After a long period of subdued investment, US capital expenditure has accelerated sharply over the past year. This has been led by the US hyperscalers and concentrated in IT equipment, software and data-centre infrastructure. The scale and concentration of this spending have been sufficient to revive the US exceptionalism narrative, with investors increasingly extrapolating the idea that AI will drive a new US-led productivity boom.

Given the healthy margins, internal financing and explicit intentions of firms to continue investing in AI, it is reasonable to expect that (at least for now) the capex narrative powers on. However, it is not a one-sided story. Key vulnerabilities include competition for capital and associated inflationary pressure, potential deterioration in corporate balance sheets, the increasing concentration of US growth in a narrow set of AI-driven activities, and the timing gap between investment today and returns tomorrow.

The AI capex cycle is clearly a very powerful driver of the US economy and markets, and the near-term fundamentals remain strong. Profit margins across many AI-linked businesses continue to expand, returns on capital are high, in aggregate the investment is still being funded from cash flows rather than excessive borrowing, and firms are signalling strong spending intentions. Little in the current data suggests an imminent turning point.

However, the uncomfortable fact is that two things can be true at once: the AI capex boom can be economically significant and durable in the near term, whilst also becoming increasingly extended in market expectation and asset terms and unlikely to persist indefinitely. Strong fundamentals do not necessarily translate into good investment opportunities.

This dynamic has clear historical precedent. Previous capex booms, most notably during the late 1990s, began with strong earnings growth but ultimately gave way to overinvestment, increasing competition and falling returns. Whilst today’s starting point is stronger – particularly given current levels of profitability – the underlying dynamic of self-reinforcing investment and the potential for misallocation of capital are unchanged. High levels of spending support near-term earnings, which in turn encourage further investment, creating the conditions for an eventual overshoot.

It is therefore useful to distinguish between the build phase and the return phase of the AI cycle. What we are seeing today is predominantly the former: a period when capital is deployed at scale, driving activity, earnings and sentiment. The critical question for markets over the coming months and years is whether these investments ultimately generate sufficient returns. The risk is that the focus shifts from the benefits of capex to the returns on that capex.

There are already signs of this tension. Whilst spending remains robust, visibility on the ultimate profitability is limited, and the business models required to justify the scale of investment are not yet fully established. The US Census Bureau’s Business Trends and Outlook survey, which includes AI adoption data, suggests corporate uptake is gradual rather than explosive, creating a gap between what markets are pricing and what the real economy is delivering. For example, according to the survey, more than two‑thirds of US firms said they did not use AI across any business functions, and more than half of firms said they did not plan to use AI over the next six months. The technology may prove transformative, but the timing mismatch is key – markets already discount benefits that may take longer to materialise than investors are willing to wait.

This cycle is also more capital-intensive than previous waves of technological innovation, with large-scale infrastructure replacing the asset‑light models of the past. This raises the risk of increased competition for capital (both between corporates and across the broader economy) and introduces the risk that balance sheets deteriorate if returns fail to materialise.

This matters for inflation. Whilst much of the AI debate is framed around whether the technology will ultimately be disinflationary, the near-term impact may well be the opposite. A large-scale build-out drives demand for capital, energy, labour, equipment and financing and with it brings the risk that further inflationary pressure could come first, with productivity-driven disinflation arriving later. The effects are already becoming visible. Demand from AI data centres has pushed up memory and storage component costs, leading companies including Apple and Microsoft to increase prices for consumer products.

At a macro level, whilst this wave of investment should support stronger global growth, the effects are likely to be most pronounced in the US and in Asian export economies such as South Korea and Taiwan, with spillovers into Japan and beyond. Stronger activity, higher employment and rising wages are all plausible near‑term outcomes.

However, these dynamics also sow the seeds of their own reversal. Stronger labour markets will put upward pressure on wages, whilst competition and supply chain constraints lead to higher input costs, which would compress margins. Over time, rising costs and tightening financial conditions are likely to constrain further investment, leaving the cycle to eventually collapse under its own weight.

Our view is that AI will continue to support earnings and growth in the near term, but that the anticipated productivity boom may not materialise quickly enough to sustain the current cycle. The risk is that a repricing in earnings expectations and asset values occurs before those gains are realised. To that end, evidence of the transition from building infrastructure to generating sustainable returns will be a key focus of markets in the months ahead.

Today’s challenge for investors is that the most important source of market strength may also be its greatest vulnerability. The AI investment cycle continues to support growth, earnings and asset prices, yet both economic and market performance are becoming increasingly dependent on this single theme continuing to deliver.

The most significant risk facing markets therefore is a reassessment of the returns AI-related investment can generate. This matters because recent equity market performance has been unusually concentrated in a small number of AI-linked companies. The ten largest companies now make up around 40% of the S&P 500 by market capitalisation, more than double their share a decade ago and well above the dot-com peak of c27%. As a result, both market leadership and broader investor confidence have become increasingly dependent on the continuation of the AI investment cycle.

Given the importance of this spending to both US growth and market leadership, any meaningful scaling back of capex plans would represent a significant demand shock. In a highly financialised economy, the consequences would be far-reaching, with weaker growth, lower asset prices and a disinflationary impulse as investment slows. We view this as a genuine risk, but one with highly uncertain timing.

However, a reversal is not the only risk. The cycle may instead continue and broaden beyond AI-related capex and today’s narrow group of beneficiaries. In the near term, this would support stronger growth, rising employment, and firmer household incomes. Yet the success of this cycle may create its own challenges. Sustained investment in AI infrastructure, energy systems and associated industries would increase demand for labour, capital and raw materials, placing upward pressure on wages, inflation and interest rates.

This would lead to a more conventional end to the cycle. Rather than a collapse in spending, overheating eventually forces a tougher policy response. Higher real rates and tighter financial conditions eventually squeeze margins, slowing investment. Higher borrowing costs would also compound the pressures facing already indebted governments, limiting their ability to cushion any eventual downturn and leaving economies more vulnerable to shocks. In such an environment, the geopolitical risks discussed remain firmly in the wings, with the potential to amplify already fragile conditions.

Whether the cycle ultimately ends through disappointing returns, or overheating and tighter financial conditions, investors will eventually shift their focus from the scale of investment to the returns generated by that investment. The productivity benefits of AI may ultimately prove transformative, but there is a meaningful risk that earnings expectations, asset prices and government finances come under pressure before those benefits are fully realised.