Download the PDF
Will AI produce a structural lift in productivity growth?
As 2025 draws to a close, the global economy and financial markets look to be settling on a Goldilocks trajectory. That is, a path that is not too hot that it is inducing too high a rate of inflation and interest rates, and not too cold that we risk slipping into recession necessitating a sharp cut in interest rates. As covered in a recent Brief (2025: a year that will go down in history) the major global economies of the US, Europe and China are all tracking toward trend growth rates this year and are expected to grow at trend next year. Inflation is slightly elevated but below 3% with expectations of convergence toward central bank targets over 2026, and interest rates are close to levels we would expect when the economies are close to equilibrium. One global market that does not seem in line with historical trends, however, is the equity market. Here, equity valuations seem stretched. In the case of the US, the price/earnings multiple of the S&P 500 is touching levels last seen just prior to the bursting of the dot.com bubble 25 years ago. Over the first two weeks of November, we have seen a lift in equity market volatility, albeit modest in extent, as markets take out Fed easing in the face of stubborn inflation numbers. However, sentiment around the equity market remains robust based on an assumption that tech stocks, and AI in particular, will deliver the economic gains that justify current valuations.
Analysts continue to forecast exceptional earnings growth for the S&P 500 of 13.5% over the coming 12 months, double the historical average and double the rate one would expect of an economy growing at trend. And, according to equity analysts’ projections, this is to be followed by a 14% earnings growth rate over 2027, again when the US economy is forecast to grow at trend. The rally in the US stock market amidst a normalisation in debt markets following Covid has seen a severe compression in the US equity risk premium. For example, the premium for the S&P 500 currently sits at about 2.5%, based on IBES forward earnings and the real bond yield as given by US 10-year Treasury Inflation-Protected Securities yield. At this level, the current US equity risk premium is little over a half of the premium that has existed on average over the last 40 years. Can valuations and risk premia be sustained at these levels? The short answer is, yes. There are several factors that could justify current valuations in the longer term, but the most obvious is the potential for AI to deliver unprecedented productivity gains that lift the earnings potential of not just tech companies, but all companies within the economy. While history is not always a faithful guide to the future, a review of the record is useful, if only to isolate the key assumptions that are driving current thinking. So, what does the historical record tell us about US productivity growth? Over the last half century, growth in US labour productivity has averaged 1.5%. Of course, there has been volatility around this average, and ignoring the distortions of the Covid period from 2020-22, productivity has temporarily spiked as high as 3.6% in 1992 and as low as zero growth a decade earlier, in 1982. The period from 1975 to the present is a rich sample of negative productivity shocks (oil price shock of the 1970s and 1980s, the GFC and European debt crises) and positive productivity shocks (the IT Revolution of the 1990s with the introduction of personal computers and the development of the internet). We will focus on the period from 1994 to 2003, the period of the IT Revolution as the period holding the greatest resemblance to current circumstances. During this period, US productivity growth averaged 2.2%, around 50% higher than its long-term average.
It is generally accepted that during this period, productivity growth was lifted, first, by the IT sector as it developed the use of personal computers and the internet, and then, by other industries as they adopted the latest technology. An oft-cited example of the later effect is the use by companies in the retail industry of applications such as Excel to better manage inventories. However, the productivity growth boom did not last and was followed by a nine year period, from 2011 to 2019, where productivity averaged 0.8%. What do we make of the historical record? First, it would be historically unprecedented (at least given the historical record of the last 50 years) for a technological breakthrough to permanently shift the average annual rate of growth in productivity. Second, previous technological breakthroughs have resulted in non-permanent, but prolonged increases in productivity growth with the most recent example of the 1990s IT Revolution leading to a productivity spike lasting for almost a decade. What does this mean for the current AI Revolution? The hurdle for AI to generate a permanent boost to the growth rate in productivity is elevated, but its potential to have a years-long impact is possible with the recent historical experience of the IT Revolution a case in point.
As the dot.com period showed, investor exuberance can drive equity market valuations ahead of fundamentals. This eventually leads to a price correction once the true contribution that the technology will deliver is revealed. It is easy to understand how investors can get ahead of themselves, if, like the IT Revolution of the 1990s, the new technology delivers a productivity boost for many years. What investors must assess is whether that boost can be extrapolated to a permanent structural shift higher in productivity growth. If the answer to that is no, then investors must also face the decision as to when it is best to lower their exposure to the potential for disappointment and correction. During the IT Revolution, the market took five years to rally to its peak valuation with the S&P 500 forward earnings PE multiple topping out at 24.5, its highest level in data going back to 1985. By comparison, we are only halfway through the AI Revolution that has been running for about 2½ years. But over those 2½ years we have already hit a PE multiple of 23.3, within striking distance of the IT Revolution’s all-time high.
From its peak in March 2000, the fall from grace in the S&P 500 was brutal, with the price falling by around 50%, from 1527 to 776 by October 2002. Will the AI Revolution be different from the IT Revolution? Will the AI Revolution have more in common with the Industrial Revolution? If current valuations are to be sustainable, the answers to these two questions must be in the affirmative.