Nvidia CEO Jensen Huang said in Seoul that Samsung Electronics, SK hynix, and Micron Technology have been qualified to supply HBM4 memory for Nvidia’s Vera Rubin AI platform.

He said all three are in production and competing to support Vera Rubin, which is set for delivery in the third quarter.

Vera Rubin is Nvidia’s next AI platform. Nvidia said its Rubin graphics processing units use HBM4 that doubles interface width compared with HBM3E, though it has not disclosed HBM4 capacity per graphics processing unit.

HBM4 matters because the JEDEC standard supports up to 2 TB/s per stack on a 2,048-bit interface, vs. about 1 TB/s for HBM3E on a 1,024-bit bus.

The qualification expands Nvidia’s supplier base in a market where SK hynix held 62% share in the second quarter of 2025, while HBM supply remains tight.

Huang also said Nvidia is hiring for a new research and development center in Korea. Arm CEO Rene Haas described memory as “probably the toughest” bottleneck to resolve.

🔗 Source: Bloomberg

Recent Nvidia developments