Weak US Tech Earnings Trigger Share Correction
After Google reported earnings, its share price fell below the annual line and entered a death cross, with the medium-to-long-term trend weakening, Buffett's heavy bet resulting in a paper loss of about 70億美元, about NT$2240億元, while Facebook is also struggling near its annual line. US heavyweight tech firms Google, Facebook, Amazon and Intel reported weak earnings, with shares correcting sharply, institutions expect 4 major key changes and structural impacts for Taiwan's AI supply chain, worth market attention.
Funds Concentrate in Big Four Cloud Providers
Institutions note that the 兩百多家 failing emerging cloud-computing vendors cited by Musk, such as Nebius, CoreWeave and others, are themselves high-risk customers, after these companies run out of funds, AI computing power will concentrate in the hands of the 4 well-funded top cloud providers Microsoft, Google, Amazon and Facebook. Hon Hai, Quanta, Wistron and other Taiwan makers were already first-tier contract manufacturers for Microsoft and the other 4 major cloud providers, after the customer reshuffle bad-debt risk is lower, long-term earnings are more secure, and Taiwan makers' orders are more solid.
Shift From Training to Inference Drives In-House Chips
In the past 2年 the market疯狂 purchased chips for large-model training, as Microsoft, Amazon and other major firms are now required by the market to monetize AI applications, AI infrastructure is accelerating toward inference applications. To reduce reliance on expensive Nvidia chips, North American majors are increasing investment in in-house inference chips, such as Google TPU, which will drive advanced-process outsourcing for TSMC, as well as long-term opportunities for MediaTek, Alchip and Global Unichip and other ASIC design-service firms, the shift in computing focus from training to inference is clear.
Hardware Bottlenecks Unresolved, Valuations Back to Reasonable
The global data-center expansion wave has not stopped on the share correction, major firms have promised capital expenditure will continue to hit new highs this year, capacity remains tight. As Nvidia's high-power cabinets enter volume production and iteration this year, manufacturing bottlenecks in water cooling, liquid cooling, high-spec power supplies and high-bandwidth memory remain, Taiwan makers that can overcome process yield and squeeze into the high-end component supply chain still enjoy advantages in unit price and gross margin. Investment-trust institutions estimate that, driven by the AI theme, overall corporate earnings in Taiwan's tech supply chain still have very high double-digit momentum this year, the current plunge is deleveraging and digestion of high-valuation correction, when shares return to a reasonable or even undervalued range, it instead provides a long-term entry point.