Broker models three scenarios for a reversal in tech spending, with bear markets across US, UK and European equities in every case
The artificial intelligence investment boom is already 60% larger than the dotcom bubble of the late 1990s and shows increasing signs of "irrational exuberance", according to a major new strategy note from Panmure Liberum that models the potential fallout if the cycle goes into reverse.
Strategist Joachim Klement argues that technology capital expenditure now accounts for almost 100% of US GDP growth, a concentration never seen before, and that the growth expectations baked into hyperscaler spending plans are higher than anything recorded since the end of the Second World War.
The impossible maths
Panmure calculates that the five largest US hyperscalers (Amazon, Microsoft, Alphabet, Meta and Oracle) need to find between $2 trillion and $5 trillion in additional annual revenue to justify their planned data centre investments, which are projected to reach $658 billion in 2026 alone.
At current capex-to-sales ratios, the broker estimates hyperscalers would need to quadruple their revenues without any meaningful increase in costs. With the exception of Amazon, the implied return on invested capital from planned spending is deeply negative.
The note also questions whether OpenAI and Anthropic, both expected to pursue initial public offerings this year at valuations approaching $800 billion to $850 billion, have viable business models.
Inference costs at OpenAI are projected to reach $121 billion by 2028, with losses of $85 billion. Klement argues the IPOs represent "a major risk transfer from founders to equity investors".
Hallucinations and small language models
A core concern is that large language models have an inherent hallucination problem that cannot be eliminated, undermining their use in deterministic applications such as accounting, legal work and tax filing.
Panmure also highlights the growing threat from small language models, which can perform specialised tasks at a fraction of the cost, citing one case where a podcasting company replicated a function for $33,000 that would have cost $33 million using ChatGPT.
Three scenarios, no good outcomes
The note models three scenarios for a reversal in US tech spending. Even in the mildest case, a 4.5% correction that merely unwinds the most recent quarter of excess, European and UK stock markets would enter a new bear market, with the FTSE 350 falling more than 23%.
In a more typical tech recession scenario involving a 6% decline in spending, the S&P 500 would fall close to 20% while European and UK markets could drop more than 30%.
In the most severe scenario, modelled on a repeat of the 2000-01 dotcom crash, US tech hardware stocks could fall more than 70% in the first year, the S&P 500 could lose over 30%, and European indices could decline by more than 50%.
Where to hide
UK construction stocks, including Balfour Beatty, Morgan Sindall and Keller, along with UK and European retailers such as Next, B&M, Inditex and H&M, are identified as the best hedges, with projected losses of less than 10% even in the crash scenario.
Crucially, Panmure says cheap valuations would not protect investors, finding no correlation between current sector valuations and expected drawdowns in Europe.
Not yet time to sell
Despite the warnings, Klement believes it is probably too early to exit tech stocks, arguing that central bank rate cuts later this year could sustain the bubble for another one to two years. But he urges investors to prepare for an eventual reversal.