TSMC Flags Tight AI Capacity, Eyes Price Moves

Taiwan Semiconductor Manufacturing Co. told investors in early June that demand from artificial‑intelligence customers remains robust and that capacity will stay tight for the near term, a signal that shortages could persist across the AI hardware chain.

TSMC Chairman and CEO C.C. Wei said at the company’s June 4 shareholders meeting that “it will be a long time before we can meet customer demand,” and the company is working to expand output while balancing customer relations.

TSMC’s own annual meeting documents describe 2025 and early 2026 as years of “robust AI‑related demand,” and lay out ramp schedules for N2 family technologies and advanced packaging as part of the company’s response.

Executives highlighted that the tightest bottlenecks are not only at the front‑end wafer stage but also in advanced packaging and specialized processes such as CoWoS, CoPoS and 2.5D integration, where throughput and substrate supply are strained.

Industry researchers say those process constraints have prompted pricing talk inside the foundry market: a TrendForce report published in late May flagged the possibility of a mid‑2026 3nm price increase of up to 15% with further adjustments into 2027.

Independent coverage and trade reporting indicate TSMC is openly weighing price adjustments to help cover higher costs and to manage allocation across customers, though the company told shareholders it will try to avoid abrupt shocks.

The squeeze is already rippling to the assembly and testing supply chain. Outsourced semiconductor assembly and test firms and substrate makers have seen demand spike, and analysts expect some customers to shift orders to OSAT partners as TSMC’s internal packaging headcount and lines fill up.

Some customers with the biggest AI workloads have locked in disproportionate shares of advanced packaging capacity, leaving less room for smaller buyers and specialty designs, a dynamic that tightens pricing leverage for capacity holders.

TSMC is accelerating capital spending and adding fabs and packaging assets abroad, including planned expansions in Arizona, Europe and Japan, but those lines take years to reach high‑volume output, meaning near‑term tightness is likely to persist into 2026 and beyond.

The market is already pricing in the scenario. TSMC’s recent revenue and guidance reflected an AI buildout while analysts flagged higher capex and margin impacts tied to overseas builds and rising component costs. Investors and customers are watching whether formal price moves materialize later in 2026.

If TSMC raises wafer or packaging prices, the effect could be felt up and down the AI hardware stack: GPU and ASIC makers may see higher unit costs, cloud operators could face steeper refresh bills, and OSAT and substrate suppliers may pass on their own increases. That would reshape procurement and design timelines for AI infrastructure buyers.

For now, TSMC’s message to investors and customers is twofold: AI demand is structural and durable, but capacity additions take time, so allocation and selective price measures are among the tools the company is considering to balance supply, profitability and long‑term relationships. Watch formal pricing notices, CoWoS/CoPoS ramp announcements and OSAT contract flows for the clearest signals.