AI Orders Ignite TSMC: Q3 Net Profit Up 69%, 2nm Contributes 8% in Debut Quarter

On October 15, TSMC delivered its Q3 2026 results: net profit of NT$763.1 billion, up 69% year-on-year, with revenue up roughly 50% — the sixth consecutive quarter of record earnings. Driven by AI compute demand, the “shovel-seller of the AI gold rush” is spinning faster than the market expected.

AI Demand Is the Only Storyline

The driver is stated plainly across coverage: artificial intelligence. From cloud training clusters to edge inference devices, the buyer list for advanced-node wafers keeps growing. Demand-side evidence is dense — Lenovo recently announced global delivery of NVIDIA Vera Rubin NVL72 racks as “AI cloud super factories” accelerate, and every such factory is a bid for CoWoS advanced packaging and 2nm capacity.

2nm Ramp: 8% in Its Debut Quarter

The real suspense sits at the leading edge. The 2nm process entered volume production in Q3 and immediately contributed about 8% of revenue — a fast debut. On October 18, Cailian Press reported a BofA note projecting 2nm's revenue contribution to rise to 11% next year, with capacity targets likely met ahead of schedule. With major customers locking in capacity, tight supply at advanced nodes is the base case through 2027.

Certainty Down the Chain

For the supply chain, TSMC's report is a rare high-certainty signal: full leading-edge loading implies upward revisions for CoWoS packaging, HBM memory and high-speed interconnects. Downstream, it means the unit-cost curve of inference keeps bending down — denser chips, lower power, more tokens per watt.

Every leading-edge ramp eventually shows up as a falling inference-cost curve. For teams planning long-horizon compute foundations — like Jiutian's Tiansuan platform — a foundry's capacity signal is next year's cost signal. Treating process-node cadence as a leading indicator beats chasing single-card specs.

(Compiled from Wallstreetcn, Cailian Press, Sina Finance, Guandian and other public reports)