TSMC Just Lit the Fuse on the AI Gold Rush—Blowout Quarter, Fatter Margins, and a Mid-30s Growth Bombshell

Taiwan Semiconductor Manufacturing Co. (TSMC) delivered a classic “show me” quarter—and showed plenty. Revenue and profit hit fresh records on surging AI demand, management nudged full-year growth higher, and the margin line proved sturdier than feared despite FX and overseas ramp headwinds. Just as important, the company painted an operating backdrop where AI’s pull signal keeps strengthening while non-AI end markets are crawling off the bottom. In short: the world’s critical foundry sees a tight near-term supply/demand balance at the leading edge, a disciplined investment plan, and an ecosystem that increasingly depends on its capacity, packaging, and process roadmap.

On the scorecard, TSMC posted Q3 revenue of $33.1 billion (up 10% q/q; ~30% y/y in NT$ terms) versus the Street’s ~$32.1 billion, and EPS of $2.92 versus $2.63 expected. Gross margin rose 90 bps sequentially to 59.5% (well above prior guidance on better cost and utilization), and operating margin improved to 50.6%. Mix did the heavy lifting: advanced nodes (7nm and below) were 74% of wafer revenue; 3nm contributed 23%, 5nm 37%, and 7nm 14%. By platform, High-Performance Computing held at 57% of revenue, smartphones rebounded 19% q/q to 30%, while IoT and auto each landed at 5%. Guidance was similarly resilient: Q4 revenue of $32.2–$33.4 billion implies roughly down 1% q/q (vs. a previously implied down ~10%), up ~22% y/y at the midpoint, and gross margin guided to ~60%. Management lifted 2025 revenue growth to the “mid-30s” (from ~30%) and narrowed CapEx to $40–$42 billion (prior $38–$42 billion), with ~70% earmarked for leading-edge processes and 10–20% for advanced packaging and test. Needham promptly raised its TSM price target to $360 from $270, keeping Buy.

Growth drivers remain where investors want them: AI accelerators and premium smartphone/PC silicon built on N5/N3 today and N2 tomorrow. Management reiterated that AI demand has become “even stronger than three months ago,” underpinned by explosive token growth from consumer and enterprise usage and the rise of sovereign AI. The near-term constraint is not demand but capacity—both front-end wafers and back-end advanced packaging. CoWoS, the linchpin for high-end AI modules, stays tight into 2026; TSMC is expanding aggressively and coordinating earlier with customers and, crucially, customers’ customers to size capacity to real end-demand. Smartphones are no longer a drag: inventories look “healthy,” and Q3’s 19% sequential lift confirms a seasonal and product-cycle tailwind at the high end.

Headwinds are manageable but real. Overseas fabs still dilute gross margin (management now expects 1–2% full-year dilution in 2025 vs. 2–3% previously), currency remains a swing factor (every 1% move in USD/TWD moves GM by ~40 bps), and the initial N2 ramp will pressure 2026 margins even as N3 dilution fades. Policy risk sits in the background: tariffs and export rules complicate consumer-adjacent, price-sensitive segments and keep China AI demand fenced off. Even so, management's confidence in a mid-40s AI accelerator CAGR through 2029 held firm—despite limited China exposure—thanks to broad customer pipelines across GPUs and ASICs and continuous node migration that boosts compute per watt and per dollar.

The call color centered on three themes. First, capital discipline: TSMC won’t chase capacity for its own sake. Capital intensity should trend lower over time, and there’s “no plan” to add greenfield capacity for N5/N3; new buildings are aimed at N2 and beyond. Second, packaging and Foundry 2.0: advanced packaging revenue is nearing 10% of the company and is now treated as strategic, not ancillary. TSMC will build two advanced packaging fabs in Arizona and is partnering with a major OSAT that has already broken ground there—an ecosystem approach meant to shorten the supply chain for U.S. customers and de-risk logistics. Third, geography and roadmap: Arizona is accelerating (more land, pulling in N2); Japan’s first fab is in volume with good yields and a second underway; Germany’s specialty fab construction is progressing; Taiwan remains the center of gravity with multi-phase N2 capacity in Hsinchu and Kaohsiung. On process, N2 is on track for volume later this quarter with a faster 2026 ramp, N2P follows in 2H26, and A16 with Super Power Rail targets dense HPC power delivery—TSMC expects the N2 family to be “another large, long-lasting node.”

Analysts pressed on AI math, capacity, and margins. JPMorgan asked whether the previously cited mid-40s AI CAGR needs an upgrade; management said trends look “a little better” and numbers are “insane,” with a formal update likely in early next year. On CapEx pacing, the CFO reiterated spend will track opportunity, but revenue should continue to outgrow CapEx—don’t expect a sudden spike in any single year. Questions on CoWoS capacity and AI revenue mix drew tight-lipped but telling answers: everything related to AI front- and back-end remains “very tight,” and TSMC is “working very hard” to narrow the gap through 2026. UBS dug into 2026 gross margins and N2 dilution; the CFO reminded that N2’s structural profitability should ultimately exceed N3’s, even if early-ramp dilution is unavoidable. Goldman Sachs juxtaposed exponential token growth with TSMC’s ~40–45% AI CAGR; C.C. Wei’s answer was the practical bridge: node migration and architectural advances let each chip handle more tokens, so TSMC’s revenue CAGR doesn’t need to match token growth for silicon supply to keep up. Cowen asked GPU vs. ASIC mix; TSMC is indifferent—both ride leading nodes and both grow. Citi probed packaging in Arizona and OSAT coordination; management confirmed the co-location strategy to support U.S. demand sooner. Macquarie queried competition in a “Foundry 2.0” world; TSMC stressed full-stack value (front-end plus packaging) and, with a wink, noted a key U.S. “competitor” is also a good customer.

Read-throughs are straightforward. Nvidia benefits from sustained tightness and assured N5/N3/N2 progression plus a widening packaging funnel; scarcity and performance leadership support pricing power. Apple gets a cleaner supply picture for premium A- and M-series ramps on N3 today and N2/N2P later, with Arizona localization incrementally de-risking U.S. political optics. ASML’s EUV/High-NA narrative enjoys a tailwind from TSMC’s unwavering leading-edge roadmap. Advanced packaging names and OSATs—particularly those tied to CoWoS in the U.S. (e.g., Amkor)—stand to benefit from the Arizona cluster strategy. And for the broader AI supply chain (memory/HBM, substrates, test), the message is the same: capacity still chasing demand, with 2025–2026 the critical build window.

TSMC’s bottom line this quarter: structurally tight where it matters, fiscally disciplined where it counts, and still in the pole position for the AI infrastructure build-out. In a market prone to narratives, this one comes with receipts.