The world’s most important chipmaker just told its shareholders something the tech industry already suspected but didn’t want to hear out loud.
At TSMC’s annual shareholder meeting, CEO C.C. Wei said the company will not be able to fully satisfy the rapidly growing global demand for AI chips for years to come — and that even expanding production in the United States won’t be enough to cover all customer needs. He called manufacturing capacity the single biggest bottleneck in global computing infrastructure right now. Hard to argue with that.
The numbers behind the shortage
Tech giants are expected to spend roughly $725 billion on AI development this year alone. That’s not a typo. And the chips to power all of it have to come from somewhere — overwhelmingly, from TSMC’s fabs in Taiwan.
Despite the crunch, Wei confirmed TSMC is forecasting sales growth of over 30% in 2026, and announced that employees will receive average bonus increases of more than 30% — a reflection of record profits driven by the AI boom. The company is making money faster than it can build factories, which is both impressive and the whole problem.
Arizona isn’t enough
As part of a trade agreement between the US and Taiwan, TSMC plans to build at least four additional factories in the US, on top of the six already planned — representing a total investment of $165 billion, with another $100 billion in capital still needed. Wei said two land parcels acquired in Arizona are intended to meet the company’s production needs for decades ahead.
That’s a massive commitment. It’s also a long-term one — fabs take years to build, qualify, and ramp up. The shortage isn’t going away this year, or next year, regardless of how many groundbreaking ceremonies get held in Phoenix.
Customers are already hedging
Companies that depend on TSMC are quietly trying to reduce that dependence. Apple — one of TSMC’s most important customers — has already approached Intel to manufacture some of its chips, specifically older and budget-tier components that don’t require the most advanced process nodes. It’s a sensible hedge, but it has a ceiling: Apple cannot fully walk away from TSMC, because its most advanced chips — the ones that go into iPhones and Macs — still depend on processes only TSMC can deliver at scale.
The same logic applies across the industry. NVIDIA, AMD, Apple, Qualcomm — they all need TSMC for the cutting-edge stuff, and there’s no realistic alternative at the frontier node. That’s not changing anytime soon.
What C.C. Wei said this week isn’t a warning so much as a status update: the AI buildout is real, the demand is outrunning supply, and the gap isn’t closing quickly. Every company betting big on AI infrastructure is, at some level, betting on TSMC’s ability to build faster than the world’s appetite grows. Right now, it can’t.