Recently, Nvidia and AMD have announced the start of production of a new generation of chips at TSMC's factory in Phoenix, Arizona. Nvidia's Blackwell AI chips and some of AMD's products will be the first advanced semiconductors to be manufactured in the United States. Executives from both companies mentioned the significance of localized production for supply chain resilience, but the actual industrial layout is more complex than it seems.
Nvidia's Blackwell AI chips are the first to enter production at this facility, and Nvidia CEO Jensen Huang noted that U.S. manufacturing will support demand for AI chips and supercomputers, further stabilizing the supply chain and improving its resilience. Similarly, AMD CEO Lisa Su also announced the production of its products at the plant on the same day, emphasizing the critical role of the collaboration with TSMC on the N2 process and the Fab 21 in Arizona to drive innovation.
Behind this series of actions is the uncertainty caused by Trump's announcement of imminent tariffs on semiconductors. Analysts generally believe that the company's choice to announce production plans at this time is not accidental, but to actively respond to tariff policies and stabilize their own earnings. If Trump's tariff policy is implemented, it will undoubtedly prompt more chip production to move to the United States. While the scope, permanence, and specifics of the current tariffs are unclear, it is certain that the expected policy has already had a ripple effect across the industry. For example, some fabless TSMC customers may consider moving some of their old or non-critical product lines to Intel's U.S. facility for production.
Let's focus on TSMC's Arizona plant itself. From the perspective of process technology, the current state-of-the-art process of this factory is N4, which is two generations behind TSMC's N2 process in Taiwan. In terms of capacity, the capacity of the advanced node of the Arizona plant is only about 20% of that in Taiwan. This means that the fab's capacity is likely to be stretched thin once multiple customers generate high demand for wafers. The production of advanced process chips is still highly dependent on Taiwan in the short term.
Figure: Nvidia and AMD have increased their investment in TSMC's Arizona factory
In the field of chip manufacturing, the competition between Intel and TSMC has never stopped. Intel is moving fully ahead with its 18A process in the US, which is comparable to TSMC's leading 2nm node. However, for now, Intel is lagging behind in the advancement of advanced processes. Its state-of-the-art Arizona plant doesn't appear to be fully capable of high-volume production, and construction of its Ohio plant has been delayed until 2030. This has also led to the fact that Nvidia and AMD have not yet listed Intel as their second choice. However, Intel is not without its advantages. In the field of advanced packaging, Intel's capabilities in the United States may be better than TSMC's. TSMC and Amkor plan to build a joint factory, which is expected to take another year or two to officially go online, and Intel's factory in New Mexico is already able to provide advanced packaging capacity for the country. At present, TSMC is responsible for most of the advanced packaging work of NVIDIA and AMD in Taiwan, and the wafers produced in the Arizona factory will most likely still need to be shipped back to Taiwan for packaging.
Looking to the future, Nvidia has a more ambitious plan. Over the next four years, NVIDIA plans to build up to $500 billion of AI infrastructure in the U.S. with Asian suppliers such as TSMC, Foxconn, Wistron, Amkor and ASE Group. In February, Apple also announced that it would invest more than $500 billion in the U.S. over the next four years to produce servers that power its AI services, and in January it became TSMC's largest customer at its Arizona facility. These measures will undoubtedly further promote the development and transformation of the U.S. chip industry.
