What SambaNova does
SambaNova (website sambanova.ai) is a private enterprise AI infrastructure company that focuses on faster, lower-power AI inference for large language models (LLMs) and agentic workloads. The company’s core differentiator is a reconfigurable dataflow approach delivered through its Reconfigurable Dataflow Unit (RDU) chips and purpose-built systems, along with software that targets production inference behaviors (e.g., managing long-context/agent workloads and deployment at scale).
From a product standpoint, SambaNova positions an end-to-end stack that can be deployed in enterprise data centers or as managed service offerings. On the systems side, SambaRack is the rack-level deployment that integrates hardware, networking, and software for model serving. SambaRack SN40 and SN50 are designed around RDUs and dataflow principles intended to reduce bottlenecks from memory movement and to improve inference efficiency for large models and token-generation workloads.
On the platform side, SambaNova emphasizes “SambaCloud”/hosted inference and enterprise/self-managed options (including a SambaStack path in its documentation), and provides developer-facing APIs and tooling for inference deployment. SambaNova also describes an internal architecture for inference that separates stages across heterogeneous compute (for example, using GPUs for prefill while RDUs handle decode in its disaggregated/heterogeneous blueprint).
Strategically, SambaNova is pushing for enterprise adoption of faster on-prem inference and for “sovereign” deployments with local governance. In 2026, it announced a large Series F first close valuing the company at $11B post-money and publicized production-oriented wins/choices of its RDUs for customers, including a JPMorganChase selection for on-prem inference.