Nvidia and Intel Expand Chip Production in the U.S., but Critical Stages Remain Overseas

Nvidia and Intel are actively advancing semiconductor production in the U.S., highlighting significant progress in localizing key stages. Nvidia announced a partner network spanning 43 states and plans to invest up to $500 billion in U.S. AI infrastructure over the next four years. Its partners include TSMC, Foxconn, Wistron, Corning, Coherent, and Amkor. Intel, meanwhile, showcased its full-cycle chip production capabilities, including design, manufacturing, and assembly within the U.S.

Despite this progress, both manufacturers face a critical gap: even chips produced at TSMC’s Arizona facilities are shipped to Taiwan for final packaging and HBM memory integration. The U.S. currently lacks the capacity for high-performance HBM production and packaging, with these processes concentrated in South Korea and Taiwan. This creates dependence on foreign partners and increases logistical risks.

To address this, major projects are underway to build new facilities. For example, Amkor is constructing a $7 billion campus in Arizona, set to begin operations no earlier than 2028. SK hynix is investing $3.87 billion in an Indiana plant for HBM4E and HBM5 production, while TSMC plans to launch its own chip packaging facilities in Arizona by 2029. Until then, reliance on overseas capacity will persist, and even next-generation chips like Rubin will undergo final assembly outside the U.S.

These initiatives are supported by government programs, including tax incentives under the CHIPS Act. However, project timelines and high costs create uncertainty. For now, most of the added value in AI infrastructure production comes from system assembly in the U.S., while critical components remain dependent on foreign supply chains.