For a truly staggering figure, look no further than an estimate from Columbia Business School Professor Stijn Van Nieuwerburgh of the cumulative investment in artificial intelligence-related infrastructure by 2032. In a paper last month, Van Nieuwerburgh said the AI buildout could reach US$10.3 trillion, the equivalent of 3.6 per cent of annual US economic output for the next seven years.

The average annual capital expenditure dwarfs previous infrastructure booms, including railways, highways and plumbing for the internet. This year alone, the big US “hyperscalers” – the leading technology companies and cloud service providers – will sink more than US$750 billion into capital investments in a race to dominate AI.

Since the launch of ChatGPT in November 2022, the AI boom has evolved from a popular investment theme into a massive economic and financial bet. It has also been the main source of resilience in the global economy and markets. Despite a succession of severe shocks, equity markets stand at a record high, the global economy is growing at its fastest pace since May 2023 and corporate earnings have risen dramatically.

In a report on September 25, George Saravelos of Deutsche Bank said the global capital expenditure cycle – which is also driven by the ramp-up in defence spending and investments in building resilient supply chains – is the “superglue holding the market together”.

On October 13, the International Monetary Fund (IMF) admitted it had underestimated global growth three years in a row mainly because of the “positive demand shock” from AI. In Bank of America’s latest global fund manager survey on October 13, investors had the biggest overweight position in the technology sector since August 2020.

The most talked-about risks to the AI boom pertain to the economics of the technology, specifically whether there will be enough demand from end users quickly enough to justify the enormous sums of money being spent. Another concern is the increasing constraints to the AI buildout, particularly given the fierce backlash against the construction of new data centres in some countries and mounting calls for regulation.

However, it is the boom’s profound impact on markets that is the biggest risk in the short to medium term. As a report last month from asset management firm Alliance Bernstein noted, the AI boom has become “the everything trade”. Not only is it “all-consuming in terms of investors’ attention”, “its demand for capital is creating linkages across asset classes that make diversifying AI exposure significantly harder”.

Until recently, worries about concentration risk were confined to stock markets. As the hyperscalers’ free cash flows dwindle and force the tech giants to rely on external sources of finance, though, the AI trade has broadened to bond and credit markets. Hyperscalers now account for nearly 10 per cent of total investment grade corporate bond issuance, up from 0.5 per cent in 2024, according to IMF data.

Morgan Stanley estimates that credit and corporate debt markets will account for roughly half the global capital expenditure on data centres through 2028. The dramatic shift in the financing of the AI buildout exacerbates vulnerabilities in global markets.

As investor portfolios become a one-way bet on AI and the infrastructure buildout increases cross-asset linkages in markets, the probability of the boom triggering the next financial crisis has risen sharply.

A disorderly and sustained unravelling of the AI trade would be far more damaging today than just a year ago. The bigger the leverage in the AI ecosystem, the harder the landing in the event of a loss of confidence in the monetisation of the technology.

Although the US hyperscalers have strong balance sheets and ample liquidity buffers, the surge in AI-related debt increases the risk of spillovers to broader markets, especially given the build-up of leverage in stock markets amid the rise in interest rates.

South Korea is a cautionary tale. The country’s semiconductor-led stock market – the best-performing major market in the first half of this year – has succumbed to a brutal, leverage-driven sell-off, plunging around 34 per cent since June 22. It is now the worst performer among its peers in the second half.

Also, the politics, economics and financing of AI are not conducive to a prolonged, multi-year boom. AI doom scenarios are proliferating while Chinese AI model usage overtook US platforms globally for the first time in June. As Van Nieuwerburgh notes, “Long-duration debt and contractual claims are being written against assets whose utilisation, technological relevance and residual value remain unusually uncertain.”At some point, something is bound to break somewhere. Tellingly, global fund managers believe hyperscalers’ capital expenditures are the most likely source of a “systemic credit event”.

This does not mean a crisis is imminent. However, the fact that such a massive bet is based on such uncertain foundations is deeply concerning. The “everything trade” is an accident waiting to happen.