Taiwan Semiconductor Manufacturing (TSMC) posted record Q2 2026 revenue of $40.2 billion, up 34% year-over-year, and raised full-year guidance for the second time. The company now expects 2026 capital expenditure of $60–64 billion, up $4–12 billion from prior guidance, and raised revenue growth expectations to slightly above 40%, compared to previous guidance of 30%-plus. CEO C.C. Wei called AI demand 'extremely robust' and reiterated 'very high conviction in the multi-year AI megatrend.'
Advanced process nodes at 7nm and below accounted for 77% of wafer revenue. High-Performance Computing (AI accelerators) surged 20% sequentially and now represents 66% of total wafer revenue, displacing smartphones to 22%. TSMC's 2-nanometer node (N2) achieved first meaningful commercial revenue contribution of 3% in Q2, transitioning from engineering samples to paying production volume. The company also announced a $100 billion additional investment in Arizona fabs.
Wei cautioned that global AI chip supply will not meet demand for years and flagged rising prices as a margin risk. Despite the stronger outlook, TSMC shares fell ~5% in premarket trading as investors took profits amid concerns about margin compression and whether capex is keeping pace with demand or simply being consumed to stay relevant.
For architects: this is now a supply constraint confirmation from the primary chipmaker. TSMC's own exec is saying capacity additions will not satisfy demand. N2 ramping at 3% of wafer revenue tells you second-gen process nodes are still being absorbed as fast as they ship. If your foundation model training depends on next-gen GPU allocation, TSMC's 40%+ growth forecast means volume will exist, but at higher real costs and longer lead times. Spot capacity will tighten further.