TSMC investor conference: TSMC lifts 2026 revenue growth forecast to 44%, 2nm expansion, new 3nm fab

The 44% growth signal is not an order confirmation; it is TSMC asserting pricing power as AI capex enters a structural lock-in phase. On the technology stack, 2nm GAA mass production will compress ASML's High-NA EUV delivery cycle to under 12 months, while the CoWoS bottleneck pivots from chip scarcity to packaging scarcity. Once HBM4 exceeds 16-layer stacking, TSV yield becomes a more lethal constraint than lithography itself. On compliance, new 3nm fabs in Taiwan, China remain tethered to US export-control approval chains. The 2023 Dutch DUV restriction case proved that a 6-to-9-month approval delay can erode single-line ROI by 8 to 12 percentage points. Competitively, if Samsung's 2nm GAA yield ramp continues to lag, its foundry share will fall below 5% by 2026. Intel's 18A Foveros route is architecturally differentiated, yet HPC customer validation cycles mean it cannot capture mainstream workloads before 2027. The 18-month long-tail: the real variable is AI inference chips siphoning good-enough 3nm and 4nm capacity, systematically diluting 2nm's marginal premium. At least 15 of the 44 growth points originate from inference rather than training, signaling that TSMC's profit structure is migrating from process scarcity toward system-level customer lock-in.