Samsung is poised to commence mass production of its sixth-generation high-bandwidth memory (HBM4) chips in February 2026, following a successful quality assessment by Nvidia. This move comes as Samsung aims to avoid past supply issues experienced with its fifth-generation chips. Production will take place at Samsung’s Pyeongtaek campus, with rival SK Hynix also planning to start production of its own HBM4 chips around the same time.
The majority of Samsung’s HBM4 chips are set to be integrated into Nvidia’s upcoming AI accelerator system, Vera Rubin, expected in the latter half of 2026. Additionally, some chips will be supplied to Google for its seventh-generation Tensor Processing Units. Samsung has opted for a 10nm-class fabrication process for its chips, reportedly providing superior performance compared to SK Hynix’s 12nm process. Internal tests indicate Samsung’s chips can achieve speeds up to 11.7Gbps.
