Is AI Inflationary or Deflationary?

Key-points

The AI boom goes from strength to strength. Big technology companies are pouring hundreds of billions of dollars into chips, data centers and power-hungry infrastructure. One estimate puts annual AI infrastructure investment above $650 billion in 2025 and potentially over $800 billion in 2026. Meanwhile, SpaceX's record initial public offering (IPO) — raising about $75 billion and valuing the company near $1.75 trillion at issue — is a useful reminder that capital markets remain wide open for anything with a plausible claim on the AI future.

That creates an awkward problem for the Federal Reserve (Fed). AI may be a long-run productivity miracle. But the short-run impulse looks less like disinflation and more like a classic investment boom colliding with limited supply. Semiconductors are scarce and used in almost all industrial goods. Electric grids are constrained. Construction costs are rising. Land, labor, copper, turbines and transformers are not infinitely elastic. Data centers may be digital assets in the long run, but they require very physical inputs.

Some Fed officials have already expressed concern that these pressures could complicate the inflation outlook. The concern is not simply that Nvidia chips cost more. It is that the AI build-out touches several sticky categories at once: equipment, electricity, commercial construction, skilled labor and financial conditions. Add buoyant equity markets and robust IPO demand, and the Fed must also consider whether asset-price inflation is loosening conditions even as policy rates remain restrictive.

This is where the new Fed chair, Kevin Warsh, matters. Having chaired his first Federal Open Market Committee (FOMC) meeting last week, Warsh has already signaled that the central bank's inflation framework needs updating, including greater attention to trimmed or underlying inflation measures rather than simply relying on core personal consumption expenditures3 (PCE). At his Senate hearing, Warsh said he wants to understand "the underlying inflation rate" and whether shocks are creating second-round effects. After several years of overshooting the 2 percent target, the Federal Reserve's institutional credibility is no longer free.

The numbers could force the issue. If AI-related demand adds as much as 0.5 percentage points to core inflation over the coming year, policymakers cannot easily look through it. A credible central bank may need to lean against the boom by keeping rates higher, slowing credit growth, and tightening financial conditions enough to cool the most speculative parts of the cycle.