DigiTimes整理台积电法说会：魏哲家称agentic AI带动CPU需求但未计入AI加速器增速，后段封装偏紧、与GF/VIS/Amkor/SPIL合作，均为定性表态无新增数字

Taiwan Semiconductor Manufacturing (TSMC) is seeing rising demand for CPUs related to agentic artificial intelligence (AI), but has not yet included CPUs in its calculations of AI accelerator demand growth. TSMC chairman C.C. Wei said at the company's earnings conference on October 15 that the company is maintaining its original, narrower definition of AI demand, mainly because it is difficult to clearly distinguish whether CPUs are used in traditional servers or AI servers. Actual demand also depends on how customers deploy the technology and configure their products.

Smartphones, AI, and HPC ramp together
Rising demand for AI and high-performance computing (HPC) is also creating a demand mix that differs from the past for TSMC's production ramp-up strategy for new process technologies.
Demand for smartphones, AI, and HPC applications is currently strong at the same time, Wei said. Chiplets and the large dies used in AI accelerators could both adopt new process technologies simultaneously. TSMC is prepared for this scenario, and its current yield learning and mass-production execution are performing reasonably well, with customers also satisfied with the results.
Power supply and BSPDN adoption
Addressing market concerns over whether power supply could constrain demand for AI chips and drive greater adoption of backside power delivery network (BSPDN) technology, Wei said power availability is one of the factors TSMC considers when evaluating capacity and demand. The company has conducted further checks on power availability in the US, and its assessment so far has been fairly good.
Whether customers adopt BSPDN technology will depend on product requirements. Some AI accelerators may use it, while products such as smartphones may not need it.
Pricing varies by segment and location
Regarding foundry pricing, reports have suggested that TSMC plans to raise prices for both mature and advanced process technologies. Asked whether the increases would apply to existing contracts or primarily target additional capacity beyond customers' original commitments, Wei said TSMC's pricing strategy varies according to factors such as market segment, geographic location, and process technology.
For example, manufacturing costs differ depending on whether customers choose production in the US or Taiwan. TSMC hopes the market will recognize the value it creates and aims to secure the maximum value its technology can generate for shareholders.
Partners shore up back-end capacity
Beyond wafer manufacturing, back-end advanced packaging capacity has become an important part of TSMC's efforts to support customers. Commenting on the announcement by GlobalFoundries (GF) that it will supply silicon interposers for TSMC, Wei said the two companies have established a cooperation arrangement as part of TSMC's overall back-end supply strategy.
Wei said TSMC's top priority is helping customers succeed in the market and ensuring they receive comprehensive supply chain support, covering both front-end wafer manufacturing and back-end packaging.
Back-end supply is currently somewhat tight. TSMC hopes to work with partners such as GF to provide the capacity support customers need and help products reach the market smoothly. In addition to GF, TSMC has established back-end capacity cooperation with Vanguard International Semiconductor (VIS), Amkor Technology, and Siliconware Precision Industries (SPIL).