What a Soros theory can tell us about the AI boom

Unlock the Editor’s Digest for free

Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.

The writer is a financial journalist and author of ‘The Economic Consequences of Mr Trump’

It is a mug’s game trying to predict the end of a boom with any precision. They last much longer than anyone might reasonably expect. That is true of bull markets, as well as economic advances. The reason is that markets and economies find ways to support themselves. George Soros, the well-known investor and philanthropist, has a term for it: reflexivity.

In a Financial Times article back in October 2009, Soros defined the concept, in terms of its impact on markets, quite succinctly. “The participants’ views influence the course of events, and the course of events influences the participants’ views,” he wrote.

It is a positive feedback loop. The same idea was at the heart of what John Maynard Keynes, the great economist, described as “animal spirits”; if businesses are confident, they will invest money and hire more workers, and this investment will boost economic growth.

In terms of asset markets, the most obvious example of reflexivity comes from the link between banking and property prices. Initially, for whatever reason, banks start lending more money to people who are buying property. The availability of additional finance pushes up demand for property — whether it is office blocks or homes — and property prices rise. This makes the bankers more confident about lending money in the property sector, as their collateral is rising in value. And it makes investors and or speculators more willing to borrow money to buy property, since it looks like a very good bet.

Debt does not have to be involved. For much of the life of cryptocurrencies, the price of digital assets such as bitcoin and ethereum has been sustained by the belief, among some investors, that they represent the wave of the future. Any weakness is thus a buying opportunity. And a rising price is a wonderful way of proselytising the crypto religion; more people are tempted to adopt the faith.

Another way in which booms can sustain themselves, in both economic and asset-market terms, is through spending on goods and services. That is clearly the case at the moment with the rush to invest in artificial intelligence.

This spending has done a lot to prop up US economic growth, at a time when job creation has stalled and consumer confidence has declined. In the first half of the year, JPMorgan estimated that AI spending contributed 1.1 percentage points to US GDP. In market terms, it plays a crucial role in convincing investors of the solidity of the AI boom, not least in the demand it creates for the chips made by Nvidia, the world’s most valuable company.

The buzz surrounding this spending also creates a kind of Fomo (fear of missing out) among other executives. If AI is the wave of the future, then any company that doesn’t embrace it risks being left behind. And, true to the principle of reflexivity, the race to invest makes the AI boom seem all the more substantial to investors. The obvious parallel is the late 1990s when spending on fibreoptic cable, routers and telecoms equipment soared, spurring the dotcom bubble.

The intoxicating nature of bullish sentiment indicates how these booms may eventually sow the seeds of their own destruction. In the late 1990s, it seemed that every twenty-something was either launching their own website or joining a start-up internet company with the hope of cashing in their share options. The appeal of the technology was so obvious that too many businesses were founded; only a fraction of them would ever be profitable. When it became clear, in the spring of 2000, that some businesses were running out of cash, sentiment changed. 

The AI boom is different as it is focused on a few big players with strong existing business models, rather than on a host of start-ups. This means that the financial pressures are unlikely to bite as quickly.

On the other hand, AI might not be as immediately useful as many executives hope; a McKinsey study found that 80 per cent of companies that had started to use AI had yet to experience any boost to their profits. Plenty of consumers — particularly students — are enthusiastic users of AI to summarise reports and generate business proposals or essay plans. Useful stuff, but hardly the basis of a productivity miracle.

Of course, in the past, the impact of innovations such as electrification has taken decades to show up in the productivity numbers. By that stage, however, history suggests that a market boom, even if powered by reflexivity, will be long over. At some point, the growth rate in AI spending — and in Nvidia’s revenues — will slow; and then the rating that investors are willing to apply to corporate earnings will decline, along with share prices. The bandwagon will develop a wonky wheel. 

Arguing that a boom must come to an end is not the same as saying the underlying technology is rubbish. AI will be useful, just as the internet is useful and the railways were very useful. That didn’t stop the other two booms from experiencing crashes. A reflex action may prolong a boom but it can also deliver a painful kick.