TSMC's $100 billion Arizona bet rides AI chip demand

Taiwan Semiconductor Manufacturing will pour another $100 billion into its Arizona operations, lifting total US spending to $265 billion, as demand for AI accelerators keeps its advanced packaging capacity sold out and pushes revenue growth above 40 percent this year.

"AI chip demand will stay strong through at least 2029 to 2030, and it could last longer because a new industry is being created," Chief Executive C.C. Wei said on the company's second-quarter earnings call.

The world's largest contract chipmaker reported Q1 revenue of $35.9 billion, with net income up 58 percent to NT$572.48 billion and gross margin at 66.2 percent. High-performance computing, driven by AI accelerators, accounted for 61 percent of revenue, while chips built on 7-nanometer processes and below made up 74 percent of wafer revenue. Management raised full-year 2026 revenue growth to more than 30 percent and guided second-quarter revenue to $39.0-$40.2 billion.

The Arizona expansion reduces the single-point-of-failure risk of concentrating production in Taiwan, where geopolitical tensions with China have long weighed on the stock. TSMC trades at 23.5 times forward earnings, and management expects AI-related revenue to compound at a high-50s percent rate through 2029, with 2-nanometer risk production beginning in the second half of 2026.

CoWoS capacity sets the ceiling on AI GPU shipments

CoWoS, or chip-on-wafer-on-substrate, is the packaging method that bonds high-bandwidth memory directly onto GPU dies, the architecture behind Nvidia's H100 and Blackwell accelerators. Because TSMC's CoWoS capacity is sold out well into the future, the company effectively sets the ceiling on how many AI GPUs Nvidia can ship. Management is accelerating expansion of that capacity alongside the Arizona buildout.

The 2-nanometer node (which packs more transistors per square millimeter, improving performance per watt) begins risk production in the second half of 2026 and reaches high volume in 2027. CFO Wendell Huang said the ramp will be supported by "continued strong demand for our leading-edge process technologies." Nvidia, AMD, Apple, Broadcom, and Marvell, plus the hyperscalers' custom silicon programs, are all expected to migrate designs to 2nm.

The recent sell-off in AI chip stocks — Micron down about 32 percent in three weeks, Broadcom roughly 24 percent below its 52-week high, Nvidia about 12 percent off its peak — reflects fears that custom silicon from DeepSeek, OpenAI, and the cloud giants will erode Nvidia's pricing power. TSMC sidesteps that argument: those custom chips still need leading-edge manufacturing, and TSMC's capacity is overwhelmingly where they get built. The company manufactured 12,682 products for 534 customers in 2025.

At roughly 30 times trailing earnings, TSMC trades in line with Nvidia for a business whose profit growth is accelerating — net income rose 77.4 percent year over year in Q2, up from 58.3 percent in Q1 and 35 percent in Q4 2025. The main risks are the older ones: most production still sits in Taiwan, and the semiconductor industry remains cyclical. Management raised 2026 capital spending to $60-$64 billion, at least $4 billion above its prior forecast, which could pressure margins over time.

This article is for informational purposes only and does not constitute investment advice.