SK Hynix cranks up the HBM4 assembly line to prep for next-gen GPUs (www.theregister.com)

SK Hynix announced it has completed development of HBM4 and is prepping high-volume production, a move timed to supply next‑generation datacenter GPUs from Nvidia and AMD slated for 2026. Technically, SK Hynix says it doubled the number of I/O terminals to 2,048 versus HBM3e, effectively doubling per-stack bandwidth and boosting energy efficiency by more than 40%. The company also reports exceeding JEDEC HBM4 targets with a 10 Gb/s operating speed. That performance jump matters because current HBM technologies top out at roughly 36 GB per stack and ~1 TB/s bandwidth per module, limiting aggregate memory bandwidth on today’s accelerators. The shift to HBM4 unlocks much larger capacity and raw throughput: Nvidia’s Rubin is expected to use 288 GB of HBM4 to reach ~13 TB/s aggregate bandwidth, while AMD’s upcoming MI400 family aims for up to 432 GB and near 20 TB/s. Those increases translate directly into faster model training and larger in-memory models, but also raise power and supply-chain stakes — HBM already dominates GPU power draw (e.g., boosts from 250 W to ~1 kW per GPU observed with higher-capacity stacks). SK Hynix’s announcement also intensifies competition: Micron is sampling 36 GB HBM4 stacks with a 2,048‑bit interface and expects a 2026 ramp, and Samsung is pushing its own HBM4 despite validation delays. Which vendor wins volume contracts will shape GPU performance and supply resilience for the AI industry.