Rumor
NVIDIA to optimize HBM4 specifications amid yield management and HBM3E cost pressures
Friday, February 6, 2026 at 02:15 PM
NVIDIA is expected to optimize HBM4 performance metrics to balance technical specifications with production yields, while simultaneously addressing rising cost pressures affecting the HBM3E supply chain.
Context
Nvidia is reportedly adjusting its HBM4 performance benchmarks to prioritize supply stability for its next-generation Vera Rubin AI platform. While original targets demanded per-pin speeds of 11 Gbps, rumors suggest a potential relaxation to approximately 10 Gbps to ensure sufficient volume. This pivot addresses persistent yield challenges as manufacturers transition to the significantly more complex sixth-generation high-bandwidth memory architecture.
The supply chain is currently defined by a two-player split for the 2026 cycle. SK Hynix is projected to secure a dominant 70% share of initial orders, with Samsung capturing the remaining 30%. Micron was initially sidelined by the aggressive original speed requirements, but a downward specification revision could offer the U.S.-based firm a late-entry opportunity to fill production gaps if either Korean rival faces output constraints.
Initial shipments of the new memory are slated for late February 2026, ahead of the official Vera Rubin reveal at GTC 2026 in mid-March. By prioritizing volume over peak speed, Nvidia aims to mitigate potential shortages and maintain its aggressive 2026 deployment roadmap for global data center customers.
Related Companies
Micron
MU