Samsung reportedly locks up to 80% of 2027 memory supply into long-term deals

Samsung committing roughly 80% of its 2027 memory output, including HBM, to long-term agreements with Nvidia, Google, and Microsoft is not a pricing tactic. It is a structural realignment of the AI compute stack. Memory is transitioning from a cyclical commodity to a strategic input, and Samsung is replicating the TSMC-Apple co-design lock-in at the DRAM layer. The technical implication is severe: Pyeongtaek P4 and P5 fabs will be architecturally dedicated to HBM4 stacking, systematically starving the DDR5 and LPDDR5X commodity pool. Nvidia's Rubin roadmap is now reverse-engineering Samsung's process roadmap, a supplier-customer inversion that deepens every quarter. For SK Hynix the calculus is brutal. Either M15X in Cheongju scales aggressively to capture residual HBM share, or the company pivots to base-die customization as a moat. Micron is harvesting the friend-shoring premium on US sovereign AI workloads. AMD, without a guaranteed HBM allocation, faces structural weakness across its MI400 cycle. Within 18 months expect HBM pricing to bifurcate into LTA-locked contracts carrying a 30 to 50 percent premium and a thin open market. The memory oligopoly is shifting from capacity competition to allocation politics. The question is no longer who makes the most wafers but who is on whose list.