TSMC cannot keep up with AI chip demand even as it pours $165 billion into US factories, CEO C.C. Wei told shareholders on June 4, 2026. The world's largest contract chipmaker is building three additional plants in the US on top of its existing Arizona facility, plus two advanced packaging sites and a research and development center, and it still expects to be a bottleneck for American customers. Wei said it could take a "very long time" to fulfill customer needs from US-based production.
Taiwan Semiconductor Manufacturing Co. fabricates the leading-edge silicon behind virtually every frontier AI accelerator, from Nvidia's data-center GPUs to custom chips designed by Google, Amazon, and Microsoft. With AI training and inference workloads compounding faster than fab capacity can come online, TSMC has become the single most constrained link in the global AI supply chain. Deloitte projects the semiconductor industry will grow into a $1 trillion market by 2027, with AI silicon driving the bulk of the expansion.
“Customer demand is so high, and we can only support so much.”— C.C. Wei, TSMC CEO
The Arizona buildout was meant to ease that pressure. TSMC's first US fab opened there, and the additional $165 billion package — one of the largest single foreign manufacturing investments in American history — funds three more plants alongside the packaging and R&D sites. Advanced packaging, in particular CoWoS, has been the choke point for Nvidia's Hopper and Blackwell shipments, and TSMC has been racing to expand that capacity at multiple sites.
Key facts
- 01TSMC CEO C.C. Wei told shareholders on June 4, 2026 that customer demand has outstripped what the company can supply.
- 02TSMC is investing $165 billion in the US to build three additional plants, two advanced packaging facilities, and a research and development center.
- 03Deloitte projects the semiconductor industry will hit $1 trillion by 2027 on the back of AI demand.
- 04Wei said it could take a 'very long time' to fulfill customer needs from US-based production.
- 05The AI boom has already triggered RAM and NAND Flash shortages expected to persist for years.
Even with that spending, Wei was blunt that supply will trail demand for the foreseeable future. He framed it as a matter of physics and lead times rather than strategy: fabs take years to build, qualify, and ramp, and the latest process nodes require equipment and skilled labor that cannot be conjured overnight. US-based production specifically, he indicated, will lag Taiwan-based output for an extended period.
Pricing is the other lever. Wei said he would "like" to raise prices but added that TSMC would not impose an abrupt increase of the kind the memory market has already seen. DRAM and NAND Flash prices have spiked as AI servers consume HBM and high-capacity SSDs in volumes that have squeezed consumer supply, and that shortage is expected to last for years. Wei appears intent on avoiding a comparable shock for logic chips, where TSMC's pricing power is even more concentrated.
“We are doing our best to ensure TSMC does not become a bottleneck.”— C.C. Wei, TSMC CEO
The restraint is notable given TSMC's position. The company has effectively no peer at the 3nm and 2nm nodes that frontier AI chips require — Samsung Foundry and Intel Foundry are still working to close the gap — which means TSMC could raise prices sharply with minimal customer defection. Holding the line preserves long-term relationships with hyperscalers and chip designers who are already negotiating multi-year wafer commitments.
For customers, the message is uncomfortable. Nvidia, AMD, Apple, Qualcomm, MediaTek, and the in-house silicon teams at every major US cloud provider are all competing for the same wafers and the same packaging slots. Allocation, not list price, is the binding constraint. Companies that locked in capacity early are shipping; companies that did not are waiting.
The Arizona project also carries political weight. The $165 billion commitment was structured to align with US industrial policy and CHIPS Act incentives, and it gives American AI infrastructure a domestic source of leading-edge logic for the first time in decades. But Wei's comments make clear that on-shoring will not, on its own, solve the AI shortage — Taiwan will remain the center of gravity for advanced node production well into the next decade.
There are real questions about how long the demand curve holds. If AI training compute requirements plateau, or if model providers shift more aggressively to inference-optimized chips and custom silicon, TSMC's order book could soften before the new US fabs ramp. Wei did not address that scenario directly, and the company has been careful in past cycles to avoid over-committing to capacity that could go underutilized in a downturn.
TSMC's constraint is the AI industry's constraint. Every shipping delay at a 3nm fab translates into delayed accelerator launches, delayed data-center buildouts, and delayed product roadmaps at companies whose valuations assume continued exponential compute growth. The $165 billion US investment buys long-term resilience and political cover, but in the near term the math is unchanged — a single foundry sets the pace for the entire AI hardware stack, and that foundry is telling its customers, on the record, that it cannot keep up.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
