TL;DR

Silvana Tenreyro, now chief economist at the International Monetary Fund and formerly on the Bank of England’s rate-setting committee, has published research arguing that AI’s inflation effect is ambiguous rather than downward. Spending that anticipates productivity gains before those gains materialise can tighten supply, push prices up and require higher rates. The work appeared on the Bank’s staff blog with two co-authors.

The intuition that does not survive contact

The comfortable assumption runs: AI raises productivity, more output from the same inputs means lower unit costs, lower costs feed through to cheaper goods. Kevin Warsh, who chairs the Federal Reserve, has said in public he hopes the US can grow faster on that basis without inflation following.

Tenreyro breaks that chain at the timing, in work co-written with Jenny Chan, an economist at the Bank, and the doctoral researcher Ludovica Ambrosino. Firms invest and households spend on the expectation of gains, not on gains already banked — which the authors note is arguably happening right now with AI infrastructure. Demand arrives first. If the productivity has not yet shown up, that demand meets a supply constraint, and the result is inflationary.

There is already visible evidence. Memory and graphics chip prices have climbed over the past year on data centre demand, feeding through into phones, laptops and consumer electronics. That is the mechanism operating in public.

Where the gains land also matters

A second finding cuts against simple readings. The inflation effect depends on whether productivity improves in exports or in domestically produced services. Gains in services tend to reduce domestic inflation. Gains in exports tend to lift domestic wages and increase demand for services whose supply cannot easily expand — which pushes inflation up.

Looking forward

The Bank Underground blog carries staff views rather than official Bank positions, and Tenreyro contributed in her London School of Economics capacity, having left the committee in 2023. So this is not a policy signal.

It is still awkward for a particular strand of UK optimism. Much of the case for AI as a growth fix assumes it arrives disinflationary, letting rates fall while output rises. This argues the sequencing may run the other way first — the investment boom before the productivity payoff — and that the interim is when rates have to stay higher, not lower. The European Central Bank’s warning this week about a probable AI market correction points at the same interval from a different angle.