The dominant investment theme of the past two years has been under an unforgiving spotlight in recent weeks as six of the Magnificent Seven have lifted the lid on their second quarter results. Technology, and in particular artificial intelligence (AI), has been the most popular and crowded trade in the two-year-old bull market, accounting for the lion’s share of the US market’s gains so far in 2024 – 60% of them from just six stocks. But both the sector and theme have recently come under intense scrutiny.
Will the massive AI investment generate adequate returns?
In particular, investors are starting to question whether the gargantuan sums being poured into AI-related infrastructure - the key talking point of the Big Six’s earnings statements - will ever earn investors an acceptable return. Comparisons are being drawn with the dot.com bubble of 25 years ago and even, more worryingly, with the railway mania of the 1840s, with which the ongoing AI mania has some similarities.
The rotation in the leadership of markets away from technology is partly about the improving case for the smaller companies that have underperformed for so long. But it is also in large part a sign that investors are becoming anxious about the amount of unproven growth baked into the AI leaders’ share prices. Markets have become less tolerant of companies’ inevitable failure to beat expectations quarter in quarter out. Perhaps most worrying from an investors’ perspective is that where these stocks go, the rest of the market inevitably follows, given their weight in the index and the growing importance of value-agnostic passive investment.
The mistake investors made a generation ago was to think that the big winners from the internet would be the companies that built its infrastructure. The likes of Cisco and Intel went to the stars but crashed back to earth when the scale of their over-investment became clear. Investors today are becoming increasingly concerned about the daunting gap that appears to be opening up between the revenue forecasts implied by the current breakneck AI infrastructure build-out and the actual sales that are likely to be delivered off the back of it.
Sequoia Capital’s David Cahn thinks US$600 billion of annual revenue is required for a reasonable payback. But OpenAI, which dominates the industry’s sales, has reached only just over US$3 billion. He reckons there might be a US$500 billion shortfall. Total annual investment in AI might top US$1 trillion by 2027, according to former OpenAI executive Leopold Aschenbrenner. This is a large sum - worth 3% of US GDP - and it easily outstrips both the Manhattan and Apollo projects, which at their peak reached just 0.4% of economic output - US$100 billion a year in today’s money.
Comparisons to past investment bubbles
But spending US$1 trillion a year on AI investment, while dramatic, would not be unprecedented. In the five years to 2001, telecoms companies spent a similar amount. Many trillions are being spent on the green transition. For years, China has spent 40% of its output on investment. We borrowed 100% of our GDP in the Great War and then did it again 20 years later in the Second World War.
But the most salient comparison is perhaps with the British railway mania of the 1840s when a cumulative 40% of UK GDP was poured into the AI of the Victorian era. This was the equivalent today of US$11 trillion over a decade, roughly the run rate that’s being forecast for the upcoming AI investment round. Nearly 200 years ago, this did not end well.
In the two years after 1843, according to bubble historian Edward Chancellor, the value of British rail stocks doubled. Hundreds of railways were proposed, with investment peaking at £40 million, or 7% of national income. Inevitably it turned out to be a massive misallocation of resource. There were three separate lines between Liverpool and Leeds, for example. To pay investors a satisfactory return, revenues and passenger numbers would have needed to rise five-fold in five years. Growth was overestimated and costs spiralled out of control. Dividends were slashed as returns collapsed. Within five years, railway shares had lost 65% of their value.
Technology bubbles are always subtly different, but they have shared characteristics. They latch onto a new technology about which vaunting claims can plausibly be made. Investors put to one side traditional valuation measures. A massive over-commitment of capital is made and poorly directed.
There is a key difference between the railways of the 1840s and AI today. When the mania took hold in Victorian Britain, the case for the new technology was already well understood. Railways then benefited from greater intrinsic pricing power than computer processing today because there is a physical limit to how many tracks you can lay between one place and another. Without this advantage, prices inevitably are competed down to their marginal cost.
Investors often underestimate the pace at which expensive kit becomes obsolete. Sequoia’s Cahn said it well: ‘speculative investment frenzies often lead to high rates of capital incineration’. Literally, money to burn.
The four phases of AI
Goldman Sachs has identified four phases in the AI value chain. Worryingly the investment returns are falling away fast from one to the next. Nvidia was phase one; year to date its shares have risen 165%. Phase two are the infrastructure builders creating the chips, data centres and large language models. The average stock in this group has risen 26% so far in 2024. But investors are sceptical about the returns in the later phases - the software firms in phase three have already provided disappointing forward-looking commentary about their ability to monetise AI. Phase four companies are those with the biggest potential earnings boost from widespread AI adoption and productivity gains. For them the wait goes on.
While share prices were heading higher, no-one was looking too hard at reasons not to believe. If the rotation of recent weeks persists, that willing suspension of disbelief could be tested. If you don’t know how exposed your portfolio is to the AI story, now would be a good time to find out.
Tom Stevenson is an Investment Director at Fidelity International, a sponsor of Firstlinks. The views are his own. This document is issued by FIL Responsible Entity (Australia) Limited ABN 33 148 059 009, AFSL 409340 (‘Fidelity Australia’), a member of the FIL Limited group of companies commonly known as Fidelity International. This document is intended as general information only. You should consider the relevant Product Disclosure Statement available on our website www.fidelity.com.au.
For more articles and papers from Fidelity, please click here.
© 2024 FIL Responsible Entity (Australia) Limited. Fidelity, Fidelity International and the Fidelity International logo and F symbol are trademarks of FIL Limited.