The Fed Now Holds Two Prices for the Same Machine

On October 13, at the IMF and World Bank's annual meetings in Bangkok, the Fund's stability report priced the worst case for artificial intelligence: an investment bust that shaves 1.3 percentage points off global growth by 2028, funded by debt that has grown from under 1% to 10% of all investment-grade bond issuance in two years. ^[1] Days later, at the same meetings, Anthropic's co-founder priced the same asset the other way: AI expanding U.S. GDP to $44 trillion by 2030, from just over $30 trillion now. ^[2] Standing between the two documents was the Federal Reserve chair, who declined at the venue to say anything about the next rate decision. ^[3] Both receipts are already in the Fed's own file. By the Fed's own attribution, AI infrastructure investment drove between 35% and 65% of U.S. GDP growth in the first quarter of the year, and data-center construction has overtaken consumer spending as a contributor to growth. ^[4][5] The same file shows the engine throttling: high-tech equipment investment decelerated from 39.9% annualized in the first quarter to 6.9% in the second, with overall growth at 1.5%. ^[4] The Fed's June minutes name the same buildout as an inflationary pressure, with sustained demand for AI infrastructure keeping upward pressure on tech and electricity prices even as the Fed's preferred inflation gauge (PCE) rose from 3.8% to 4.1%. ^[6] The growth camp's hostage notice came the same week from BlackRock's chief executive. ^[7]

If we have a moratorium of some sort in this country, we will be in a recession. — Larry Fink

In the same appearance he ruled out the other price.

So much of our economy is based on the buildout of this capex. — Larry Fink

Turn from the Fed's file to the IMF's and the same asset looks like a liability. The debt fronting for the buildout has reached a tenth of investment-grade issuance in two years, strung through private credit funds into a leverage chain that could reach commercial banks and pension funds. ^[1] The cost of insuring that debt against default — traded as credit-default swaps — sits at record highs for the big tech names: five-year protection on CoreWeave, a neo-cloud that rents AI computing but borrows to build, passed 1,100 basis points, and Oracle was cut to one notch above junk. ^[8] And the edge is still being built — roughly $150 billion in chip-related debt lined up in a single week, including the package Broadcom arranged for OpenAI's custom chips and the $60 billion it syndicated for Anthropic. ^[9] Broadcom's chief executive, whose company arranged two of those packages, framed the problem in credit terms. ^[9]

This is not the first time we've had a sizeable market wobble on the back of AI fears, but as we move to a capex cycle funded out of debt rather than simply free cash flow, we have to increasingly respect the credit price action. — Hock E. Tan

The tell came this week. The 10-year Treasury yield posted its largest weekly drop in more than a year — not because inflation cooled, but as investors sought safety amid AI-debt concerns while inflation and oil stayed high. ^[10][11] The bond market is now pricing AI financing risk as a bigger driver of the cost of money than inflation itself. ^[11] One counter stays alive, so this reads as a collision between prices rather than between a sage and fools. Nvidia's chief executive disputes the instability account outright ^[12], and UBS still projects memory-chip shipments growing 49% by 2027 on demand for high-bandwidth memory. ^[13] The collision is cleanest in the chair's own sentences, one from the summer, one from this week. The bull case came in July.

Many participants noted that ongoing strong demand for AI infrastructure would likely sustain upward pressure on prices for technology products and electricity. — Federal Reserve System

The other came this week, in Bangkok.

It's obviously hitting the demand side of the economy before it's going to hit the supply side. — Kevin Warsh

That is the trap, laid out in the Fed's own papers: inflation says tighten, and the growth engine is the same machine the tightening would squeeze. Markets still price an 83% chance the Fed holds at its October 27–28 meeting and an 88% chance of a December hike — cornered, not paralyzed. ^[10] The one thing that could reconcile the two prices — a measured productivity payoff — is still unproven. The Fed's own staff economists, in a 2025 report, left the question open. ^[14]

The case that generative AI is a general-purpose technology is compelling, supported by the impressive record of knock-on innovation and ongoing core innovation. — Federal Reserve Board of Governors

The $44 trillion, meanwhile, turns out to be Anthropic's own "extreme scenario," a model that needs recursively self-improving AI and unemployment near 14%. ^[15] The IMF's chief gave the referee's call.

We have very high expectations. They are not going to be met because it takes time for diffusion. — Kristalina Georgieva

A rate number gets set on October 27–28. The evidence that would reconcile the two prices still isn't measured, and the debt fronting for it reprices daily. ^[10]