· AI
The IMF Found 1.4 Trillion Dollars in Circular AI Deals. Its Fix Targets the Wrong Lenders.
The risk the IMF found is real. The lender it wants to restrain is not where that risk actually sits.
The International Monetary Fund put a number on something people have been gesturing at for a year. In its October 2026 Global Financial Stability Report, the IMF reviewed 14 publicly disclosed AI financing deals struck since 2023. Hyperscalers and other investors put roughly 300 billion dollars of equity and 650 billion dollars of debt into financially weaker AI firms. In exchange, they locked in 1.4 trillion dollars in future revenue, mostly through chip orders and service contracts, as reported by TheNextWeb. About 320 billion dollars of that debt sits with the firms receiving the money, concentrated in a handful of them.
My position: the risk the IMF describes is real and worth taking seriously. The fix it is reaching for, bank-side concentration limits and sectoral capital requirements, is aimed at the part of the financing stack that already carries the least of this exposure. The part that carries the most of it is the part banks barely touch.
Start with the IMF's own case, because it is a good one. The core danger it names is what IMF assistant director Jason Wu called a "triple whammy": equity and debt tied to an AI firm's revenue can both lose value at the same moment that revenue gets impaired, since the investment and the sales contract are often the same deal viewed from two sides. The report's own framing is that a correction could be made worse by leverage sitting on top of exposures already concentrated in a few firms. Its stress test found that under a scenario combining weaker AI productivity gains with higher bond yields, banks holding 22 percent of global banking assets would fall under their post-stress capital thresholds, up from 18 percent a year earlier, a group reported elsewhere as roughly 110 banks. Hyperscalers now account for about 10 percent of global investment-grade bond issuance, up from under 1 percent in 2024. None of that is alarmist. It is a specific, falsifiable claim about where leverage sits.
Where I think the IMF gets the remedy wrong is in where that leverage actually sits. Its recommended tools, drawn from the report and summarized by Unite.ai, amount to restricting margin financing and leveraged ETFs on the trading side, plus capital requirements and concentration limits on bank lending. But look at how the biggest deals in the IMF's own dataset are actually structured. UBS's September 2026 credit analysis describes a roughly 35 billion dollar deal led by Apollo and Blackstone in which a special-purpose vehicle bought about a million Google AI chips and leased them to Anthropic, with Broadcom backing the senior notes and Google backstopping the site leases. Nvidia has lined up six Wall Street firms, Apollo and Blackstone among them, to mobilize more than 500 billion dollars on similar terms. That is private credit and direct bond issuance, built specifically to avoid sitting on a bank's balance sheet. A concentration limit written for banks does not reach it.
Meanwhile the credit discipline the IMF wants is already showing up where it belongs: in price. UBS tracked a ten-year hyperscaler credit basket widen from about 45 basis points in July 2025 to nearly 90 basis points by July 2026, and reports that a recent Meta-linked bond sale drew only lukewarm demand. Oracle, rated BBB minus, saw its remaining performance obligations jump from 138 billion to 638 billion dollars in a year, much of that resting on OpenAI, a company UBS notes burned 3.7 billion dollars in a single quarter and is not expected to be cash-flow neutral before 2030. Lenders are not ignoring this. They are charging more for it, deal by deal, which is what a functioning credit market does before a regulator has to.
A blanket concentration limit on bank lending would land hardest on the AI infrastructure players who still need banks: smaller data center operators and neocloud providers without a hyperscaler's balance sheet or a Blackstone-sized private credit line. The companies the IMF is actually worried about have already routed around the tool meant to restrain them.
If your business leans on an AI infrastructure vendor, the lesson is not to wait for the IMF to referee this. Ask that vendor how concentrated its revenue is and who backs its balance sheet, the same diligence you would run on any supplier before depending on them. That is a conversation worth having before you sign a multi-year contract, not after a credit spread answers it for you. If you want a second opinion on a vendor decision like that, that is what fractional AI advisory is for.
Sources
References used in this article. Links also appear alongside the relevant claims.
Let's make it happen.
You don't need a polished brief. A couple of lines about where you want your company to go is plenty, and we'll come back with what we'd tackle first.
We scope the work around your goals and timeline before quoting anything, so you know exactly what you're getting.