Founded by Morris Chang in 1987, TSMC pioneered the pure-play foundry model and is the world's largest and most advanced semiconductor foundry. In 2025 it held ~70% of the global foundry market, controlling over 90% of sub-7nm advanced-process capacity and over 95% of AI-accelerator chip manufacturing, with customers including NVIDIA, Apple, AMD, and Qualcomm. It builds a dual "process + packaging" moat with its 2nm process and CoWoS advanced packaging, the core "shovel-seller" of AI compute infrastructure.
Current core thesis
TSMC is one of the highest-certainty investments of the AI era, but the certainty has a time boundary (2028 is the key inflection). 2026Q1 gross margin was 66.2%, net margin 50.5%, and advanced process 74% of wafer revenue, in a "super window" overlapping AI demand exploding + 2nm effective monopoly + CoWoS sold out through 2027: 2nm wafer ASP >$30K, capacity only 1/3 of demand, Samsung 3nm a yield disaster, Intel 18A external revenue only ~$300M — unprecedented pricing power. A forward P/E of ~27x implies a ~3-5pp geopolitical discount (PEG ~0.6). Core risks: US-fab ramp dilutes gross margin 2-4pp, Samsung/Intel catch up in 2028, and Taiwan-Strait tail risk. Strategy: overweight in 2026-2027 (15-25% of a tech portfolio) to enjoy the certainty, adjust dynamically from 2028 based on the competitive landscape, with gross-margin trends and NVIDIA orders as core exit signals.
Forward catalysts
High-impact catalyst: verify whether gross margin holds the 65.5-67.5% guidance, HPC growth continuity, and 2nm ramp progress.
N2P (enhanced 2nm, target yield 80%) is expected in H2 2026; CoWoS capacity targets 120-130K wafers/month by end-2026 (from 35K at end-2024, ~4x), still sold out through 2027.
Rubin R100 uses N3P, expected to sample in Q4 2026 and reach production in Q1 2027; NVIDIA takes 36% of TSMC's 2026 3nm capacity (just 5% in 2025).
A16 uses Super Power Rail (SPR) backside power, +8-10% speed at the same voltage or -15-20% power at the same speed vs N2P, targeting HPC/AI; NVIDIA's Feynman GPU is confirmed to adopt it.
The CFO guides overseas-fab ramp to dilute gross margin 2-3pp/yr in 2025-2027 and 3-4pp after 2028; with 2nm-ramp dilution, total pressure could reach 5-7pp after 2027.
A14 (1.4nm, 2nd-gen GAA) reaches production in 2028 but at peak capital intensity; meanwhile Samsung 2nm may mature, Intel 14A may reach production, CoWoS may be in surplus, and the US fab's share rises to 15-20% — a re-evaluation point where multiple factors converge.
Occurred
Revenue NT$1.134T (+35.1% YoY, +8.4% QoQ), gross margin 66.2% (170bps beat), net margin 50.5%, EPS NT$22.08 (5.7% beat); an eighth straight quarter of double-digit profit growth.
Jensen Huang publicly confirmed NVIDIA is now TSMC's "largest customer," surpassing Apple in 2025 at ~19% of TSMC revenue; Apple fell from 25% to 17%.
TrendForce reported TSMC raised prices 5-10% on all sub-5nm processes from January 2026; customers accept the premium for lack of alternative suppliers — a pricing-power pass-through rarely seen in industry history.
Revenue NT$3.81T (+31.6%), net income NT$1.70T (+46.4%), gross margin 59.9%, operating margin 50.8%; free cash flow topped NT$1T for the first time at NT$1.003T (+15.2%).
N2 entered production in Q4 2025 at Hsinchu Fab 20 and Kaohsiung Fab 22, the first GAA nanosheet node, with SRAM yield >90% and logic density 313 MTr/mm² (the industry's highest).
The US Commerce Department revoked the Nanjing fab's VEU (Validated End-User) license, confirming the substantive threat of an export-control escalation path.
US investment was raised from $12B in 2020 to $165B, the largest single foreign direct investment in US history; Fab 1's 4nm is in production but at only ~8% gross margin.
Buffett cut TSMC ADRs ~60% on Taiwan-Strait geopolitical risk (not fundamentals), a classic case of the geopolitical risk premium being explicitly priced into institutional positioning.