New York Fed's Williams says AI hardware supply may lag demand beyond two years

Williams put a condition on his own two-year estimate. Asked whether AI supply and demand would become more balanced over the next two years, he said: "I think it's going to take longer. Assuming the demand kind of impulse continues as expected, and that's a big assumption." ^[3] The supply gap is therefore a forecast of how long orders keep arriving. If they slow, the gap closes sooner and the price pressure he describes closes with it. Neither account says how far chip prices have risen.

On rates, he described a sector that mostly ignores them. "I think investors there are relatively price insensitive," he said ^[6], and he added: "I don't expect monetary policy to be really having a deep effect on that." ^[7] Elsewhere in the interview he said the labor market is not contributing to price pressures and that underlying inflation measures have already moved lower ^[16], and, in his words, "This is not an economy that is overheating overall." ^[9]

Futures are less patient than he is. They reflect a December increase and two more in 2027 ^[14], which is three moves against his one ^[17], or rather a gap of two hikes, since both counts include this year's. Of anything beyond the next hike, Williams said: "I don't think at this time I have any reason to believe that we would need to do more." ^[5] Cleveland Fed President Beth Hammack and Dallas Fed President Lorie Logan have each signaled that more than one additional move will be needed, according to Bloomberg ^[13].

His path can fail in at least three ways. AI demand could stall, which is the "big assumption" he flagged. The tariff and energy effects he expects to fade ^[8] could linger. Or the hawks could carry the committee.

Our view is that if chip and related-goods prices stay high for more than two years ^[1] and rates do not slow the orders behind them ^[7], that slice of inflation does not respond to the hike, and the 2027 question stays open whatever happens in December. The counter-thesis is Williams's own: AI-driven productivity gains are disinflationary in the end, and underlying inflation has already come down ^[16]. We would be wrong if AI orders slow early, or if underlying measures keep falling while chip prices stay up.

The financing side sits awkwardly with "price insensitive." Cointelegraph reported on Monday that miners pursuing AI projects already face substantial financing requirements and increasingly selective lenders ^[20], even as IREN signed $2.8 billion in new AI cloud contracts in July ^[19]. Williams was describing AI investors in general. Lenders to miners are a narrower group, and both descriptions can be true at once.