After Google reported earnings, its share price fell below its annual moving average and entered a death cross, weakening the medium- to long-term trend. Buffett's heavy bet is now facing book losses of about US$7 billion, or about NT$224 billion, while Facebook is also struggling near its annual line. With US tech heavyweights Google, Facebook, Amazon and Intel reporting disappointing earnings and seeing sharp corrections, institutions expect four key changes and structural impacts for Taiwan's AI supply chain.
Capital concentrates in top-tier clouds, Taiwan makers' orders more stable
Institutions point out that the more than 200 emerging cloud-compute vendors at risk of failure cited by Musk, such as Nebius and CoreWeave, are themselves high-risk customers. After these firms run out of funding, AI compute will concentrate in the hands of four well-funded top-tier cloud providers: Microsoft, Google, Amazon and Facebook. Hon Hai, Quanta, Wistron and other Taiwan makers are already first-tier contract manufacturers for Microsoft and the other three major cloud providers, after the customer reshuffle bad-debt risk will fall and long-term profitability will be better protected.
Compute pivots to inference, in-house chips drive Taiwan chain
Second, over the past 2 years the market aggressively purchased chips for large-model training, as Microsoft, Amazon and other large firms are now being pushed by the market to monetize AI applications, AI infrastructure is accelerating its shift 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 outsourcing of advanced processes to TSMC, as well as long-term business opportunities for ASIC design service firms such as MediaTek, Alchip and Global Unichip.
Bottleneck remains spec upgrades, prices await re-rating after decoupling
Third, capacity remains tight, the global data-center expansion wave has not stopped despite the share-price correction, with large firms promising record-high capital expenditure this year. As Nvidia's high-power racks enter volume production and iteration this year, manufacturing bottlenecks such as water cooling, liquid cooling, high-spec power supplies and high-bandwidth memory remain, Taiwan makers that can overcome process yields and enter the high-end component supply chain will retain major advantages in long-term unit prices and gross margins. Fourth, domestic institutional investors estimate that, driven by the AI theme, Taiwan's technology supply chain will still post strong double-digit overall corporate earnings growth this year, the foundation for a medium- to long-term bull market remains intact, the current plunge is a deleveraging and overheated-valuation correction, when share prices return to reasonable or even undervalued levels it will instead provide entry points for long-term value investors.