SambaNova and Supermicro team up on SN50 inference racks, starting with 10,000 chips alongside Nvidia GPUs

AI chip company SambaNova announced on November 3 a strategic collaboration with server maker Supermicro to build AI inference racks powered by its fifth-generation SN50 chip. The racks will be offered as part of Supermicro's Data Center Building Block Solutions, with SambaNova's reconfigurable dataflow units (RDUs) running alongside Nvidia GPUs in the same systems; the collaboration is initially planned around 10,000 SN50 RDU chips. (Cover photo is a file photo of data-center racks, not the companies' products.)

Supermicro builds the systems; SambaNova tunes the models

Under the arrangement, Supermicro will integrate, test, deliver and support the systems, including on-site service, while SambaNova engineers help customers get models running and tuned and support deployments. The companies describe the systems as disaggregated inference architecture, splitting inference work between dedicated inference chips and GPUs rather than running everything on GPUs.

SN50: a fifth-generation chip built for inference

The SN50 is SambaNova's fifth-generation Reconfigurable Dataflow Unit, purpose-built for high-performance inference. The companies did not disclose pricing, a delivery timetable, first customers, or measured performance figures for SN50-based systems versus GPU-only systems; the announcement covers product development and supply planning, not deployed scale.