There are good reasons for investors to be bullish about artificial intelligence (AI). The earnings of technology firms in the S&P 500 are expected to increase 54 per cent this year. For the index as a whole, profits will rise 33 per cent, turbocharged by the tech and energy sectors, data from FactSet shows.

In Asia, the impact of the AI boom is even more profound. Corporate earnings in South Korea and Taiwan, the two leading semiconductor hubs, are expected to grow a staggering 283 per cent and 73 per cent respectively this year.

There is even a silver lining in the recent turmoil in equity and corporate debt markets caused by worries over the financing of the AI infrastructure buildout. The sharp fall in the tech-heavy Nasdaq 100, which briefly entered correction territory on October 30, has helped drive down yields on US Treasury bonds as investors search for safe havens.

Yet a cursory glance at the scale and severity of the sell-off last month shows the AI spending spree has entered a new and more dangerous phase. Exceptionally strong profits and revenues from the United States’ “hyperscalers” – the leading tech companies and cloud service providers – are no longer a source of reassurance.

Mounting concern over the rapidly escalating buildout costs – hyperscalers’ combined capital spending is expected to increase 96 per cent this year to US$807 billion – and whether the investments will start to pay off in a meaningful way has put the economics and financing of AI under sharper scrutiny.

Anthony Saglimbene of Ameriprise Financial says the “AI math problem” has become more acute. The growth in tech firms’ capital expenditures far exceeds the increase in their revenues. “In our view, that gap can’t stay this wide indefinitely. Either revenue catches up more quickly, or spending plans get reset,” he said.

The problem, however, is that the risks in financing the boom – tech companies’ debt issuance has risen dramatically in the last two years as the spending spree eats into their free cash flow – are two-sided. Capital expenditure reductions would “raise questions about the durability of the investment cycle” while “further increases raise questions about financing, execution, and the eventual returns”, said Citadel Securities.

Once a rising tide that lifted all boats, the AI trade has become much more discriminating and volatile. Companies with aggressive capital expenditure plans and negative free cash flow are being punished.

A good example is Google parent Alphabet. On October 29, its shares plunged 13 per cent – the sharpest daily decline ever – after it reported a second straight quarter of negative free cash flow and revealed “AI mode” searches were eating into sales at its core advertising business. Microsoft’s share price, by contrast, has held up in part due to its positive free cash flow.

The growing pressure on tech companies to be more prudent in their capital spending while maintaining confidence in the sustainability of the AI boom leaves little margin for error at a time when other risks – higher interest rates and the backlash against the construction of data centres in several countries – are weighing on the AI trade.

This makes the timing of the stock market debut of Anthropic, the maker of the chatbot Claude, this month inopportune. The first company focused on frontier AI to go public, Anthropic plans to spend a staggering US$895 billion on computing power in the years ahead even though the growth in its revenues from AI services is already slowing, possibly due to growing competition from cheaper Chinese models.

In a sign of the scale of the challenge confronting the leading tech companies, a report by Gartner last month found that over US$3.5 trillion of AI infrastructure spending could be “exposed to impairment” by 2030 because of insufficient revenues to cover financing costs and generate attractive returns.

What is clear is that AI’s mathematical problem will become more acute given the unprecedented surge in capital spending. While tougher curbs on data centre development – a near certainty in the US following the sweeping victory of the Democratic Party in the midterm elections – would make current AI infrastructure even more valuable, a sharp unwinding of the boom would wreak havoc on the global economy and markets.

Asia is the most exposed. The region’s tech sector – in particular the semiconductor industry in Taiwan, South Korea and Japan – has been the biggest beneficiary of the surge in AI spending, which “helped to insulate East Asian economies from what would have otherwise been a severe terms-of-trade shock”, said Carlyle.

Yet in Taiwan, whose economy grew a staggering 14.3 per cent in annualised terms last quarter, foreign outflows from the stock market reached a record high last month. The AI sell-off is no blip.