Will the AI trade unwind?

SINGAPORE – “To light a candle is to cast a shadow,” said acclaimed American science fiction doyenne Ursula K. Le Guin. Her moral was that progress inevitably comes with costs.

Today, that duality is playing out in artificial intelligence. Of 1,500 US consumers recently polled by Bain, 25 per cent of millennial and Gen Z respondents said they mostly or always default to AI chatbots instead of traditional search engines. AI’s undeniable utility has already quietly rewired decades of internet habits.

Yet its extraordinary promise has come at an extraordinary cost, including massive infrastructure investment, and surging compute and power constraints. Frontier labs such as OpenAI and Anthropic are spending aggressively to stay ahead, while lower-cost Chinese open-weight models are narrowing the capability gap. And with the semiconductor index down more than 20 per cent over the last month, investors are questioning whether the AI trade could unwind more broadly.

To be sure, this sentiment shift is happening against a still-favourable backdrop. The global economy remains resilient, the second-quarter earnings season has begun strongly, and market leadership has broadened to include cyclicals and previous laggards. Market momentum has persisted amid robust sector and factor rotations, and multi-decade-low stock index correlations, while single-stock volatility has surged.

Looking ahead, we are still constructive on the AI structural theme. But risks are indeed building, requiring a more balanced portfolio approach. Specifically, the AI trade’s next phase likely hinges on three questions.

First, can hyperscalers translate massive AI investments into sustainable revenues, cash flows and returns on capital? Second, will the intensifying global AI race fuel more chip, compute and infrastructure demand? Third, will governments seeking technological sovereignty become significant new sources of investment? The answers will determine whether the current AI boom matures into a durable investment cycle or faces a more challenging adjustment.

From AI ambitions to AI returns

For investors, the most immediate question is whether AI revenues justify AI investment.

The current megacap technology earnings season, which began with Alphabet, is being heavily scrutinised for evidence of AI monetisation, revenue durability and free cash flow discipline. The four largest US hyperscalers are likely to spend around US$730 billion (S$936 billion) on AI-related capex in 2026, supporting an unprecedented infrastructure build-out. Globally, we forecast AI capex investment of approximately US$900 billion in 2026 (up 84 per cent year on year) and US$1.2 trillion in 2027 (up 33 per cent year on year). Meanwhile, cash capex should exceed operating cash flow for some of the industry’s biggest companies by end-2026.

None of this suggests the AI investment cycle is ending. Demand for compute, memory and power remains robust. Supply constraints are unlikely to disappear soon. Component shortages and pricing strength could even provide further tailwinds into 2027.

Beyond that, however, visibility becomes less certain. We view 2028 as a potential period of capex digestion, and investors could focus increasingly on the risk of slowing spending.

In short, markets are moving from funding AI ambitions to measuring AI returns.

Iron sharpens iron

While investors have become focused on near-term economics, they should not overlook the strategic nature of AI competition.

The emergence of increasingly capable Chinese models – from DeepSeek’s R1 in January 2025 to Moonshot AI’s latest Kimi K3 model, the largest open-weight large language model globally – underscores how the global AI race remains intense. China’s advances are clearly not one-off and will likely invite more US scrutiny.

Much of the debate has centred on export controls, intellectual property and AI “distillation” techniques. Regulatory pushback remains a near-term risk as Washington explores measures that could hinder the mass adoption of Chinese AI models.

For investors, the more important implication is that competition should sustain demand. In fact, tech giants currently view AI as an existential battle. Companies could keep buying hardware despite low near-term returns to prevent rivals from gaining a lasting technological advantage. Paradoxically, greater competition from open-weight models such as Kimi K3 may prolong, rather than shorten, the AI investment cycle.

Sovereign AI’s rise

In the longer term, governments themselves could become a significant source of AI demand.

Policymakers globally are paying more attention to the strategic importance of AI infrastructure, data control and technological independence. Domestic AI strategies under way include the European Union’s tech sovereignty package, India’s Sarvam platform for multilingual AI models and Canada’s push to reduce reliance on US technology providers.

While it remains too early to assess the ultimate impact of localised, state-funded AI strategies, they spotlight how the future customer base for AI infrastructure may extend well beyond Silicon Valley and traditional hyperscalers. Industry estimates already suggest sovereign AI comprises about one-third of compute demand today.

Diversify, not divest

To invest, our overall view is that the AI supercycle is intact, but investors should be mindful of a “capex taper tantrum” risk. Volatility could remain elevated, but it does not mean investors should abandon the AI trade.

In fact, the agentic AI shift should accelerate monetisation and significantly increase token demand, boosting the Asian AI hardware supply chain. With earnings growth set to remain strong into 2028, we still see around 10 per cent upside for the AI theme over the next 12 months.

Within portfolios, we favour a balanced, barbell-style approach to AI exposure. On the growth side, we prefer semiconductor capital equipment companies, where revenue growth could compound at well above 30 per cent annually over the next three years. Foundries also offer more attractive growth-adjusted valuations than many other AI-linked segments, and sustained demand should continue to support the compute layer.

As a balance to that growth exposure, we favour defensive technology such as payment networks, data centre real estate investment trusts, select smartphone makers and consumer electronics. These businesses offer resilient earnings profiles and lower sensitivity to the AI capex cycle while preserving exposure to AI’s growth story.

Meanwhile, in semiconductors, we have significantly trimmed our overweight exposure and prefer to see overleveraged positioning unwind before buying the dip.

Le Guin’s insight remains relevant today. The light of technological progress continues to illuminate new possibilities. But, equally, investors should stay attuned to the growing shadow of risks – and retain the discipline to adjust their portfolios – to stay ahead.

Source : https://www.straitstimes.com/business/will-the-ai-trade-unwind