Federal Reserve Chairman Warsh watches AI token prices for clues on productivity and inflation

The Federal Reserve now has a new data point on its radar. Fed Chairman Kevin Warsh used his Jackson Hole keynote on August 28 to spotlight an unusual metric: the price dynamics of AI tokens, the units that measure how much data an AI model processes per query. AI companies typically charge customers based on token consumption, making token pricing a surprisingly clean proxy for how aggressively businesses are adopting the technology.

Warsh noted that AI labs’ token sales have surpassed $100 billion, representing growth of more than 500% from the prior year.

What Warsh actually said, and why it matters

The chairman described artificial intelligence as a potential “new factor of production,” placing it in the same conceptual bucket as labor, capital, and land. That framing implies AI could fundamentally alter how the Fed thinks about the economy’s productive capacity, and by extension, how it sets interest rates.

But Warsh didn’t just cheerlead. He posed a pointed question: will customers continue paying premium prices for tokens generated by the most advanced frontier models, even as older models see their prices collapse toward marginal cost? The answer to that question could reveal whether AI is creating genuinely new economic value or just running on hype-cycle fumes.

A tiered pricing structure, where frontier models command premiums while older models become nearly free, would suggest genuine differentiation in AI capabilities. If instead all token prices converge downward, it could signal that AI is becoming a commodity utility rather than a transformative production input.

The $100 billion signal

For the Fed, revenue is only half the equation. The central bank cares about productivity: whether AI is enabling businesses to produce more output per hour of labor. Revenue growth tells you that companies are buying AI. It doesn’t tell you whether AI is making them meaningfully more efficient.

Warsh was explicit on this point. Despite the promising signals, he said the productivity effects of AI remain uncertain and are not currently influencing monetary policy decisions. The Fed’s approach remains data-dependent, focused on achieving sustainable inflation targets.

Forward guidance and the AI variable

One of the more notable elements of Warsh’s speech was his emphasis on reducing forward guidance. If AI does deliver a genuine productivity surge, it would raise the economy’s potential growth rate without necessarily stoking inflation. But if AI spending turns out to be a capital expenditure bubble without corresponding productivity gains, the Fed could face overinvestment followed by correction.

The token pricing dynamic Warsh flagged could serve as an early warning system for either outcome. Rising or stable prices for frontier tokens would suggest sustained demand for genuinely superior AI capabilities. Rapid price deflation across all model tiers would suggest commoditization and potentially diminishing returns on AI investment.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.