The five biggest cloud providers and the lenders financing their AI buildout named the risk in the same two week window, with five signals worth tracking and three questions worth asking your AI vendor.
The five largest cloud providers, including AWS, Microsoft Azure, Google Cloud, Oracle, and Meta, are on track to spend more than $1 trillion on AI infrastructure across 2025 and 2026, according to the Bank for International Settlements' Annual Economic Report 2026. That figure has been in headlines for a year. What changed in late June is who started saying it out loud, and in what document.
In the same two-week window, the BIS, the central bank of central banks, published the chapter "Progress and peril," warning that this capex is outpacing the earnings and free cash flow of those firms, with some already issuing debt to finance the build-out. Separately, Oracle filed its Form 10-K for the fiscal year ended May 31, 2026 and became the first major hyperscaler to put a quantified "could lose the farm" frame around its AI infrastructure bet in a primary regulatory disclosure. The Register cataloged the risk-factor language on July 1; Fortune summarized the BIS report on June 29; and the same outlet's July 6 roundup is what put the convergence on the desk of anyone who follows this beat.
The lenders underwriting the AI buildout and the hyperscalers running it are the ones with the most to gain from continued optimism. They are now the ones publicly hedging their own thesis, in primary disclosures that are harder to walk back than a press quote. The wire summaries frame this as "experts are worried." The more useful read is that the people long the trade are the ones introducing doubt.
The BIS does not predict a crash. It does something more useful: it names the mechanism. Chapter I of the 2026 report draws a direct line from prior capital manias, including the dot-com era, noting that those episodes "shared one common trait: a genuine technological breakthrough that attracted capital in excess of what commercial returns could ultimately justify." The same chapter flags the specific transmission channel: "Disappointment in returns could trigger a sudden pullback in financing and turn the capex boom into a protracted investment bust." A $1 trillion spending wave financed partly by debt, against revenues that have not yet caught up, is a debt-structure problem before it is a forecast problem.
Oracle's 10-K is a different kind of signal, and a different kind of disclosure. Risk-factor language in a SEC filing is not management prediction. It is, however, a legally certified inventory of what could go wrong, and the AI-infrastructure risk section is unusually long. The Register's read of the filing is that Oracle has effectively told shareholders the AI bet is now a binary outcome. That is not Oracle saying the bet will fail. It is Oracle saying the SEC's standard for material risk disclosure is now met by the bet's failure modes.
For anyone signing a multi-year AI contract, deploying GPU capacity, or modeling capex exposure, the practical question is what to watch. Five signals are worth tracking over the next two earnings cycles:
Three questions worth putting to your own AI vendor, cloud provider, or finance and treasury team:
The BIS report and Oracle 10-K do not say a bust is coming. They say the people long the trade are now formally flagging the risk in documents they cannot easily walk back. The next earnings season is the first test of whether the words match the disclosures.