The global AI trade ran into a financing problem on October 15. Asian and European semiconductor stocks fell sharply after hotter U.S. inflation revived expectations of further Federal Reserve tightening and pushed investors to question who will fund the next wave of data-center construction.
South Korea's SK Hynix fell more than 10% and Samsung Electronics dropped 8.7%. In Japan, Advantest lost almost 14% and SoftBank Group fell 7.9%. ASML, Infineon and ASM International also declined in Europe, extending the prior Wall Street selloff.
The macro trigger was clear. The U.S. consumer price index rose 0.6% in September and 3.7% from a year earlier, according to the Bureau of Labor Statistics. The Federal Reserve had already raised its target rate by a quarter point in September to a range of 3.75% to 4.00%. Another inflation surprise increased the risk that capital remains expensive for longer.
Why financing matters to AI
AI infrastructure requires enormous upfront spending on chips, power systems, networking and buildings. Higher interest rates raise the cost of debt and reduce the present value of cash flows expected many years in the future. The effect is most severe for projects that rely on optimistic utilization assumptions or customers with weak balance sheets.
This does not mean AI computing demand has disappeared. Memory producers and equipment suppliers still report strong orders, and Micron's latest results show exceptionally high revenue and cash generation. The market's concern is that demand can remain strong while financing becomes less attractive.
That distinction changes which evidence matters. Announced capital-expenditure budgets are useful, but signed contracts, customer credit quality, prepayments and project-level financing terms offer better protection against a funding slowdown.
Competition adds a second risk
China's ChangXin Memory Technologies reported progress on a denser memory architecture, prompting fears that new capacity could pressure supply and pricing later in the decade. Analysts cited in CNBC's report noted that CXMT still trails leaders in advanced high-bandwidth memory and has not demonstrated large-scale production of the new design.
The immediate competitive threat is therefore less certain than the share-price reaction suggests. Yet the announcement reminded investors that today's shortages can attract tomorrow's capacity, especially when governments view semiconductors as strategic.
The selloff should not be read as proof that the AI cycle has ended. It is a repricing of two assumptions that had supported high valuations: that capital would remain readily available and that incumbent suppliers would preserve unusually strong economics. Investors should now watch hyperscaler spending, cancellations, GPU rental prices, memory inventories and the cost of financing. Those indicators will reveal whether the move is a valuation reset or the start of a weaker demand cycle.