The real AI risk is overcommitment, not low returns

The AI debate is no longer about whether AI investments can be monetised, but whether the industry can manage the scale of its long-dated capital commitments.

Sundeep Gantori Chief Investment Officer, Equities

The latest correction in AI-related equities has revived a familiar question – are technology companies spending too much on AI? At first glance, the concern is understandable.

A year ago, investor scepticism around AI was largely about monetisation. Companies were spending heavily on AI infrastructure, but the revenue opportunity still felt uncertain. The latest Q2 corporate earnings season has helped address this concern as AI-related cloud demand has continued to surprise positively. Moreover, enterprise adoption remains healthy, and customer revenue backlogs have expanded meaningfully. Demand visibility is also increasingly extending into 2027 across advanced semiconductors, packaging, networking, power and cooling infrastructure. At the same time, the market is demanding more discipline and is no longer prepared to simply reward AI spending.

In our view, investor concerns now centre on the execution risks associated with the substantial AI investments. Infrastructure must be built first, while revenues and profits arrive only when capacity becomes operational and utilisation improves. As a result, near-term free cash flow can look weak even when long-term returns remain attractive.

A closer look at recent Big Tech earnings reinforces why we continue to favour Big Tech for its steady risk-reward profile.

Viewed through a traditional capital allocation lens, AI infrastructure investments’ return profile appears supportive. The key takeaway is that AI monetisation appears to be improving, not weakening. Cloud growth remains resilient, AI-related workloads continue to scale and management commentary across the industry continues to point to demand exceeding available supply.

Based on our estimates, every incremental USD1 invested in AI infrastructure needs to generate only around USD0.15-0.20 of annual revenue to earn a cost of capital of approximately 10%. Current industry performance suggests Big Tech cloud businesses are generating returns well above this hurdle, with AI-related returns on investment (ROI) estimated at around 25-30%. This attractive ROI is supported by strong enterprise adoption, rising cloud utilisation and continued supply constraints across critical parts of the AI value chain.

We believe market uncertainty now is more around the execution risk in realising these attractive returns, with hyperscalers having committed significant amounts of capital. Much of the commitment to suppliers is effectively locked in for years, reducing management flexibility should AI demand slow unexpectedly or the macro environment deteriorate sharply. Put differently, while the return on each incremental dollar remains attractive, the uncertainty is more about the ability of hyperscalers to pause, defer or reverse the investment cycle should macro conditions deteriorate. However, we expect AI adoption to be a steady tailwind, possibly rising even through an economic downturn, given the productivity gains it offers. Given AI adoption is currently at only around 20-25%, overall market penetration remains relatively low with a long runway ahead.

In the AI era where “big is getting bigger”, we expect the largest platforms to make larger infrastructure spending, secure long-term customer demand and monetise AI across their multiple revenue streams, including cloud, advertising and subscriptions. This gives Big Tech a more diversified and durable return profile than companies exposed only to one part of the AI supply chain – firmly reinforcing our confidence in Big Tech’s ability to execute well on the substantial AI investments.

Semiconductors – one of the key components in the AI value chain – were at the epicentre of the recent tech correction. The sector has been under pressure as investors questioned whether the AI capital expenditure cycle could continue at its current pace. However, our industry checks across Taiwan and South Korea suggest demand remains strong, with visibility extending into 2027.

We believe the market may be underestimating how broad the semiconductor opportunity has become. AI infrastructure requires far more than just leading-edge processors. Each new generation of AI system requires more advanced packaging, substrates and optical interconnects, denser printed circuit boards and greater power capacity. In light of these strong fundamental drivers, the recent correction appears overdone relative to the underlying demand backdrop. Much of the weakness seems to have been driven by positioning, deleveraging and headline risk rather than a meaningful deterioration in fundamentals.

That said, after a strong multi-year run, the semiconductor opportunity is now becoming more selective. The easy money from the early phase of the AI trade may be behind us, and the next phase will likely require a healthier mix of structural and cyclical exposure.

For investors, the message is to stay invested in the AI theme, but with greater discipline. We continue to view Big Tech as a core allocation within technology portfolios, complemented by selective exposure to semiconductors and other beneficiaries of the broader AI infrastructure cycle.