AI chips are pushing everything else off TSMC's most advanced production lines

The Decoder
AI Hardware AI Business

In 2026, nearly every major AI accelerator family is moving to TSMC's N3 manufacturing process at the same time - Nvidia's Rubin, Google's TPU v7/v8, Amazon's Trainium3, and AMD's MI350X. The resulting demand shock is hitting a foundry that simply can't expand fast enough. According to SemiAnalysis, TSMC has been caught flat-footed, with wafer capacity expansion falling well behind surging AI demand.