The Machine the Fed Built

In the same room, on the same day, in the same set of minutes, the Federal Reserve held two opposite convictions about artificial intelligence.

a significant disinflationary force — Kevin Warsh

the biggest productivity wave our generation will experience — Kevin Warsh

Kevin Warsh called it a significant disinflationary force, the biggest productivity wave our generation would experience. The same committee document reached the opposite conclusion.

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

One diagnosis said AI would make everything cheaper. The other said it was already making critical inputs more expensive ^[1]. The Fed was of two minds. Then it wasn't. Over the six weeks that followed, the central bank built a machine designed to treat AI infrastructure spending as a distinct inflationary force, one requiring monetary policy action rather than patience. Each institutional move narrowed the range of possible responses. The architecture is now largely in place, and it points at the one category of spending keeping U.S. economic growth above water. The construction began on June 14, when Warsh announced what he called a regime change in how the Fed communicates. He shortened the policy statement, removed forward guidance, and eliminated the dot plot — the quarterly chart of where each policymaker expects rates to go. His reasoning was explicit.

I think the financial markets work less efficiently when they ask a question, 'How will the Federal Reserve react to that incoming information?' — Kevin Warsh

In a fast-moving economy shaped by AI investment, Warsh argued, telling markets the plan in advance was a liability. The reform gave the Fed room to move without telegraphing its next step ^[2]. Then came the task forces. Warsh created five expert advisory groups, including one charged specifically with studying the impact of AI infrastructure on prices, with recommendations due by December ^[3]. He appointed Marc Andreessen, former RBI governor Raghuram Rajan, and Stanford economist Raj Chetty, among others, to institutionalize the AI-inflation question inside the Fed's policy apparatus ^[4]. The task force came with a deadline: December, the same month markets now assign roughly 90% probability to a rate hike ^[5]. The committee itself crossed a threshold in the June FOMC minutes. For the first time, the Federal Open Market Committee as a body — not an individual governor, not a staff paper — named AI infrastructure spending as a source of inflationary pressure. The minutes cited ongoing strong demand for AI infrastructure driving up prices for technology products and electricity ^[1]. The language was careful. The institutional signal was not: AI capex had moved from a corporate growth story the Fed passively observed to a macroeconomic variable it was preparing to counteract. On July 15, three governors delivered the same message in unison. Lisa Cook quantified it.

If we do not see signs of disinflation soon, I am prepared to act. — Lisa D. Cook

She was prepared to act if disinflation did not materialize. Christopher Waller invoked 2021, the year the Fed waited too long, as the cautionary tale. Warsh, who weeks earlier at Sintra had celebrated AI as a paradigm shift and predicted business sentiment would turn positive within six months, now declared simply that prices were too high ^[5][6]. The coordinated signaling was itself the message. The Fed had stopped debating whether AI infrastructure spending was inflationary and started deciding how hard to push back. The paradox is not that the Fed is wrong about the inflation. It is that the inflation and the growth share a single engine. In the first quarter of 2026, U.S. GDP grew 2.1%, but the number concealed a lopsided economy. Business investment surged 39.9%, driven almost entirely by AI data center equipment. Consumer spending grew 0.5%, the weakest in four years. Residential investment fell 7.8% ^[7]. AI capital expenditure is, right now, the only component keeping the expansion alive. The same spending that the Fed's new institutional machinery is designed to suppress, by making debt-funded data center buildouts more expensive, is the spending that prevented the first quarter from printing negative. And the White House is demanding the opposite of what the Fed is building toward. Trump declared that low interest rates would solve everything, pushing for cuts that would accelerate the very AI investment the Fed now frames as an inflation driver ^[2]. Warsh's response, delivered in congressional testimony, was a flat refusal: they chose an independent guy to do the job and that is exactly what he planned on doing ^[3]. The governor Trump tried to remove, Lisa Cook, whom the Supreme Court protected on procedural grounds in a 5-4 ruling, is now the most specific voice in the hawkish chorus — the one who put a dollar figure on the AI spending she wants the Fed to counteract ^[8]. The counter-evidence is real, and the Fed has acknowledged it only to dismiss it. June consumer price inflation eased to 3.5%, below the 3.8% forecast, with core CPI at 2.6%. The dollar weakened on the news, and markets priced just 14% odds of a rate hike at the July 29 meeting ^[9]. The same day, producer prices fell 0.3%, the first decline since August 2025. Warsh dismissed it with two words ^[10].

There might be some that look at this morning's data and say, 'Oh, mission accomplished. Everything is swell.' That is not my view. — Kevin Warsh

The institutional machinery is in motion whether the monthly data cooperates or not. The task force reports in December. The communication architecture that once telegraphed the Fed's intentions has been dismantled. Nine of eighteen FOMC policymakers already expect at least one rate hike by year-end ^[2]. The Bank for International Settlements has now supplied the warning that makes this architecture look less like resolve and more like a trap. In its annual report, the BIS noted that five U.S. hyperscalers are projected to spend over $1 trillion on AI infrastructure across 2025 and 2026, with capital expenditure outpacing earnings — a pattern it compared to the dotcom bubble and the 1830s canal mania. The supply-side bottlenecks in semiconductors and electricity are already impacting consumers, contributing to inflationary pressures. And then the BIS said the thing that should keep a central banker awake.

Policymakers must act now. Delay will only make the necessary adjustments more costly. — Pablo Hernández de Cos

Tightening into an AI investment bubble, the BIS cautioned, could cause cascading defaults if investor optimism fades or central banks tighten policy ^[11]. The Fed has built a machine that works exactly as designed. It identified an inflationary force, constructed institutional capacity to analyze and counteract it, removed the communication guardrails that would slow its response, and coordinated a signal that rate hikes are coming. Every step was deliberate, defensible, and grounded in the central bank's mandate. The machine is pointed at the one category of spending keeping U.S. GDP above water, and the institution that knows the most about AI-driven inflation has explicitly warned that pulling the lever could trigger the very collapse the lever is meant to preempt.