The Fed Calls AI a Paradigm Shift While Filtering Out Its Price Signals

On July 1, at the European Central Bank's Sintra forum, Fed Chair Kevin Warsh made a declaration no sitting Fed chair had made before.

This is a big paradigm shift both for the conduct of our policy and for our economies. — Kevin Warsh

He placed artificial intelligence inside the monetary-policy frame, treating it not as a sectoral investment story but as a variable in how the Fed sets interest rates ^[1]. The BIS, the central bankers' central bank, reached the same conclusion in its June annual report, finding that AI supply-side bottlenecks in semiconductors and electricity are already hitting consumers and adding to inflationary pressures ^[2]. Morgan Stanley had told clients weeks earlier that AI had crossed from a tech story into a macroeconomic force ^[3]. The price signals are not subtle. In May, U.S. CPI hit 4.2 percent. DDR5 and DDR4 memory rose 290 percent year over year, producer prices for electronic components climbed a record 27 percent, and computer software and accessories prices jumped 14.5 percent — the largest such increase since 2000 ^[4]. Nine U.S. industry trade groups, including the Alliance for Automotive Innovation and the National Retail Federation, warned Treasury and Commerce that AI data centers are consuming a disproportionate share of memory chip capacity and threatening sustained near-term price increases for American households ^[5]. Apple raised prices on Macs, iPads, and the Vision Pro by as much as 50 percent. The company's CEO told analysts the speed and magnitude of the component price surge had no precedent.

We have never seen a component price increase this much, this quickly. — Apple

The chain runs from data-center demand through memory scarcity to consumer checkout. But Warsh's preferred analytical response to this inflation is to strip those signals out of the data. He is advocating trimmed-mean inflation averages, a method that drops the items with the largest price movements — up or down — before computing the average. His description of what the method does is candid.

What I’m most interested in is what’s the underlying inflation rate, not what’s the one-time change in prices because of a change in geopolitics or a change in beef. — Kevin Warsh

He told Congress he cares about the underlying inflation rate, not one-time price changes driven by geopolitical events ^[6]. The difficulty is that AI's inflationary channel looks exactly like the kind of extreme movement the method discards. Memory up 290 percent, component prices up 27 percent, Apple raising hardware prices by as much as 50 percent — these are extreme by definition. In a trimmed-mean calculation, they get trimmed. Dallas Fed President Lorie Logan, one of the system's most respected inflation technicians, put the risk plainly.

But if recent trends continue, it may soon be appropriate to act. — Beth M. Hammack

The contradiction sits right there: the Fed chair has named AI a paradigm shift for the conduct of policy, and the inflation method its chair is advocating is mathematically calibrated to filter out the evidence of that shift. No major central bank or international institution has built a framework to separate AI's inflationary channels from its deflationary ones — or from the concurrent Iran oil shock, which sent energy costs up 23.5 percent in the same May CPI print ^[4]. Each institution is reaching for AI through the lens it already owns:

How each institution frames AI

BIS: Bubble risk comparable to the dot-com boom and 1830s canal mania, despite identifying semiconductor and electricity bottlenecks as inflationary ^[2]

IMF: Growth cushion that has mitigated the impact of the Middle East energy crisis, with a warning that frothy valuations could correct sharply if productivity gains fail ^[7]

Bank of England: Financial stability risk — testing AI agents for herding behavior; concluded AI does not yet pose systemic risk ^[8]

U.S. Treasury: Sectoral policy fight — career analysts' bubble warning dismissed by Secretary Bessent in favor of the administration's Golden Age narrative ^[9]

The deflationary channel exists — AI-driven layoffs reached 87,714 in the first five months of 2026, surpassing combined 2024-2025 totals ^[10]. But the expected productivity dividend is arriving slowly and unevenly. AI computing costs at some firms now exceed employee salaries ^[11], and companies are spending on internal GPU infrastructure to escape cloud token pricing — capital spending that feeds the inflationary channel while attempting to capture the deflationary one. The Fed's own regional presidents are already split. Cleveland Fed President Beth Hammack explicitly named AI capital expenditures alongside the Iran energy disruption as primary inflation drivers and warned rates may need to rise ^[12]. New York Fed President John Williams said the next day that policy is well-positioned and inflation will peak soon ^[12]. The disagreement is about whether AI's inflationary channel is large enough to act on — a question the method Warsh advocates cannot answer because it is built to remove the evidence. Meanwhile, the spending driving AI's inflation channel is still accelerating. Hyperscaler capital expenditure is estimated near $725 billion in 2026 ^[13]. ERCOT, the Texas grid operator, received 519 data center interconnection requests in two years that, if fully approved, would demand roughly one-third of total U.S. power generation capacity, forcing it to ration connections ^[14]. A $2.3 trillion tech sell-off in June raised the possibility that capital markets might impose the discipline the grid cannot ^[15]. They have not. Alphabet raised $80 billion in equity for AI investment amid the volatility ^[16], Meta and Microsoft are planning combined 2026 capex above $250 billion despite stock declines of 12 and 20 percent ^[17], and SpaceX executed a record $85.7 billion IPO pivoting to AI infrastructure ^[18]. Equity skepticism is not translating into cutbacks. Warsh has said he wants to deploy new technologies for real-time economic monitoring within nine to twelve months ^[19]. But his five task forces are deploying AI as a data-collection tool, not building a framework to decompose AI as an inflation variable. He is simultaneously ending forward guidance — the practice of telling markets where rates are headed — which narrows the Fed's analytical transparency at the moment it most needs it. The chair who called AI a paradigm shift is investing in AI-as-tool while not building AI-as-variable analysis, and the method he is advocating for reading inflation is one his own Dallas Fed president warns will miss the shocks the paradigm is producing. The question is whether the constraints on AI spending — the power grid, the capital markets, the capability ceiling — slow the inflationary channel before the measurement gap becomes a policy error. None has bitten yet. The 4.2 percent CPI, the 290 percent memory spike, and the Apple price increases are arriving in real time, and the Fed's instrument is calibrated to call them noise. The last time the Fed dismissed a broad price surge as transitory, the error was visible only in retrospect. This time the warning is already on the record, inside the institution, before the decision is made.