Nvidia Says Custom Memory Offers 30% Higher Bandwidth
Key Facts
Nvidia said its custom NVHBM memory offers up to 30% higher bandwidth per stack than standard HBM4e memory, according to a technical blog post the company published on August 26, 2026. The statement comes amid a report published by the Seoul Economic Daily on October 18, which said Samsung would use TSMC’s 3-nanometer process to produce base dies for Nvidia’s custom memory. The report concerns production of the base die, not the memory chips themselves. NVHBM integrates Nvidia’s own memory controller into the HBM base die, according to the company. The design places some memory control functions on the die, while Nvidia compares bandwidth performance per stack with the standard HBM4e product.
Nvidia says NVHBM is built on the same technology it will use in future graphics processing units. The company presented the higher-bandwidth feature as a direct comparison with standard HBM4e and put the maximum increase at 30% per stack. On manufacturing, the October 18 report points to Samsung using TSMC’s 3-nanometer process to make the base dies, a detail that distinguishes the two companies’ roles in the production chain. This is part of Nvidia’s broader effort to expand supply-chain capacity for AI infrastructure components. On October 21, the company announced agreements with suppliers to expand capacity for core materials, advanced packaging, testing and photonics engines.
The agreements Nvidia announced on October 21 included Amkor Technology, IBIDEN, King Yuan Electronics, Samsung Electro-Mechanics, Siliconware Precision Industries and Unimicron Technology Corporation. The company said it is working with these suppliers to enable new and existing suppliers to expand manufacturing capacity in those areas. Nvidia is targeting a 2.5-fold increase in production capacity for core materials for its products over the three years following the announcement. It also expects to double total semiconductor testing capacity between 2026 and 2027. The company links these steps to strengthening supply-chain resilience and meeting demand as it seeks to expand production of AI infrastructure components.
In the market, TSMC shares closed at $353.1 on November 5, compared with a close of $395.75 in the previous session, according to EL7 data. Nvidia shares closed at $181.6 on the same day, according to the same data. On its operating outlook, TSMC said it expects revenue of between $52.3 billion and $54.3 billion in the fourth quarter of 2026, equivalent to quarter-on-quarter growth of 14% at the midpoint. September revenue was 511.857 billion New Taiwan dollars, up 54.6% year over year. The company is scheduled to release October sales on November 10, 2026.