DigiTimes综述：英伟达/AMD加码高端载板与封测产能，但核心内容为10月21日GTC Berlin官方旧闻与AMD 5月旧计划

AMD and Nvidia have separately invested in supply-chain capacity expansion, making substrates, packaging, and testing key priorities as AI demand diversifies. The moves show that chipmakers are extending competition beyond performance and architecture design to control over critical manufacturing resources.

High-end substrate capacity is expected to fall short of demand in 2026, with the gap likely to widen in 2027 and 2028. AMD and Nvidia have responded with investment partnerships, sending clearer demand signals for future AI chips and extending Taiwan suppliers' capacity-planning visibility into 2029 and 2030.
Nvidia prepays key partners
Nvidia CEO Jensen Huang announced at the GTC Berlin conference that the company would provide capacity prepayments to three key partners—Japan's Ibiden, South Korea's Samsung Electro-Mechanics (Semco), and Taiwan's Unimicron—to support their expansion. The goal is to increase their substrate capacity to 2.5x current levels within three years and to meet demand as subsequent product platforms ramp up.
With Nvidia's financial support, Unimicron will expand its Guangfu facility in Hsinchu to increase KF substrate capacity. The new capacity is expected to begin ramping in the second quarter of 2027 and reach full utilization by mid-2028.
Unimicron's broader new-capacity plans in Taiwan are concentrated at its Guangfu and Yangmei facilities. The company's Hukou plant, acquired from Bridgestone in September 2026 for more than NT$10 billion (US$307.7 million), is also reportedly receiving strong backing from Nvidia.
Nvidia is also reportedly in talks over capacity partnerships with Austria's AT&S, Kinsus Tech, Nan Ya PCB, and Zhen Ding Technology Group subsidiary LIST. The discussions include support for investments in certain plants or equipment to secure a stable supply of high-end substrates for multiple product platforms.
Substrate makers spread risk
Substrate makers are receiving a succession of orders from major customers, including AMD, Intel, and Broadcom, which are willing to share expansion risks. Given the potential for continued volatility in AI demand, they cannot commit all their resources to a single customer. Spreading capacity across multiple customers helps them avoid idle capacity and investment-recovery pressure if market conditions reverse.
AMD extends its planning horizon
During her most recent visit to Taiwan in May, AMD CEO Lisa Su announced a supply-chain capacity expansion plan involving more than US$10 billion in investment in Taiwan. The plan's substrate and OSAT partners include Unimicron, Kinsus Tech, Nan Ya PCB, ASE, SPIL, and Powertech.
Su said demand for AI computing products such as CPUs and GPUs continues to accelerate. AMD is maintaining close cooperation with Taiwan's supply chain, including TSMC, OSAT providers, and substrate makers, while capacity discussions have expanded from the next one to two years to three to five years. The company could increase its investment further in the future.