Samsung, Nvidia discuss HBM4E, HBM5 deals 

Samsung Electronics and Nvidia discussed long-term memory supply deals covering seventh-generation high-bandwidth memory (HBM4E) and next-generation HBM5 during Jensen Huang's final day in Korea on Monday.

Jun Young-hyun, Samsung Electronics vice chair and head of the semiconductor division, met with Huang at Nvidia's Korea AI Ecosystem Reception at The Shilla Seoul in central Seoul.

"From next year onward, we discussed longer-term cooperation covering HBM4E and HBM5, as well as foundry," Jun told reporters after the meeting.

"We had a very good conversation — in fact, I'd say it was the most productive meeting we've had in all our years of working together," he said.

Samsung sent samples of its HBM4E to clients, including Nvidia, last month, ahead of rival SK hynix and a month earlier than its own planned mid-year shipping timeline.

"We are also working together on autonomous driving chips and Groq's LPU at the 4-nanometer and 8-nanometer nodes, and we are discussing cooperation on the next generation of those as well," he added.

The chipmaker's first priority this year is to supply Nvidia with enough sixth-generation HBM4 and server low-power dynamic random access memory (DRAM) modules known as Socamm2. It is currently supplying HBM4 running over 11.7 gigabits per second (Gbps) for Nvidia’s AI accelerator, Vera Rubin, which has entered mass production. For the Vera CPU, Samsung is supplying LPDDR5X-based Socamm2.

Jun declined to comment on whether Nvidia had sought or secured a two-year memory supply agreement, a trend that has emerged this year amid a prolonged chip shortage driven by the AI boom.

BY LEE JAE-LIM [[email protected]]