Samsung HBM4E World-First at GTC 2026 — Jensen Huang's "World's Best" Endorsement and 4.0 TB/s Specs
HBM4E (High Bandwidth Memory 4th-generation Extended) is Samsung's 7th-generation high-bandwidth memory chip, delivering 16 Gbps per pin and 4.0 TB/s bandwidth per stack — roughly 21% faster than the current HBM4 generation. Samsung Electronics unveiled HBM4E for the first time at NVIDIA GTC 2026 in San Jose on March 16, 2026, drawing a personal booth visit and public endorsement from NVIDIA CEO Jensen Huang.
Why Did Jensen Huang Personally Visit Samsung's Booth?
During GTC 2026, Jensen Huang walked into Samsung's booth and signed the HBM4E display frame, a Groq chip prototype, and a Vera Rubin test unit — all with a gold marker. Accompanied by Samsung's DSA division chief Sangyun Cho, Huang repeatedly exclaimed "Samsung is world's best!" and "Let's go Samsung!" The visit was far from ceremonial. Samsung now supplies both HBM memory and Groq 3 LPU foundry manufacturing (on its 4nm process) to NVIDIA, making it a dual-role strategic partner. During his keynote, Huang publicly thanked Samsung: "I want to thank Samsung, who manufactures the Groq 3 LPU chip for us, and they're cranking as hard as they can."
What Makes Samsung's HBM4E a Generational Leap?
Samsung's HBM4E brings three key upgrades. First, per-pin speed jumps to 16 Gbps, compared to HBM4's 13 Gbps — a 21% improvement. Second, per-stack bandwidth reaches 4.0 TB/s, up from 3.3 TB/s. On NVIDIA's upcoming Rubin Ultra platform with 16 HBM4E sites, total memory bandwidth hits 64 TB/s with 384 GB capacity. Third, Samsung's proprietary Hybrid Copper Bonding (HCB) technology enables 16+ layer stacking while reducing heat resistance by over 20% compared to conventional thermal compression bonding.
How Does This Shift the HBM Competitive Landscape?
SK Hynix currently holds roughly two-thirds of NVIDIA's 2026 HBM4 supply allocation. Samsung began HBM4 mass production in February 2026, and unveiled HBM4E just one month later — signaling aggressive intent. Samsung's unique advantage lies in being the only company that can supply memory, foundry services, and storage to NVIDIA simultaneously, an IDM (Integrated Device Manufacturer) strategy that no competitor can replicate. SK Hynix also showcased HBM4, HBM3E, and a liquid-cooling SSD co-developed with NVIDIA at GTC 2026, underscoring that the HBM race remains a two-horse competition between South Korea's semiconductor giants.
Key Takeaways
① Samsung HBM4E world-first — 16 Gbps per pin, 4.0 TB/s bandwidth, 48 GB per stack using Hybrid Copper Bonding
② Jensen Huang's public endorsement — Visited Samsung's booth, signed HBM4E with a gold pen, thanked Samsung for Groq 3 LPU foundry manufacturing in his keynote
③ HBM war Round 2 — Samsung's IDM total-solution strategy challenges SK Hynix's supply dominance, intensifying competition in AI memory
Jensen Huang's booth visit symbolizes a shifting dynamic in the HBM market. As AI infrastructure spending accelerates, the rivalry between Samsung and SK Hynix will ultimately determine how fast the entire AI semiconductor ecosystem advances.
👉 NVIDIA Groq 3 LPU Unveiled at GTC 2026 — GPU+LPU Division and Samsung 4nm Foundry — also worth a read.
📌 Sources: The Investor, WCCFTech, Samsung Newsroom, WinBuzzer (2026)
댓글
댓글 쓰기