Artificial intelligence could raise productivity without producing the lower inflation that might intuitively follow, according to new research examining how investment and consumption respond before expected improvements in productive capacity have arrived.

The work by Silvana Tenreyro, the International Monetary Fund’s new chief economist, Bank of England economist Jenny Chan, and researcher Ludovica Ambrosino challenges a simple assumption around AI economics: if workers and machines can produce more with the same inputs, prices should face downward pressure.

That mechanism still exists because higher productivity can reduce marginal production costs and expand the amount an economy can supply. The researchers find, however, that the inflation effect depends on whether demand rises simultaneously, how quickly productivity improves, which sectors receive the gain, and how monetary policy responds.

AI provides a useful test because expectations are already influencing expenditure. Technology companies are building data centres, organisations are purchasing software and infrastructure, and investors are valuing businesses partly on productivity improvements expected several years ahead even though aggregate evidence of those gains remains comparatively limited.

Demand can arrive before productivity

The researchers distinguish between a one-off improvement in the level of productivity and a sustained increase in the rate of productivity growth. A temporary improvement can lower costs and place downward pressure on prices, whereas persistent growth changes behaviour because companies and households begin expecting higher future returns and incomes.

Businesses may invest more in anticipation of larger markets and cheaper production, while households can increase consumption because they expect to be wealthier. If that demand arrives before productive capacity expands, prices can rise rather than fall.

Anticipation is particularly important where investment in new equipment, facilities, skills, or infrastructure is required before the productivity benefit becomes available. AI infrastructure already displays some of those characteristics.

Demand for accelerators, memory, power equipment, data-centre construction, and specialist engineering has expanded as companies prepare for future workloads. Capacity in those supply chains cannot necessarily respond at the same speed, creating price pressure even if the technology eventually allows more output to be produced with fewer inputs elsewhere.

That sequence complicates the argument that central banks can simply look through an investment boom because productivity will eventually expand the economy’s capacity. Monetary policy has to respond to the demand and inflation occurring now, whereas uncertain future improvements cannot be treated as productive capacity before they materialise.

Higher productivity can raise rate pressure

The model also connects sustained productivity growth with the natural real interest rate — the theoretical rate consistent with stable inflation when an economy is operating around its potential. If households expect higher permanent income and businesses identify more profitable investments, desired consumption and investment can rise together.

Absent tighter monetary policy, that additional demand can generate inflation even while the economy becomes more productive. Central banks may therefore need higher interest rates during part of the transition rather than being able to cut because AI has expanded supply.

That possibility affects companies financing AI investment. Business cases may assume productivity improvements will lower operating costs, while the infrastructure needed to deliver those improvements is simultaneously exposed to borrowing costs because data centres, networks, energy systems, and hardware require large amounts of upfront capital.

An AI boom that produces higher productivity while keeping real interest rates elevated would distribute gains unevenly. Software businesses requiring relatively little physical investment could experience the transition differently from the companies constructing computing and energy infrastructure.

The researchers also find that the sector in which productivity improves influences the inflation outcome. Gains concentrated in domestically consumed services can affect domestic prices differently from improvements in internationally traded sectors that raise incomes and spending elsewhere in the economy.

That distinction is relevant to Britain and other European economies where services make up a large share of output. Productivity changes in professional services, administration, finance, healthcare, and other domestic activities can feed through to inflation differently from improvements concentrated in export manufacturing or globally traded technology.

AI’s economic effect will not arrive as one number

The research does not conclude that AI is inherently inflationary. Instead, it shows why a productivity forecast cannot be converted mechanically into a price forecast without considering how households, businesses, supply chains, and central banks respond to the same technological change.

The economic debate is consequently moving beyond whether AI raises productivity towards the timing and distribution of any gains. The expenditure required to adopt AI may initially increase demand for scarce labour, computing hardware, electricity, construction capacity, and specialist services before organisations become more efficient.

Businesses encounter the same mismatch on a smaller scale. Software can be purchased before processes have been redesigned, employees trained before workflows are automated, and data-centre capacity contracted before new AI services generate corresponding revenue.

Some of that expenditure will eventually produce additional output, while some projects will fail to justify their cost. Macroeconomic statistics reveal the balance only after those decisions have worked through the economy, leaving central banks to set policy while much of the expected supply improvement remains uncertain.

The appeal of the disinflationary AI narrative is that it appears to offer stronger growth without the accompanying price pressure. The new research presents a less tidy transition: productivity can increase economic capacity, but expectations of that increase may encourage companies and households to spend first, putting pressure on prices before the additional output arrives.