For nearly two years, investors have treated artificial intelligence as a one-way trade. Every announcement of a new data centre, every increase in computing capacity and every surge in capital expenditure has been interpreted as confirmation that demand for semiconductors would remain virtually unlimited. Valuations followed accordingly. This week, the market received its first serious reminder that even the strongest investment themes are ultimately governed by supply and demand.
South Korean equities suffered one of their sharpest declines in years after reports that Meta Platforms is preparing to commercialise part of its computing infrastructure by selling access to excess AI computing power and foundation models. The announcement immediately triggered fears that the industry may have built more capacity than it can currently absorb. The reaction was brutal. The Kospi fell by more than 8%, while SK Hynix lost as much as 14% and Samsung Electronics almost 10%. US semiconductor names such as Micron also came under heavy pressure. The market was not simply reacting to one company’s strategy; it was questioning one of the central assumptions underpinning the entire AI investment cycle.
If Meta is considering monetising unused computing capacity, investors naturally ask a simple question: why? One possibility is that demand remains strong but is becoming more cyclical as customers optimise spending. Another, more concerning explanation is that hyperscalers collectively overestimated the immediate need for computing power. In that scenario, the industry would temporarily shift from a shortage to excess capacity. Markets tend to react long before the data confirms such shifts. The concern extends beyond Meta. Reports that Apple is exploring sourcing memory chips from Chinese manufacturers such as CXMT and YMTC have added another layer of uncertainty. For Korean manufacturers, which have enjoyed exceptional pricing power in high-end memory products, increased competition could gradually compress margins precisely when investors were assuming the opposite. None of this suggests that artificial intelligence is coming to an end. Far from it. The structural transformation driven by AI remains one of the most powerful technological revolutions of our generation. Computing demand will continue to expand as models become larger, enterprise adoption accelerates, and entirely new applications emerge. The long-term direction remains intact.
What is changing is the investment narrative. Markets had priced an almost-perfect scenario in which demand would continuously outpace supply, allowing semiconductor manufacturers to maintain exceptional pricing power while hyperscalers expanded capital expenditures indefinitely. That assumption was always optimistic. Every industrial revolution eventually reaches a phase where infrastructure temporarily exceeds immediate demand. Railways, fibre-optic networks, renewable energy, and telecommunications all experienced similar cycles. Initial overinvestment was followed by consolidation, after which long-term growth resumed. Artificial intelligence is unlikely to prove different.
The Korean market illustrates another important lesson about concentration risk. The country’s equity market has become increasingly dependent on just two semiconductor companies. When sentiment towards AI deteriorates, the entire index suffers. Leveraged retail positioning further amplifies both rallies and corrections, transforming relatively modest changes in expectations into violent market moves. This episode should therefore be viewed less as the bursting of an AI bubble than as the normal transition from euphoria towards discrimination. Investors will become more selective. Companies generating sustainable returns from AI infrastructure will continue to attract capital. Those relying purely on expectations of unlimited capacity expansion may find the coming quarters considerably more challenging.
As we have argued repeatedly, the winners of the AI revolution will not necessarily be those investing the most, but those capable of generating durable returns on that investment. The AI story is far from over. But markets are finally beginning to distinguish between growth and profitable growth. That distinction could define the next phase of the technology cycle.