TSMC warns AI chip shortage to persist into 2027; signals 15% 3nm price increase H2 2026

TSMC CEO C.C. Wei told shareholders at the company's June 4 annual meeting that the AI chip shortage will persist for years, and the foundry is already raising prices on its most advanced nodes. While avoiding the sudden memory-market style spikes, TSMC is executing measured, negotiated double-digit price increases: reports cite 3% to 10% increases across advanced nodes for 2026, with a further 15% hike on 3nm slated for the second half of the year. The company posted record gross margins around 66% in Q1 2026, confirming that pricing power is translating directly to profitability.

The binding constraint on AI chip supply is no longer wafer fabrication—it is advanced packaging. CoWoS (Chip-on-Wafer-on-Substrate), the 2.5D packaging process that bonds logic dies with high-bandwidth memory stacks, is the bottleneck. TSMC's CoWoS capacity cannot keep pace with demand from accelerator makers like NVIDIA, AMD, and Broadcom, meaning lead times for AI-related packaging are expected to stretch into 2027. This constraint explains why supply-side relief is not expected before 2027 at the earliest, when expanded capacity for advanced nodes and packaging is forecast to ease pressure.

Architects and infrastructure teams should budget for sustained HBM costs through 2027 and expect NVIDIA H200/B200 and AMD MI300X/MI355X component costs to increase by 10%–15% in coming quarters. The shift in TSMC's pricing strategy from flat to steady incremental hikes means that devices built on cutting-edge silicon (data center GPUs, custom AI accelerators, inference chips) will see cost growth absorbed downstream into cloud pricing. For organizations locking capacity via long-term agreements, securing allocations through 2027 is now the primary strategy.