TSMC to build another 3nm fab in Tainan - Evertiq

TSMC's additional 3nm fab in Tainan, Taiwan, China is less about capacity expansion and more about hedging N2 yield-ramp risk. GAA architectures typically land at 60-70% initial yield, and no hyperscaler will commit full AI/HPC volume to an unproven node. This second 3nm line locks in transitional demand for Apple A19, AMD Zen 6, and NVIDIA's next-gen GPU. Upstream, the ripple is immediate: ASML's EUV backlog thickens further, SUMCO and Shin-Etsu 12-inch wafer utilization stays above 95%, and Tainan's power grid—already near saturation at roughly 200 GWh annual draw per 3nm fab—becomes the binding constraint. On competition, Samsung's SF3 pricing aggression is being neutralized by TSMC's 18-24 month capacity lock-in contracts. When customers hold long-term allocation, the cost-performance narrative loses its leverage. Intel's 18A foundry business remains in customer qualification through 2025 and cannot divert advanced-node volume in the near term. The 12-24 month tail: 3nm transitions from frontier to volume workhorse, mirroring 5nm's 2020 role. Profit centers shift to N2 and A16. The Tainan fab's real value isn't the 3nm node itself—it's the 12-18 month strategic buffer it buys the entire advanced-node pipeline.