Management published a rebuttal of sorts to the bubble comparison, arguing the more useful question is not whether today’s deals resemble the late-1990s but whether the underlying fundamentals do. Its analysts point to three real differences. First, the largest hyperscalers, Microsoft, Amazon, Alphabet, and Meta, are funding most of this buildout out of their own free cash flow and healthy margins rather than relying on external credit the way telecom firms did in the 1990s, which historically made those earlier booms far more fragile once credit tightened. Second, AI is monetizing while it is still being built. OpenAI, for instance, has gone from near-zero revenue a few years ago to roughly $13 billion, with a stated target of $200 billion by 2030, an extraordinary ramp if it holds. Third, unlike the dot-com era, when only about 7% of fiber-optic capacity was actually being used, data center vacancy rates are near record lows and utilization is running around 80%, meaning demand for compute is not obviously running ahead of what is actually being consumed.

Those are legitimate points, and they explain why plenty of serious investors are not fleeing the trade. But notice where the reassurance is strongest: at the biggest, best-capitalized companies. It gets shakier the further you move toward the edges of the ecosystem, toward neoclouds, leveraged infrastructure vehicles, and companies like Oracle that are financing growth with borrowed money rather than free cash flow. The AI trade may not be one bubble. It may be a core that looks reasonably sound wrapped around a periphery that looks a lot more fragile, and the two get lumped together every time the word “AI” appears in a headline.

What Q2 Earnings Need to Prove

That brings us to the next few weeks, which matter more than usual. The four largest hyperscalers are on track to spend a combined $725 billion on AI infrastructure in 2026, up roughly 77% from about $410 billion in 2025, based on aggregated guidance and analyst tracking reported by Yahoo Finance. Microsoft and Meta are both expected to report second-quarter results on July 29, with Amazon following on July 30 and Alphabet reporting in the same window. Those four calls will do more to settle the bubble debate, at least for this quarter, than any think tank report.

What investors should actually watch for is not just the size of the next capex number, which will almost certainly go up again. It is whether cloud revenue growth and AI-linked product revenue are accelerating at a pace that closes the gap with spending, whether management teams offer anything concrete on return on invested capital rather than vague language about long-term opportunity, and whether any of the four start sounding more cautious about the pace of new commitments the way Oracle’s costs have forced it to. Nvidia’s own commentary on customer concentration and order backlog, expected around the same window, will add another data point on whether the demand underpinning all of this spending is broadening out or still resting on the same handful of names.

None of this requires picking a side today. The BIS is not saying the AI boom will collapse next quarter, and JPMorgan is not saying it is riskless. What both are really pointing at is the same thing: the AI trade has grown large enough, and financially interconnected enough, that its risks are no longer contained to a handful of software stocks. That is exactly the kind of moment that rewards paying closer attention to balance sheets and cash flow statements, not just headline growth numbers, before deciding how much of this story to own.

____________________________________________________________________________________________________

This article is written for educational and informational purposes only and does not constitute financial or legal advice. The views and analytical frameworks presented draw on publicly available information and reported commentary from industry participants. Readers are encouraged to consult primary sources and form their own informed views on these complex topics.