AMD CEO sees chips from TSMC's US plant costing 5%-20% more
Cost Premium and Product Impact
- Many commenters are surprised the US-fab premium is only 5–20%, calling it a “bargain” for leading-edge nodes.
- Rough back-of-envelope math suggests that for a ~$350 Taiwan-made GPU die, a 20% fab cost increase adds ~$70, which is only a few percent of a $2,000 retail card.
- Others note margins differ: AI GPUs and high-end GPUs can absorb this easily; commodity CPUs and lower-margin parts feel it more.
- There’s debate over how much manufacturing cost matters versus total development and BOM (HBM, packaging, etc.).
Resilience, Security, and Geopolitics
- Strong support for paying a moderate premium to reduce dependence on Taiwan given invasion/earthquake risk and China–US tensions.
- Some argue that even maintaining 5–10% of global leading-edge capacity in the US is strategically huge: it preserves skills, equipment, and a base to scale in crisis.
- Others counter that for non-US buyers, Taiwan or “anywhere-but-US” is safer given US sanctions, export controls, and tariff volatility.
- A minority worries that more US capacity could reduce US willingness to defend Taiwan; others argue the opposite: the Arizona fab increases Taiwan’s leverage.
Partial Onshoring and Supply-Chain Gaps
- Several point out that wafers from Arizona are often still packaged in Asia; however US-based packaging (Amkor Arizona, Intel NM, others) is ramping.
- Boards, passives, specialty films, and many “low-profit, high-barrier” components remain overwhelmingly Asian; without these, US fabs don’t fully solve dependency.
- Shipping cost and carbon from moving dies back and forth is argued to be minor versus fab energy use; leading-edge chips are often air-freighted anyway.
Tariffs, Subsidies, and Who Pays
- One camp sees this as validation that modest tariffs plus CHIPS-style subsidies can level costs without massive consumer pain.
- Another stresses that any 10–20% cost rise, once passed through layers of the stack and compounded by tariffs, is effectively a large indirect tax on consumers.
- Some insist businesses, not end-users, should bear more of the cost (lower margins, government subsidies), while others argue customers ultimately pay either via prices or taxes.
Intel, TSMC, and Long-Term Industry Structure
- Long side-thread debates whether Intel is “cooked” or can recover; many blame decades of financialization, mismanagement, and fab delays.
- Consensus that TSMC currently dominates state-of-the-art; Samsung is a distant second; Intel’s future as a competitive foundry is uncertain.
- Several note that long term, leading-edge fab economics may support only one or a very small number of global players, making geographic diversification politically but not economically natural.