Artificial inflation

So much ink has been spilt debating potential stock market winners and losers from recent AI advances that there’s a danger of not enough ink being left to debate another issue that’s vital but sometimes overlooked. This is the question of what AI means for inflation and interest rates. It’s a question that will directly affect the performance of bonds in clients’ portfolios, with knock-on effects on stocks too.

In the US, this item has accelerated up the agenda following May’s confirmation of Kevin Warsh as the US Federal Reserve’s Chair. Before selection as President Donald Trump’s nominee, Warsh argued that AI would reduce inflation, opening the door to lower interest rates. This may have been to boost his chances of landing the job, given Trump’s repeated calls for big interest rate cuts. Since taking over at the Fed, Warsh has made the same argument, albeit less dogmatically.

The argument rests partly on history. One of Warsh’s predecessors, Alan Greenspan, resisted calls to hike interest rates in 1996 in response to strong growth. Greenspan thought widespread adoption of IT was driving up growth in productivity: how much each worker can produce in a given period. This would increase the rate at which the economy could grow without generating inflation. If each worker can produce more, they can also earn more and consume more without the extra demand just bidding up prices.

There was no evidence of this in the official aggregate data at the time. But Greenspan had a hunch it was the case based on developments elsewhere, and knew that measuring productivity (especially in real time) is tricky. His suspicion ultimately proved correct: inflation held steady and revised data eventually showed productivity growth had, in fact, been accelerating.

It’s easy to see the appeal of returning to the economic conditions of the late 1990s. The US economy boomed, with growth above 4% per year. The unemployment rate was falling to lows not seen since the 1960s, while inflation stayed under control. This allowed the Fed to lower interest rates from 6% in early 1995 to 4.75% by 1998.

The Greenspan gamble

Strong growth and falling unemployment didn't stop the Fed from cutting rates as inflation remained under control

Inflation in the short term

However, we’re sceptical that Warsh will be as fortunate as Greenspan. In the near term, the evidence suggests that AI will be inflationary, largely due to the demand being created by the enormous scale of AI investments. After all, part of the reason that many stocks perceived to benefit from AI investment have performed so well lately is because demand-supply imbalances have driven up prices for their goods and services.

This pressure is feeding through into measures of inflation that shape the Fed’s interest rate decisions. Higher prices for memory chips and other electronic components are already leading to price hikes for some consumer goods, such as MacBooks and iPads. Meanwhile, power-hungry data centres are putting upward pressure on electricity prices. Building those data centres is causing bottlenecks in other areas, too. Wage growth for construction workers has rebounded, unlike in other sectors. This rebound is increasing construction costs throughout the economy.

Inflation in the long term

Looking further ahead, what matters is whether AI can deliver stronger productivity growth. It’s not clear yet whether such gains are occurring for economies as a whole, given the real-time measurement issues. However, in the long-term assumptions underpinning our portfolios, we’re now including some conservative forecasts, grounded in the latest academic literature, for how much AI could boost productivity across various economies.

If we’re correct, there would be some potential for AI to bear down on inflation – temporarily. Producing more with less creates savings that can be passed on (at least in part) to households. There are two channels through which this happens: lower prices and higher wages. But firms tend to be able to drop prices faster than workers can bargain for the savings to be directed towards their pay packets. That means we tend to see some downward pressure on inflation.

Will that allow Warsh to cut rates? We don’t think it will be that simple. One reason: economic theory tells us that higher productivity growth should lead to a higher real neutral interest rate. The real interest rate is the headline or ‘nominal’ rate, minus expected inflation. The real neutral rate is what’s consistent with an economy operating at full capacity, with stable inflation. It’s ‘neutral’ in the sense that it doesn’t either stoke or reduce demand.

Why is faster productivity growth apt to do this? First, it increases companies’ expected returns on capital. That boosts their demand for capital. Second, households could save less if they expect stronger future wage growth, enabled by productivity improvements. That reduces the supply of capital. More demand and less supply means a higher price – the price in question here is the rate of interest.

This raises the question of how the Fed avoided hiking rates for most of the late 1990s. First, it’s important to recognise that here we’re talking about nominal interest rates. Because inflation expectations were falling, the real policy stance wasn’t becoming quite as accommodative. Second, there were several other factors pushing in a disinflationary direction. Rapid globalisation was putting downward pressure on inflation as cheaper goods from Asia and elsewhere became available.

At the same time, the US federal government was reducing inflationary pressure by tightening fiscal policy and taking demand out of the economy. The benign geopolitical backdrop and limited effects of climate change also helped prevent major supply shocks from stoking inflation. That world is very different from today’s.

AI comes at a price

Investment in AI infrastructure is contributing to higher prices across a range of manufactured goods, shown here as factory gate inflation.

A lesson from the dotcom boom

As a side note, we have some sympathy for the argument that it was a mistake to keep rates lower in any case, because this allowed the stock market bubble to keep inflating. The Fed eventually responded by raising rates sharply from 1999.

Ultimately, there is still considerable uncertainty about how AI will develop and its likely impact on economies. But unlike Warsh, we think the balance of risks is tilted towards interest rates remaining higher over the long run. This suggests no return to the halcyon days of the 2010s. That, in turn, would mean higher (though not necessarily rising) yields for fixed income. It may also prove a headwind to equity markets. Though there are potential tailwinds, such as stronger earnings growth, that could prove more powerful.