SK Hynix targets January 2027 for HBF standard, eyes DRAM stacked on AI chips

SK Hynix's HBF play isn't a product launch—it's a structural decoupling of AI inference from HBM's pricing monopoly. By inserting a flash-based bandwidth tier between HBM and SSD, they're weaponizing NAND's non-volatility and cost curve against the DRAM oligopoly. The chain reaction is asymmetric: Kioxia and SanDisk gain their first credible entry into AI's core compute stack, while TSMC's CoWoS "scarcity premium" erodes if DRAM-on-die becomes viable. This hits the packaging layer in Taiwan, China directly. Geopolitical risk is embedded in the standard itself. If Washington folds HBF into export controls, SK Hynix's Dalian NAND line faces the same compliance severance as HBM. If Seoul leads spec governance, BIS will likely intervene under "technology security"—a replay of the 2022 EUV playbook. Competitive positioning: Samsung's dual NAND+DRAM portfolio makes it the natural HBF player, but it will likely accelerate HBM4 to defend margin. Micron is most exposed—HBM is its AI valuation anchor, and HBF success directly attacks that narrative. NVIDIA will almost certainly dual-source in post-Blackwell architectures. The 12-24 month inflection: by H2 2026, prototype validation shifts the fight from bandwidth specs to "who defines inference memory allocation." The real killer feature isn't speed—it's near-zero standby power. Under data center PUE constraints, HBF's non-volatile nature delivers a TCO advantage HBM structurally cannot match. That's SK Hynix's actual endgame.