Samsung Electronics’ next-generation high-bandwidth memory, HBM4 (sixth generation), will make its official debut alongside NVIDIA’s AI accelerator “Rubin” at the company’s GTC 2026 conference in March.
The semiconductor industry reported on the 25th that Samsung has passed final quality tests for HBM4 from both NVIDIA and AMD. The company will begin mass production next month. HBM4 units mass-produced and shipped from Samsung in February are set to reach NVIDIA for use in performance demonstrations of Rubin at the March GTC event.
Samsung’s HBM4 operates at 11.7 Gb/s, exceeding the 10 Gb/s required by NVIDIA and AMD. Last year, it passed verification without redesign, even after customers requested performance upgrades, which confirms its technological completeness.
Industry assessments indicate that this shipment signals the normalization of Samsung’s memory technology. The product resolves the technological gap with competitors that appeared in HBM3 and HBM3E generations, positioning Samsung to regain its previous product leadership.
Full-scale, large-volume supply of HBM4 is projected for around June. As HBM4 integrates into AI accelerators such as Rubin, its supply aligns with customers’ final product mass-production schedules. Major customers currently produce next-generation chips through foundries, so Samsung will adjust HBM4 shipment volumes to match their actual mass-production timelines and required quantities.
The details come from an exclusive article on biz.sbs.co.kr.
