The AI valuation bubble is now getting silly | Nils Pratley

It is easy, at the distance of a quarter of a century, to forget the full madness of the late-1990s dotcom bubble. The tech-heavy Nasdaq index in the US rose by 86% in 1999 alone. Companies only had to announce a hopeful “internet strategy” to see their share prices soar. And the key point is that it went on for ages. Alan Greenspan, the then chair of the Federal Reserve, used the phrase “irrational exuberance” to describe the mood in stock markets as early as December 1996, more than three years before the bubble finally burst. Infamously, Greenspan himself went on to supercharge the latter stages of the mania by cutting interest rates three times, after a currency crisis in Asia in 1998 and the blow-up of a huge hedge fund, Long-Term Capital Management. Everybody at the time half-knew it had all gone too far too soon but the greater danger, career-wise, was often to sit it out. Poor Tony Dye, one of London’s top fund managers, earned himself the nickname “Dr Doom” for predicting an imminent stock market disaster for about five years; he lost his job at Phillips & Drew just before his prophecies came true. Related: Do OpenAI’s multibillion-dollar deals mean exuberance has got out of hand? The turn-of-the-century experience explains the current crop of analysts’ notes asking where today’s AI bubble – because that’s what it clearly is – fits in a 1990s timeline. The broad parallels are genuinely close. The development of AI will clearly have society-changing effects, just as the arrival of the internet and mobile communication did. The impossibility lies in knowing the speed of adoption, and which lumps of capital will earn extraordinary returns and which will end up getting torched. Are we in the merely exuberant foothills of 1996? Does a last manic 1999-style “melt-up” come next? Or is 2000’s bell about to ring? Many of those notes end up being well-argued versions of “dunno, guv” because timing is impossible. But here are four points that suggest it’s getting late in the day. First, valuations are extremely stretched by usual standards. Everybody – including the Bank of England’s financial policy committee in its report on Wednesday – is staring at market historians’ favourite yardstick, the Cape ratio, with amazement or alarm. The Cape measures cyclically adjusted price-to-earnings ratios over the past decade taking account of inflation. On that score, we’re back at the peak dotcom bubble. The better news is that, on a plainer, forward-looking, price-to-earnings ratio, the S&P 500 index is at 25 times, which is not as severe as 1999 levels. But the overall picture is clear: prices are very high. Second, concentration risk in markets is off the charts. The “Magnificent 7” tech companies – Alphabet, Amazon, Meta, Tesla, Apple, Microsoft and Nvidia – now represent slightly more than a third of the whole S&P 500 index. To varying degrees, all are bets on the future of AI. If, or when, the AI bubble bursts, there will be few places to escape the pop. One doubts the concentration is appreciated by US holders of tracker funds who look to invest in a diversified portfolio of companies. And, given the dominance of the US in global indices, the same applies to a lesser degree for your everyday global tracker, too. Third, the correlation risk is becoming worse given the AI firms’ obsession with cross-shareholdings and partnerships. The deal whereby OpenAI will pay Nvidia for chips, and Nvidia will invest $100bn in OpenAI, has been criticised as circular because that’s exactly what it is. In the latest move, OpenAI has pledged to buy lots of AMD chips and take a stake in AMD over time. Fans of these arrangements see an alignment of interests for mutual gain. One can equally view the fancy financial footwork as a formula for ensuring capital is misallocated. Certainly, the strain of getting the AI infrastructure built – before returns are remotely clear – seems to be showing. Fourth, the wider economic backdrop for an IT revolution is not as helpful as in the 1990s. “Inflation wasn’t a problem in the 90s,” argued Dario Perkins of TS Lombard recently. “Outside the IT sector, there were powerful structural forces keeping prices down, such as rapid globalisation and the collapse of the Soviet Union.” As he says, that’s not today’s world of deglobalisation, supply shocks, anxiety about public finances and lurches towards populism. Conditions are less favourable for a sustained “melt-up”, he concludes. That sounds right. On the other side of the ledger, one can point to Nvidia’s sheer revenues – £46.7bn in the last quarter – as evidence that the foundations are more solid this time. Certainly, Nvidia puts Cisco, the leader in the dotcom years, in the shade. Yet, viewed the other way around, it’s actually more concerning that so much sentiment about AI revolves around a single stock. It’s not as if OpenAI is threatening to make profits soon. To repeat, timing is impossible and no two bubbles are exactly alike. But does it all feel as silly as last time? Short answer: yes.