The latest market news on Friday night said that after many setbacks, Samsung Electronics’ 12-layer HBM3E high-bandwidth memory chip finally received approval from computing power giant Nvidia. According to people familiar with the matter,Samsung’s fifth-generation 12-layer HBM3E recently passed Nvidia’s qualification test. This approval comes 18 months after Samsung completed chip development, and several tests during this period failed to meet Nvidia's demanding performance requirements.
Technical information shows that NVIDIA's flagship B300 artificial intelligence accelerator, as well as AMD's MI350 and other systems all deploy this large-capacity memory chip. Samsung has previously shipped the chip to AMD, but has been stuck at Nvidia's supplier threshold. Therefore, this certification also plays an important role in restoring Samsung's technological credibility.
According to industry insiders, this achievement is largely due to Samsung Electronics’ chip business leader Jeon Yong-hyun’s decision at the beginning of the year to redesign the DRAM core of HBM3E, which solved the thermal performance problems of the early version.
However, since Samsung is the third approved supplier after SK Hynix and Micron Technology, sources said Nvidia’s order volume will still be relatively small.. Therefore, an industry executive commented: "For Samsung, supplying Nvidia is more about pride than revenue. Being recognized by Nvidia means that its technology is back on track."
HBM4 The war is fierce
The return of Samsung Electronics also means that the battle for HBM4 is heating up.
As the sixth generation of high-bandwidth memory,Nvidia plans to use this chip on its next-generation Vera Rubin architecture, due out next year.
It is reported that Samsung plans to use its most advanced 10-nanometer (1c) DRAM process and equip it with 4nm logic chips produced by its foundry. Other competitors use the previous generation 1b process and 12nm logic chips.
As a key competitive indicator, Nvidia has asked suppliers to push HBM4's data transfer speed to more than 10Gb per second, significantly higher than the 8Gb industry standard. According to people familiar with the matter, Samsung has demonstrated the ability to reach 11Gbps, which is higher than SK Hynix's 10Gbps, while Micron Technology has difficulty meeting this requirement.
People familiar with the matter revealed,Samsung plans to ship a large number of HBM4 samples to Nvidia this month, aiming to obtain certification as soon as possible. The company has previously stated that it is negotiating with major AI chip manufacturers such as Nvidia, Broadcom and Google for HBM4, and large-scale shipments can begin as early as the first half of 2026.
If it can obtain Nvidia certification early in the competition for HBM4, Samsung is expected to regain market share in this key memory segment.
As a key competitor of Samsung,SK Hynix said on Friday that it has completed the internal verification and quality assurance process of HBM4 and is preparing for large-scale mass production.The company said its products have doubled the bandwidth and improved energy efficiency by 40% compared to the previous generation. Driven by this news, SK hynix rose by more than 20% last week, and rose by 7.46% this week, makingThe increase this year has reached 102%.
(SK Hynix weekly chart, source: TradingView)
Micron Technology also announced on June 10 that it had delivered 12-layer HBM4 36GB chip samples to a number of key customers.