TSMC Posts Record Quarter, Lifts Capex as AI Chip Demand Surges

Taiwan Semiconductor Manufacturing Company (TSMC) posted one of the strongest quarters in its history and raised its 2026 capital expenditure outlook as AI chip demand continues to outrun supply.

Second-quarter revenue reached roughly $40 billion, with net profit jumping about 77% year-over-year. High-performance computing now accounts for roughly two-thirds of revenue, roughly three times the smartphone share, underscoring how thoroughly AI accelerators have rewritten the foundry mix.

Capex and capacity go vertical

TSMC lifted 2026 capital spending guidance to $60 to 64 billion, up from a prior $52 to 56 billion range. Industry trackers say 3nm wafer output is on track to hit about 180,000 units a month in early Q4 2026, months ahead of the original plan, driven by strong orders from Nvidia, AMD, and Broadcom.

Construction of the company’s 1.4nm (A14) Fab 25 in Taichung is also ahead of schedule. The first building is expected before April 2027, with pilot production possible in 2027 and mass production targeted for mid-2028. Packaging remains a bottleneck: CoWoS capacity is scaling rapidly but still tight for non-anchor customers.

Geopolitics and diversification

Outside Taiwan, TSMC continues expanding its global footprint, with Japan already in production, Arizona ramping, and Dresden scheduled for late-2027 mass production aimed at automotive and industrial customers. The dual pressure of AI demand and geopolitical risk is forcing the company to build redundancy without surrendering process leadership.

Intel Foundry is positioning EMIB packaging and its 14A roadmap as an alternative for hyperscalers seeking relief from TSMC constraints, but TSMC’s lead at leading-edge nodes remains the industry’s central fact.

Decoded Take

TSMC’s quarter is not just a Taiwan earnings story, it is a map of the AI economy. When HPC is two-thirds of foundry revenue and capex jumps by nearly $10 billion in one guidance revision, the bottleneck has moved from “will AI demand persist?” to “who can manufacture and package fast enough?”

That reframes competitive strategy across the stack. GPU vendors, custom silicon teams, and hyperscalers are now competing as much for capacity allocation as for model quality. Packaging, not just transistor density, is the new scarce resource. Watch CoWoS and next-gen panel packaging timelines as closely as node roadmaps, and watch whether Intel or Samsung can convert “alternative foundry” talk into meaningful share before A14 locks in another multi-year lead.