Nvidia Reportedly Reduces Rubin Ultra HBM Stack Height as Memory Costs Rise

Nvidia is reportedly changing the planned HBM configuration for its Rubin Ultra AI accelerator, moving to an eight-high stack design from an earlier twelve-high approach, according to DigiTimes. The report links the change to higher memory costs and a greater focus on bandwidth efficiency and total system economics. HBM stack height is a key design choice for AI accelerators because it affects memory capacity, packaging complexity, power, and cost. A move toward fewer stacks or lower stack height could signal tighter optimization of memory resources as accelerator vendors balance performance against the cost of high-bandwidth memory. The supplied source does not provide a confirmed Nvidia product specification, product timing, memory supplier list, or expected HBM volume change. The report should be read as a roadmap signal rather than a confirmed shift in Nvidia procurement.