Nvidia's Rubin expected to use mostly 12-layer HBM4 despite reported 8-layer shift
newsDigiTimesOct 15, 2026
Nvidia's next-generation Rubin AI accelerators are expected to rely predominantly on 12-layer HBM4 high-bandwidth memory, despite reports that the chipmaker has asked Samsung Electronics and SK hynix to increase supplies of 8-layer products. The differing a...
Impact: MediumConfidence: LowNovelty: Incremental
newsdigitimes