Nvidia's Rubin expected to use mostly 12-layer HBM4 despite reported 8-layer shift

DIGITIMESOctober 15, 2026en

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 assessments suggest that changes in memory configurations do not necessarily signal weaker HBM demand, as manufacturing constraints and supply shortages are expected to support memory prices in 2027.

This is a short summary published by AI Global Wire. The full article is owned and hosted by DIGITIMES — open it there to read it in full.