Why AI Productivity Gains Could Raise Inflation Before They Lower It

Artificial intelligence is often expected to deliver a rare economic combination: faster growth without renewed inflation. But research by Ludovica Ambrosino, Jenny Chan and Silvana Tenreyro argues that the outcome is not automatic. AI can reduce inflation when new productive capacity arrives quickly; it can also increase price pressure when spending and investment move ahead of realized productivity gains. 34

The key issue is timing

Productivity affects inflation through both supply and demand. Better technology can allow firms to produce more efficiently, but expectations of higher future income and profits can also encourage businesses to invest and households to spend sooner. The inflation result depends on which effect arrives first and how quickly the economy’s capacity expands. 45

If AI-related investment and consumption rise before the promised productivity gains are delivered, demand can outpace available supply. That imbalance may create shortages, increase inflationary pressure and require monetary policy to keep demand under control through higher interest rates. 23

The researchers distinguish between a temporary productivity improvement, a permanent one-off increase and a gradual rise in productivity growth. Those different paths do not have the same inflation consequences: a productivity gain that lowers firms’ costs quickly can be disinflationary, while a persistent improvement that raises expected future income can pull consumption and investment forward. 45

AI infrastructure shows how demand can arrive first

The current AI build-out illustrates the potential problem. Companies are spending heavily on data centres and related hardware before the full economy-wide productivity benefits of AI have been realized. That demand has tightened markets for memory and graphics chips, creating cost pressure that can spread to phones, laptops and other electronics. 3

This does not prove that AI will generate permanently higher inflation. It shows why the transition itself can be inflationary: the economy may need to devote scarce chips, equipment, energy and other inputs to building AI capacity before that capacity begins expanding overall supply. 35

Why the sector receiving the productivity gain matters

The location of the productivity improvement is also important.

Domestic services can put downward pressure on prices

When productivity improves in services that are consumed domestically, firms may be able to provide more output with the same resources or at lower marginal cost. If supply expands in a sector that was previously constrained, the result is more likely to reduce domestic inflation. 35

Export gains can increase local service inflation

Productivity gains in export industries work differently. Prices for internationally traded goods are influenced by global markets, so greater efficiency may not translate into lower prices for domestic consumers. Instead, stronger export-sector productivity can raise domestic incomes and wages, encouraging households to spend more on local services such as housing, healthcare and other areas where supply is limited. That additional demand can push domestic service prices higher. 3

The same technology can therefore have different effects depending on whether it expands supply in a domestically consumed sector or primarily increases income connected to internationally traded production.

A more cautious view than the optimistic AI-growth scenario

The findings qualify the more optimistic hope expressed by Federal Reserve Chair Kevin Warsh: that widespread AI adoption could accelerate U.S. economic growth without producing higher inflation. 23

The research does not say that this outcome is impossible. Its conclusion is conditional. If productivity gains are rapid, broad and well matched to demand, AI could ultimately be disinflationary. If expectations generate spending before capacity catches up, the economy could experience an inflationary phase first. 34

What the publication represents

The work was published as Productivity and Inflation Dynamics, Bank of England Staff Working Paper No. 1,201, authored by Ambrosino, Chan and Tenreyro. It was also discussed through Bank Underground, the Bank of England’s staff blog. Reuters described the research as published by Bank of England staff; it should therefore be read as economists’ research rather than automatically as an official policy position from the IMF, Federal Reserve or Bank of England. 245

The practical lesson for policymakers is straightforward: AI’s effect on inflation cannot be judged from the technology’s eventual productivity alone. The timing of investment, the response of household spending, the sector where productivity improves and the speed at which supply expands will all shape whether AI first eases price pressure or adds to it.