When to leave the AI party

How do you know the perfect time to leave a party? There’s an art to staying long enough that you don’t miss out on a good time, while still departing before suffering the consequences of earlier overindulgence. Investors in AI-related stocks may be asking themselves a similar question. How can we know when enthusiasm about AI is about to go too far, and it’s time to call a taxi home?

We don’t believe we’re at that stage in the AI cycle now, but it always pays to be prepared. Trusting gut feel, as you would at a real party, isn’t enough. That’s why – and here’s a key difference with partygoing – we’ve created a comprehensive set of conditions to monitor for signs of a major, imminent deterioration in AI sentiment. To do that, we’ve drawn on both the academic evidence from historical bubbles and the specifics of the AI ecosystem today.

A more reliable guide

The valuation of stocks – their prices relative to their profits, sales or other accounting measures – is a useful starting point in assessing market exuberance. But valuations alone are a notoriously unreliable guide to short-term performance. And US tech sector valuations are still far below the highs they hit in the dotcom bubble at the turn of the century. Focusing on valuations in isolation risks missing a lot of other potentially relevant information.

A much broader view of market activity and performance can help us gauge the mood more accurately. For example, there’s clear evidence, across countries and in data going back all the way to the 1800s, of a link between rising private-sector debt (as opposed to government debt) and subsequent crises. Booms fuelled by private debt are inherently unstable, as the economist Hyman Minsky famously illustrated. Any surge in corporate debt issuance – something which is not happening currently – would therefore be a potential warning sign that financial markets could be near a tipping point, or ‘Minsky moment’.

When the party gets too crowded

This chart shows US private equity fundraising as a share (%) of overall corporate equity value, a useful barometer of investor exuberance.

A message from the past

The final innings of several recent booms have also been accompanied by rapid increases in firms issuing new equity on public markets, coupled with surging fundraising in private markets. Both those things are essentially symptoms of a party getting out of control. It pays to be cautious when investors are rushing headlong into private equity without due care for the details – as they did ahead of the dotcom crash, the global financial crisis, and the period of speculative excess in US equities just after the pandemic.

It’s a similar story with corporate dealmaking – mergers and acquisitions. A rapid rise in dealmaking can be a sign of overconfidence, with buyers paying over the odds for targets which are a poor fit. The disastrous US mega-deal between email service provide AOL and traditional media conglomerate Time Warner at the peak of the dotcom bubble is the classic example. Both equity raising and corporate dealmaking are still currently subdued – but we continue to monitor them in case that changes.

Another related factor is the volume of trading activity in markets. The late stages of bubbles have often been associated with frenzied increases in trading, particularly in options markets in more recent history.

Options trading has been on the rise again, but not yet as sharply as it was in the speculative boom of 2021, or immediately prior to the dotcom crash and global financial crisis.

Speculation heats up

The rolling 12-month growth (%) in equity options trading provides a window into rising risk-taking and late-cycle market behaviour.

A different crowd at the party

We’re also closely watching the relative performance of different types of stocks. In the late stages of the dotcom bubble, there was an extraordinary outperformance from the least profitable stocks, from very young stocks (those of firms founded only a few years ago), and from recent winners (top-performing stocks maintained their winning streaks to a very unusual extent). This period of sustained outperformance from unprofitable new stocks portended the eventual crash. Markets today are not, to quote the singer Prince, partying like it’s 1999. But once again, any change on that front would make us more cautious.

The factors we’ve outlined so far are drawn from experience. But no historical boom is a perfect mirror of the current one. The partygoers this time are a different crowd. There’s no historical equivalent of the AI industry. We’ve identified a basket of 169 global stocks whose business models have a strong exposure to AI. As a starting point, we monitor the price performance and sales growth of this basket. But we also go well beyond that, delving into the minutiae of the AI supply chain.

A whole new ecosystem

This starts with an understanding of the structure of today’s AI ecosystem. Just five companies – the so-called ‘hyperscalers’, Amazon, Alphabet, Meta, Microsoft and Oracle – dominate global investment in AI. They directly account for nearly two-thirds of the investment spending for our basket of AI stocks, and indirectly an even larger proportion. If their AI investment spending slowed sharply, the whole supply chain would dry up.

To keep an eye out for any early warning signs of this, we monitor developments at key points throughout the supply chain, both ‘upstream’ and ‘downstream’ of the hyperscalers. Ironically, AI-based tools help us here. Especially a tool that reads and synthesises the information from huge volumes of documents, including companies’ regulatory filings and earnings calls with investors – more than even a large team of human analysts could.

We use this tool to summarise what users of AI, the hyperscalers’ customers, are saying about their plans for adopting AI, including the extent to which they can generate new revenue or cut costs from it. And we use it to summarise what the hyperscalers’ suppliers, plus their suppliers in turn, are communicating. In this way we can monitor demand for key products related to AI investment, and information about new orders and backlogs.

We track sentiment separately from firms supplying AI chips, memory and servers – key ingredients for the data centres behind the AI revolution. We even look at firms providing equipment to chipmakers, firms supplying cooling solutions to data centres, and real estate investment trusts that specialise in leasing data centres. This could help us quickly spot any faltering either in demand for AI products or of the hyperscalers’ investment plans. That might show up in weakening orders to key suppliers before any official announcement from the hyperscalers themselves, for example.

Not too early, not too late

It’s impossible to anticipate turning points in markets exactly. But compared to relying on valuations and anecdotes alone, our evidence-based framework should give us a much better chance of spotting whether optimism about AI has reached unsustainable levels. It should help us balance the twin risks of leaving the party too early, or too late.