SambaNova Systems and Super Micro Computer Inc. (Nasdaq: SMCI) are partnering to develop and sell AI inference systems, initially planning deployments involving 10,000 SambaNova SN50 Reconfigurable Dataflow Units (RDUs). The collaboration will combine specialized inference processors with graphics processing units (GPUs) to support AI workloads across data centers worldwide.
Under the agreement, Supermicro will integrate SambaNova’s SN50 chips into its AI disaggregated inference systems, which will pair the processors with GPUs. The architecture is designed to divide AI inference tasks between different types of processors, with NVIDIA GPUs handling prompt processing and SambaNova’s SN50 chips accelerating response generation.
The companies are targeting the growing demand for inference infrastructure as businesses deploy agentic AI applications that require models to process and respond to requests continuously.
“Agentic AI is the next wave of inference, and it has to run everywhere,” said Rodrigo Liang, co-founder and CEO of SambaNova. He said the collaboration would make SN50 accelerators available alongside GPUs through Supermicro’s global infrastructure business.
Supermicro President and CEO Charles Liang said the partnership would expand the company’s Data Center Building Block Solutions (DCBBS) portfolio with heterogeneous computing architectures combining GPU processing and specialized inference acceleration.
Supermicro to Handle System Integration and Support
Supermicro will be responsible for integrating, testing, delivering, and supporting the systems, including providing on-site customer service. SambaNova’s engineering teams will assist with model deployment and optimization on SN50 processors and support customer implementations.
The companies plan to initially target neocloud providers and sovereign cloud operators before expanding into enterprises operating AI infrastructure within their own data centers.
The collaboration reflects a broader industry shift toward heterogeneous AI computing, in which specialized processors complement GPUs to address different stages of model execution. As inference workloads expand, infrastructure providers are exploring architectures designed to improve performance and operational efficiency.
