The IMF Called the AI Trade a Global Stability Risk. Its Own Growth Forecast Depends on It.
On October 13 the IMF's Global Financial Stability Report named three channels through which an AI stock correction could spread into the financial system: leverage, circular financing between hyperscalers and weaker AI firms, and concentrated exposure to US AI stocks. The day before, its managing director said AI infrastructure investment is what is keeping global growth near 3 percent. Here is why the two statements describe one machine, what the same day's report of companies scaling back AI rollouts adds, and the one IMF channel you can check for yourself in a deal record.
In July we wrote that the AI boom was starting to fund itself, and that this should worry you. On October 13 the International Monetary Fund put the same worry in its Global Financial Stability Report.
## What the IMF said, twice
The report warned that a sharp correction in AI-related equities could trigger global market turmoil, and named the channels: heavy leverage, circular financing between hyperscalers and weaker AI firms, and high exposure to US AI stocks. The same day the fund nudged its 2026 global growth forecast up to 3.1 percent, with a warning that surging investment and speculation around AI, along with rising debt, pose mounting risks.
Read the three together. The spending that holds up the growth forecast is the spending whose financing the stability report calls a risk. The IMF is not contradicting itself. It is describing one machine from two sides: the capital expenditure is the growth, and the way it is being financed is the fragility.
## The channel you can check yourself
Two of the IMF's three channels are hard to see from outside. Leverage sits on bank and fund balance sheets; concentration shows up only when a correction starts. The third, circular financing, is visible in the deal record, and we have been tracking it since July.
On July 29 we wrote that the chip vendors were financing their own customers, and suggested one question: ask who funded the purchase order. On September 2 we counted about $90 billion of compute commitments Anthropic made in four weeks without buying a single computer, and noted the difference between capex and a commitment: capex can be cut in a bad quarter. In September, Fitch modeled a scenario in which an AI investment bubble bursts, Wall Street falls 35 percent in six months and the US enters recession in 2027. The IMF now lists the circle as a transmission channel. Our deal tracker lists the source and the target on every deal, which makes the circular ones easy to spot once you look for them.
## The demand side moved the same day
The stability report is about the supply of capital. The Guardian's Reworked series reported on October 13 that Meta, Microsoft and Walmart are tightening internal AI use after early deployments flooded workflows with errors, low-quality content and security concerns; executives and HR leaders say they are reinstating managers, restricting certain tools and rethinking expectations after AI-linked productivity gains failed to materialize.
That is the other end of the machine. The capex is justified by productivity gains at the customer, and the IMF's own September paper for EU finance ministers put the European gain at about 1 percent over five years. If large employers are pulling tools back, the revenue that is supposed to repay the compute commitments arrives later, and the leverage and the circular deals have to carry the gap for longer.
## What the exposure looks like in one account
Since October 12 this site's homepage has been a hypothetical, long-only paper portfolio of AI stocks, run by a model under rules enforced in code. It was seeded from the AI holdings of a real brokerage account and started at $25,975. After two trading days it holds 17 positions and 36 percent cash, is down 0.33 percent against a 0.07 percent fall in SPY, and has made no trades. The model's only proposal on both days was to trim Palantir, inherited at 17.4 percent of the portfolio, down to the 15 percent cap. The rules engine rejected it both times because the order, about $620, was below the $1,000 minimum.
So the house portfolio is more concentrated in one AI stock than its own rulebook allows, and the rulebook, not the model, is what keeps it there for now. We publish it so the exposure the IMF describes has a worked example with every number on the page.
## Hedges
The IMF report reaches us through the Guardian and AP; we have not quoted the report's own text. The Guardian's rollback piece reports what executives and HR leaders said, not a survey with a number. Fitch's scenario is a scenario, not a forecast. Two trading days of a paper portfolio say nothing about skill; we cite it for its exposure, not its results.
## What to do with this
First, take the IMF's three channels and write down which one you are exposed to. Leverage and concentration you will have to ask your broker or fund about. Circular financing you can check today: for each AI deal in your thesis, look up who sits on both sides. If the investor is also the supplier being paid, that deal belongs in the IMF's third channel, whatever the press release calls it.
Second, watch the demand side for the turn, not the supply side. Capex guidance is committed in advance and moves last. The signal is whether the Guardian's three companies are followed by a fourth with a number attached: seats cancelled, tools withdrawn, a productivity target missed. Microsoft, Meta, Nvidia and Palantir are on our tracker.