Temasek's chief investment officer called an unwind in the artificial-intelligence trade the largest current market risk, even while saying the Singapore investment company remains bullish on the technology over the long term. The apparent contradiction reveals the portfolio issue: a theme can be structurally attractive and still become dangerous when valuations, ownership and liquidity depend on uninterrupted earnings delivery.
Rohit Sipahimalani said Temasek did not see an AI reversal as imminent but expected possible “bumps” in 2027. He pointed to strong earnings among leading AI companies as a reason U.S. indexes had remained near records despite higher Treasury yields. He also noted that roughly half of Russell 3000 stocks were at least 20% below their June highs, illustrating how a small group of winners can conceal weak breadth.
Why liquidity changes the risk
About half of Temasek's AI exposure was in publicly traded assets, and Sipahimalani said the firm would prefer that share to reach roughly 70%–75%. Public holdings can be resized more quickly when product cycles, regulation or customer economics change. Private stakes in companies such as OpenAI and Anthropic may offer direct exposure to frontier models but can be harder to exit and value during a downturn.
Moving from half to three-quarters public exposure does not necessarily mean reducing total AI exposure. It changes the option value of the portfolio. Listed securities provide daily prices and deeper markets, while private assets can remain locked through changing assumptions. The trade-off is that public names may also react faster and more violently to an earnings miss.
What could trigger the unwind
Temasek identified tighter safety regulation and evidence that customers are not earning adequate returns on AI spending as possible catalysts. The second mechanism is especially important. Hyperscalers and enterprises can sustain heavy capital expenditure only if new products, efficiency gains or pricing justify the compute bill. If revenue trails depreciation, power and financing costs, suppliers across chips, servers and data centers could face lower orders.
Concentration would amplify that adjustment. When a few companies represent a large share of an index, their disappointment affects passive funds and benchmarked portfolios even if most listed companies were never priced for an AI boom. Weak breadth therefore reduces the market's shock absorbers.
The comment is not a crash forecast. It is a reminder to separate technology adoption from security-level return. Durable AI demand may coexist with poor returns for projects bought at excessive valuations or financed without clear cash generation. Temasek's preferred mix suggests maintaining exposure while increasing the ability to pivot.
The decisive evidence will come from customer return disclosures, hyperscaler capital budgets, AI-linked earnings and any change in regulation. If those metrics support cash generation, volatility may create opportunity. If they weaken together, liquidity and concentration will matter more than the long-run promise of the technology.