One leg, one trillion dollars of debt: the IMF’s warning about AI financing

The world economy has narrowed to a single source of growth, and the institution that watches cross-border stress has begun to say so in public.

Kristalina Georgieva, managing director of the International Monetary Fund, told CNBC this week that the global economy is “standing on one leg”, and that the leg is artificial intelligence. If that leg shakes, she said, the result is a financial stability problem rather than a technology problem.

The mechanism she pointed to is a loop. The largest technology companies invest in one another, then spend the money on one another’s chips and cloud capacity. Every transaction in that loop is real and booked at a price. What is unproven is the end demand underneath it, which is the part that eventually has to pay for all of it.

The size of the loop is why the fund is talking. The IMF puts the chain above $1 trillion, with about two-thirds of that debt rather than equity. Capital commitments by large technology companies for AI already exceed $700 billion for 2026 alone.

Debt changes the character of a boom. A build-out funded by equity can be written down quietly when a project disappoints. One funded by borrowing carries maturities, covenants and creditors who cannot simply decide to wait.

The transmission channels the fund listed run well past the data centre. Semiconductors first, then electricity generation and the grids that carry it, then construction, and from there into the finances of the counties and towns where the campuses are being built.

That last link is where an investment story becomes a public finance story. Local governments have traded tax abatements and land for jobs and property tax revenue on the assumption that the buildings keep filling with servers. The assumption now has a date attached to it.

No single supervisor sees the whole chain: chip supply crosses export control regimes, the borrowing sits with banks, private credit funds and bondholders in different jurisdictions, and the power contracts are settled by state utility commissions. Private credit has also made the debt harder to price, because loans held by funds rather than banks are marked on a calendar instead of a screen. The first sign of stress may arrive as a valuation notice rather than a falling price.

The mismatch between asset life and loan life is the part analysts keep returning to. Most industry estimates put the useful accounting life of a graphics processor at three to six years, while the debt raised against the buildings that house them often runs longer. A project can be fully financed and still fail on that arithmetic.

The fund has made similar points before. Its global financial stability reports in recent years warned that equity markets had become unusually dependent on a handful of companies, and that disappointment at one of them would be felt in every index that holds the rest. What is new is the claim that the dependence has moved out of portfolios and into the real economy.

There is a fair counter-argument and it deserves a hearing. The companies at the centre of the loop generate cash at a scale with little recent precedent, and much of the build-out is funded from operating profit rather than borrowed money. A slowdown in spending would hurt suppliers and shareholders; it would not by itself break a bank.

The fund’s answer is that the two-thirds debt share is what turns a spending slowdown into a credit event, and credit events do not stay where they begin. A missed payment at a power supplier, a downgrade at a construction group, a county that cannot service the bonds it sold for a substation: each is small, and each is a link.

Geopolitics sharpens the problem. Export controls have already split the chip supply chain into permitted and prohibited buyers, so the loop is less global than the revenue figures imply. A company selling into a restricted market is holding inventory it may not be able to move at the price it paid.

The question that follows is not whether AI investment is excessive. It is whether the financial system has priced the possibility that the spending stops faster than the debt matures. On the fund’s arithmetic the answer is no, because the pricing assumes the loop keeps turning.

The managing director’s own framing was narrower than a forecast. She described a confidence shock and a collective withdrawal of funding as the trigger, and neither requires the technology to fail. Both are decisions, and decisions can be made inside a week.

That is the uncomfortable part for policymakers. The tools available to a central bank address liquidity, not conviction, and none of them can restore a buyer’s belief that the end demand is finally there. The fund is not calling a crash, and it said as much; it is pointing out that an economy which has removed its other legs has less room to absorb a stumble in the one that remains.

Related Posts

  • November 20, 2026
  • 15 views
Sandisk Completes Eight-for-One Split as NAND Outlook Darkens

Eight for one. Sandisk’s stock split took effect on Nov. 19, and the shares began trading on the adjusted basis on Nov. 20, a change that divides the price by…

  • November 20, 2026
  • 13 views
A Lawmaker’s Privacy Suit Puts Model Design on Trial

For three years British courts have sidestepped the question of whether the company that builds an AI model answers for what its users produce with it. A written defence filed…