
SambaNova
Full-stack AI inference infrastructure on custom dataflow chips, cloud or on-prem.
About
SambaNova Systems builds custom AI chips, hardware systems, and software purpose-built for inference — running trained models in production — as an alternative to NVIDIA GPU infrastructure. Its reconfigurable dataflow architecture (SN40L, and the SN50 shipping in H2 2026) keeps data close to compute across SRAM, HBM, and DRAM to cut data movement, with the company claiming a single rack can serve multi-trillion-parameter models at 10kW.
Customers access the stack via SambaNova Cloud or deploy on-premises appliances, which appeals to banks and sovereign clouds needing secure in-house inference — JPMorgan Chase selected SambaNova as an inference infrastructure partner in 2026. The company raised a $1 billion Series F in July 2026 led by General Atlantic at an $11 billion valuation, bringing total funding past $2.5 billion, and is reportedly considering a US IPO. Independent benchmark data and public revenue figures remain thin, so performance claims rest largely on vendor reporting; an earlier reported Intel acquisition discussion did not materialize.
What it can do
- Inference-optimized RDU chips (SN40L/SN50) with dataflow architecture
- Cloud API plus on-premises appliances for secure in-house inference
- Single-rack serving of very large models at 10kW power draw
- Enterprise partnerships including JPMorgan Chase and SoftBank
Setup
Sign up for SambaNova Cloud for API access, or engage their sales team for on-premises SN40L/SN50 system deployment.
Prerequisites
For cloud: an account and API key. For on-prem: data-center capacity and an enterprise agreement.
Use cases
Works with
Trust score — 9/10
Scored 9/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 65; Trending score 86/100 from researcher assessment.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.