At COEX in Gangnam-gu, Seoul in October, HBM4 and HBM3E are on display at the Samsung Electronics booth at Semiconductor Exhibition (SEDEX) 2025./Courtesy of Yonhap News

Samsung Electronics is said to be in the final stretch of negotiations with Nvidia to supply next year's sixth-generation high bandwidth memory (HBM4). After SK hynix said its HBM4 volume for next year was "sold out," Samsung Electronics appears set to become the second-largest supplier of HBM4 after SK hynix. While this year's fifth-generation HBM (HBM3E) volume amounted to only a small supply to Nvidia, dealing a blow to its pride, the company appears to be moving to restore its standing starting with HBM4.

According to industry sources on the 15th, Samsung Electronics is expected to supply more than 30% of the HBM4 volume that Nvidia requested from the memory chip industry for next year. Samsung Electronics is said to be completing quality certification by mounting the HBM4 samples provided to Nvidia on the next-generation artificial intelligence (AI) accelerator "Rubin platform" that Nvidia plans to release to the market next year.

A source familiar with Samsung Electronics' situation said, "Unlike the HBM3E market, Samsung Electronics' share within Nvidia's supply chain for HBM4 is expected to exceed 30%," and added, "While meeting the performance requirements requested by Nvidia, it appears the company will supply a larger volume than Micron."

Samsung Electronics has struggled to supply HBM to Nvidia. While rivals SK hynix and Micron have steadily supplied HBM3E 8‑high and 12‑high products, Samsung Electronics suffered multiple failures in quality certification. To regain competitiveness in the HBM4 market, Samsung Electronics has taken a bold step by using its own foundry (contract chip manufacturing) 4‑nanometer (nm; 1 nm = one billionth of a meter) process and equipping a generation-ahead DRAM compared with rivals. Samsung Electronics is also pushing to expand HBM4 production capacity at its Pyeongtaek Campus to meet HBM4 demand.

SK hynix is expected to solidify its position as the largest supplier by providing HBM4 close to 70% of the total volume requested by Nvidia for next year. SK hynix has agreed to supply the maximum HBM4 volume that can be handled within its scheduled HBM4 production capacity next year. SK hynix is also ramping up production capacity as much as possible to meet HBM demand. A semiconductor industry source said, "For the HBM4 volume that Nvidia requested for next year, SK hynix finalized a contract to supply the maximum volume the company can accommodate," adding, "If plans to expand production capacity become more concrete, additional talks could take place."

Micron is reportedly set to supply less than 10% of the volume, trailing SK hynix and Samsung Electronics. Micron is said to have faced technical issues, including a partial redesign of HBM4, which delayed negotiations somewhat. While Micron met Nvidia's HBM4 product criteria and delivered a customer sample (CS), it was assessed to be inferior in performance compared with rivals.

Unlike SK hynix and Samsung Electronics, Micron uses its own DRAM process, not a foundry process, for the logic die that serves as the brain of HBM. While SK hynix and Samsung Electronics used advanced foundry processes to meet the product performance that Nvidia requested and went into supply talks, Micron struggled to improve performance and thus failed to gain an advantage in volume negotiations.

※ This article has been translated by AI. Share your feedback here.