Samsung Electronics’ next-generation high-bandwidth memory HBM4 will debut alongside NVIDIA’s AI accelerator Rubin at the GTC 2026 conference in March, following final quality tests passed with NVIDIA and AMD.
An exclusive article from biz.sbs.co.kr states that Samsung has cleared the final quality evaluations for HBM4 from both NVIDIA and AMD. Mass production of HBM4 begins next month. Units mass-produced and shipped from Samsung in February will reach NVIDIA for use in demonstrating Rubin’s performance at the March GTC event.
HBM4 from Samsung operates at 11.7 gigabits per second, the highest specification in the industry. This exceeds the 10 gigabits per second required by NVIDIA and AMD. Last year, the memory passed verification without any redesign, even after customers requested performance enhancements. This outcome demonstrates the technological completeness of Samsung’s HBM4 design.
The semiconductor industry reported these developments on the 25th. Evaluations within the sector indicate that Samsung’s memory technology has stabilized with this HBM4 shipment. Previous technological gaps with competitors, evident during the HBM3 and HBM3E phases, have been addressed in HBM4. Samsung now enters a recovery phase for its prior product leadership position.
Full-scale supply of HBM4 in large volumes is projected for around June. HBM4 integrates directly into AI accelerators such as NVIDIA’s Rubin, linking its availability to customers’ schedules for final product mass production. Major customers currently produce next-generation chips through foundries. Consequently, Samsung adjusts HBM4 shipment volumes to align with these customers’ actual mass production timelines and specified quantities.
