Bubbles good and bad

Last week US chipmaker Nvidia became the first company in the world to reach a market value of $5tn, equivalent to £3.8tn. At this staggering valuation Nvidia’s shareholders could buy every one of the 600 or so companies in the UK equity market.

Over the last three years, Nvidia has evolved from a niche graphics-card designer to an AI behemoth, with the boom in AI driving demand for its chips. You get a sense of the scale of enthusiasm for AI from the fact that Nvidia reached a market value of $1tn in June 2023, hit the $4tn level three months ago and has now reached $5tn.

It's not just Nvidia's shares that have benefited from AI euphoria. The so-called Magnificent Seven US tech giants, which include the likes of Apple, Microsoft and Alphabet alongside Nvidia, have returned 275% since ChatGPT launched in November 2022, more than three times as much as the return on the S&P 500 index.

Such stellar outperformance has prompted talk of an AI stock market bubble and numerous comparisons with the dotcom crash of 2000. How worried should we be?

Asset bubbles are a recurring feature of market economies, often sparked by technological innovations. A 1998 US study examined a sample of 51 major innovations spanning the period 1825 to 2000, starting with the steam train and ending with the smartphone. About three-quarters of these technologies generated stock market bubbles. Some transformational inventions, such as the electric washing machine, the colour TV and the digital camera, emerged bubble free. It is perhaps unsurprising that the electric toaster, which hit the market in 1908, or the electric can opener, which arrived in 1956, were not accompanied by bubbles.

It is counter-cultural to think of stock market bubbles as benign. Bubbles, after all, distort the allocation of capital and, in the bust phase, wipe out investors and can cause recessions. The memories of the 1929 Wall Street crash or, on a lesser scale, the dotcom bust of 2000, haunt policymakers to this day.

Yet many of the greatest innovations of the last 200 years, everything from trains, to cars, radio, the PC and the internet, have generated stock market bubbles. By sucking capital into a sector and fuelling investment, market euphoria accelerates the exploitation of new technologies, boosting growth and spreading prosperity.

For all the distortions and risks they cause, stock market bubbles can be beneficial. The question is whether the gains in terms of growth and human welfare from the rapid deployment of a new technology more than offset the collateral damage to the economy during the bust phase.

How does the dotcom bubble fare on these criteria? The US Nasdaq tech index dropped almost 80% from peak to trough during the dotcom bust, pushing the US economy into recession and triggering widespread business failures and layoffs in the tech sector. But the recession was short and shallow, and the economy and productivity bounced back, aided by new tech and internet infrastructure. Telecoms giant Global Crossing, for instance, laid 100,000 miles of undersea fibre-optic cables during the boom before collapsing into bankruptcy in 2022. But those cables support the internet to this day. For the US economy, so complete was the recovery that by mid-2004 the Federal Reserve was raising interest rates in an attempt to dampen US growth.

The technologies of the dotcom boom were, like the railway, automobile and the aircraft before them, transformational. The accompanying stock market booms and busts were, at least to me, a price worth paying to get these technologies deployed quickly.

Not all booms have benign outcomes. Investors can shovel capital into technologies and businesses that fall far short of expectations. As The Economist drily observes of Japan’s tech boom in the 1980s, “much of the capex by Japanese electronics firms… ultimately served no useful function”. The past is littered with revolutionary ideas, from nuclear-powered vacuum cleaners to Betamax and 3D TV, that failed.

The case for the AI boom ending up in the bad sort of bust runs as follows. Vast levels of investment in AI have not yet delivered demonstrable gains in productivity or commensurate growth in revenues. A widely cited MIT survey of 300 companies released in August found that 95% of US AI pilot projects fail. Data security concerns, shortages of technical expertise and poorly organised data seem to be hampering the successful deployment of AI. Meanwhile, the assets created in today's tech boom are not likely to be as long-lived as the internet networks or railways of the past. The advanced semiconductors and GPUs that make up the bedrock of modern AI systems are likely to be obsolete in around three to five years. The AI boom, so the argument runs, is investing in a fast-depreciating hardware infrastructure while the technology is not yet cutting the mustard in the workplace.

If the market mood changes, things could turn nasty. Last month, Gita Gopinath, the former chief economist at the IMF, estimated that a stock market correction of the scale of the dotcom crash could wipe out over $20tn in US household wealth, equivalent to roughly 70% of US GDP, given the significantly higher exposure of retail consumers to US equities these days. This is far larger than the losses incurred in 2000 and could have a devastating effect on consumer confidence and spending.

This is alarming stuff. But at least for me the case for the bust, should and when it comes, being less apocalyptic, is more persuasive.

Tech valuations are not yet as high today as they were in the 1990s. Unlike some of the big players in the dotcom boom, today’s tech majors, like Meta and Alphabet, have huge revenues and are profitable. Corporate cash, not borrowing, is funding much of the AI capex boom. As a result, banks are less involved in the provision of capital than in the dotcom era. That reduces the risk that an AI bust could bankrupt numerous tech behemoths and, by hitting bank profits, collapse bank lending. Given their low levels of debt, strong balance sheets and other established sources of income, the tech majors would be better placed to absorb a hit to valuations than many of the dotcom players.

Even the depreciation story is more nuanced than it first appears. The powerful chips being installed in data centres depreciate quickly, but the clean energy infrastructure needed to support this data centre expansion will last.

The IMF is hardly complacent about the risks posed by AI valuations. But its latest Global Financial Stability Report, released last month, judges that the risks of a deep, systemic crisis are lower than in previous bubbles, largely because corporate cash, not debt, is a major driver of investment.

Most major new technologies generate asset bubbles. When valuations return to earth, many investors lose their shirts. What matters is whether the boom leaves behind productive, growth-enhancing knowledge and assets.