TSMC Lifts Capex Toward $64 Billion as Foundry Rivals Fall Further Behind

TSMC raised 2026 capital spending to as much as $64 billion as Q2 foundry share reached 72.5% and advanced AI chip capacity stays tight.

TSMC is pouring record sums into manufacturing capacity to defend a lead that keeps widening. The Taiwanese contract chipmaker now expects capital expenditure of as much as $64 billion this year, a range revised upward from an earlier plan of $52 billion to $56 billion, as demand for cutting-edge silicon outruns anything its rivals can supply.

The scale of that dominance shows up in the numbers. Industry estimates put TSMC's global foundry market share at 72.5% in the second quarter of 2026, with quarterly revenue climbing 12.1% from the prior three months to nearly $40.2 billion. Samsung Foundry, its closest competitor, ceded ground again over the same period.

AI Demand Keeps the Most Advanced Lines Sold Out

The boom rests on insatiable appetite for high-performance chips tied to artificial intelligence. Data-center graphics processors and specialized accelerators alike are keeping the most sophisticated production lines running at full tilt, and supply is nowhere near matching orders.

TSMC chairman C.C. Wei told Intel chief Lip-Bu Tan that capacity is barely sufficient to satisfy every customer request. Analysts at Piper Sandler reckon the company's manufacturing slots are effectively sold out through 2028. Even Intel, a would-be competitor, continues to rely on TSMC for core components and calls its services the industry standard.

Leading-edge bookings stretch years into the future. Nvidia is shifting its Rubin graphics processors to 3-nanometer production, Google is building its Tensor chips on the same node, and specialized processors from Amazon, Microsoft and Arm are adding further pressure to advanced wafer capacity. Pure-play foundry work accounts for 73% of TSMC's business, rising above 90% for leading-edge AI and high-performance computing chips. Advanced nodes already generated 77% of total wafer revenue in the most recent quarter.

That tightness has major customers exploring alternatives. Apple has reportedly held preliminary talks with Intel and Samsung about future U.S.-based manufacturing options, though no firm orders have been placed with rivals — a technical switch would take years to prepare. Market watchers read such evaluations less as a verdict on TSMC than as a symptom of its capacity crunch. Intel, for its part, aims to close the gap with a next-generation process targeting performance near TSMC's upcoming nodes.

Seven U.S. Approvals, $265 Billion Arizona Plan

Geopolitical hedging is running alongside the capacity build-out. Taiwan's Ministry of Economic Affairs has approved seven U.S. investment applications totaling $44 billion since late 2020. In Arizona, TSMC's total planned investment reaches as much as $265 billion. At the same time, the company is expanding 13 plants on its home island.

The stock has tracked the operational momentum. Shares closed Friday at EUR 396.50, up 54% since the start of the year and just 5.7% below their record high of EUR 420.50.

2nm in Production, A14 Targeted for 2028

Management is advancing its technology roadmap on schedule. The 2-nanometer class has been in commercial production since the fourth quarter of 2025. By 2028, TSMC aims to start manufacturing the next-generation A14 node, which is designed to run 10% to 15% faster at unchanged power consumption.

On lithography, the company is taking a measured path. While rivals test early High-NA EUV tools, TSMC does not plan broad deployment of the technology in mass logic production until 2030. Executives intend to stick with established process steps through the end of the decade to protect yields and keep depreciation costs on new equipment in check.

Investors get their next read on the boom in October, when TSMC holds its regular third-quarter earnings conference. Beyond the year-end outlook, attention will center on progress in 2-nanometer mass production and the capital budgets of the major hyperscalers.