If artificial intelligence truly ushers in a productivity revolution, why are we not already seeing a rapid fall in inflation? Current evidence and historical parallels suggest AI’s macroeconomic impact is likely two-phased: an initial investment-driven phase that can be mildly inflationary, followed—after several years—by a productivity-led disinflationary period.
Why the early phase can be inflationary
Major technological breaks typically trigger large investment waves that, in the short run, create supply bottlenecks and upward price pressures. AI development requires substantial physical infrastructure: data centers, expanded semiconductor capacity, electricity grids, energy supply, cooling systems and increased demand for certain raw materials such as copper. If these build-outs occur simultaneously, capacity constraints can emerge and push prices up.
Macro data illustrate the strength of this capex wave: in Q1 2026 technological investment in the United States amounted to about 4.9% of GDP, exceeding the dotcom-era peak of roughly 4.5% of GDP. So far, this wave has mainly manifested as demand-side pressure linked to investment, while a measurable economy-wide productivity surge has not yet appeared.
Equity market dynamics can also indirectly support inflation: the AI optimism-driven bull market has raised household equity holdings to about 230% of GDP (compared with roughly 130% at the dotcom peak). Larger wealth effects can sustain stronger consumption, which in the short run helps keep upward pressure on prices.
Potential for later disinflation
Over the medium term—typically with multi-year lags—AI could raise productivity: automation of white-collar tasks and scalable services can lower unit costs and exert downward pressure on inflation. This is particularly important in economies where services make up a large share of inflation; in the United States services represent roughly 70% of inflation.
The timing and magnitude of such disinflation are, however, highly uncertain. They depend on how fast AI spreads, whether firms pass through cost savings into prices, the degree of competition in sectors, and how broad-based productivity gains become across the economy.
Monetary policy and political risks
Kevin Warsh, recently appointed to lead the Federal Reserve, appears more open to rate cuts than his predecessor Jerome Powell, in part because he expects large productivity gains from AI to lower inflation over time. Historical experience counsels caution: Warsh has compared himself to Alan Greenspan, who in 1996–1998 resisted raising rates by pointing to productivity gains from the internet boom—a stance many economists later linked to overheating and the dotcom bubble expansion.
The article warns that if central banks prematurely rely on AI’s expected but unrealized disinflation and ease policy too soon, that could worsen inflation outcomes.
External shocks and labour-market signals
An additional factor blocking quick disinflation is an energy-price shock related to the Iran war, which currently has a larger influence on inflation than AI’s prospective structural effects. Total inflation reached 4.2% in May 2026, while core inflation rose to 2.9%, both above the Fed’s 2% target. Market-based inflation expectations have also increased, suggesting participants are pricing in more persistent inflation.
Labour-market data do not yet clearly signal the large labour-cost savings some attribute to AI: employment figures over the past three months surprised on the upside, and the anticipated sharp negative employment effects in some sectors have not materialised broadly. At the same time, ramped-up data-center construction is showing up as higher employment in construction and manufacturing categories. Net, AI’s effect on employment may have moved from slightly negative toward slightly positive.
Historical parallels: railways, electrification, IT
Past major technological revolutions—the railway boom in the mid-19th century, electrification around the turn of the 20th century, and the IT revolution from the 1990s—generally started with an investment-driven, inflation-raising phase and only years later delivered widespread productivity gains and sustained disinflation. For example, at one point a third of U.S. investment focused on railways; electrification and IT followed similar temporal patterns of a lag between capex and economy-wide productivity benefits.
Conclusions
Current evidence suggests AI will not solve inflation in the short term; the ongoing investment wave is more consistent with an early inflationary phase seen in previous technology revolutions. Over time, AI is likely capable of reducing inflation through productivity gains, but when and how much that will occur remains uncertain.
Monetary policymakers are therefore advised to judge developments cautiously and rely on incoming data rather than prematurely assuming future disinflation. The previous parts of this series were published on 2026-06-10 and 2026-06-11, and reflect the authors’ opinions.
Cover image is illustrative.
