Uncommon truths: Is AI delivering?

AI has certainly delivered for investors in the enablers but what about users? Unemployment is trending up in some countries, US productivity growth is rising and profits have been strong. However, there are many caveats, and it remains difficult to sort the AI wheat from the cyclical chaff. 

At the end of a second successive week of falling oil prices (taking the Brent first future to $91), the optimism expressed by equity markets since the end of March appears warranted. Given that we started the year in optimistic mood and made only limited changes to our Model Asset Allocation in March, we hope the mooted US/Iran agreement comes to fruition (in mid-March we boosted the US equity allocation and reduced energy commodities to zero – see Figure 6).

Of course, relying on logical outcomes in this volatile geopolitical environment is never comfortable, which is why we prefer to focus on longer term drivers of investment performance. The AI theme has been among the most powerful of such engines, with the benefits spreading to enablers outside the US (South Korean stocks have nearly tripled and Taiwanese stocks have more than doubled in the last 12 months, based on MSCI indices). 

The question being, when will the benefits of AI spread to users? This is important not only for users and economies but also for enablers, since it is hard to imagine their massive investments paying off if users are extracting no benefit. 

Apart from anecdotal evidence that we are all spending a lot of time trying to extract gains from AI, how will we know it is delivering on its promise? I wrote about this in December 2024 (see How will we know AI is delivering?). I concluded that we need to keep an eye on trends in productivity, employment, inflation and profit margins but that sorting the underlying wheat from the cyclical chaff will be difficult. 

Interestingly, unemployment has been gently trending up since early 2022 in Germany, the UK and the US (Figure 1a). This is despite divergent growth trends, with the German economy treading water, while US GDP has grown at an annualised rate of around 2.5% since 2022 Q1 (the UK in the middle with growth of around 1.0%). At the same time, unemployment has been relatively flat in Japan (GDP growth of around 0.1%) and falling in the Eurozone (1.3%). That unemployment in Germany and the UK has risen is not surprising given the low/no growth backdrop. However, it is more surprising that US unemployment has risen when there has been reasonable growth. Could this be a sign that AI is impacting the US economy (though the benefits of AI can accrue anywhere in the world)?

This brings us to productivity. Figure 1b shows 10-year annualised growth in the broadest measure of productivity (GDP per capita) across a range of economies, using GDP in 2015 US dollars, converted using purchasing power parity (PPP) exchange rates. China is excluded because its productivity growth has been so much higher than in the other countries in the chart (5%-6% in recent years, down from 9%-10% in the 2007-14 period). Based on that chart, US productivity growth has been higher than elsewhere since the pandemic but there is no evidence of an uptrend. Indeed, outside the US the reverse seems to be true. 

So, there seems no evidence that productivity has accelerated in recent years. However, there are two problems with the data in that chart. The first is that it shows 10-year annualised gains, and recent changes may take a while to show up. Unfortunately, the conclusion doesn’t change even if we use year-on-year growth rates, which by their nature are more volatile than the 10-year numbers and more prone to cyclical distortions. 

The second problem is in what is being measured. GDP is a full economy measure, including the public sector and agriculture, which may mask AI benefits accruing to the non-farm business sector. Further, the denominator in GDP per capita is the total population and not the workforce, which again may distort what is happening to worker productivity. GDP/capita may be useful when assessing the wellbeing of a country, but other measures may better capture productivity. 

Figure 2 shows two alternative measures for the US economy: non-farm business output per worker and non-farm business output per hour worked. These measures are not only more precise than GDP/capita, but they also suggest that productivity growth is rising. 

However, there are several caveats: first productivity growth was starting from a low base, and the recent increase may imply a normalisation rather than a new upward trend. Indeed, the year-on-year growth rates have been in the 2.0%-3.5% range since mid-2023 but have not trended up in that period. Nevertheless, if they were to stay at that level over the next 10 years, the annualised 10-year growth rates would start to resemble those seen in the late 1990s and early 2000s, which may themselves have been the result of the technology revolution of that time, which included the roll out of the internet. 

Second, the recent rise in productivity growth has been accompanied by a pick-up in inflation. Some of that was because of the energy price rises that came with the Russian invasion of Ukraine but I think it had more to do with the fiscal and monetary policy settings of 2020 and 2021, which helped economic recovery during the pandemic. Hence, the rise in productivity growth in recent years may have a cyclical element, in contrast to the late 1990s/early 2000s. 

Turning elsewhere for signs that AI is delivering to users, what are business profits telling us? Though Q1 earnings reports were strong in both the US and Europe, much of the strength in the US came from the technology and communications sectors, which may owe more to AI enablers than AI users. Also, the strongest gains in Europe were in the energy sector, which I think owes more to high energy prices than to the benefits of AI. Hence, it again seems premature to suggest that users are extracting durable productivity and margin gains. 

Overall, there are signs of hope that AI is delivering for users, with rising unemployment, some gains in productivity and strong earnings reports. However, there are many caveats, and I think it is too early to conclude that users are extracting the benefits that would justify recent gains in the share prices of enablers and to hold out the promise of AI gains spreading across the stock market. 

All data as of 29 May 2026, unless stated otherwise. 

Investment risks 

The value of investments and any income will fluctuate (this may partly be the result of exchange rate fluctuations) and investors may not get back the full amount invested.