Full-stack AI platform with Reconfigurable Dataflow — from SN40L chips to enterprise generative AI
HQ: Palo Alto, CA, United States | Founded: 2017 | Employees: 500-1,000 | Stage: Late Stage VC | Website: https://sambanova.ai
SambaNova Systems was founded in November 2017 in Palo Alto, California, by Rodrigo Liang (CEO, former EVP at Oracle and Sun Microsystems), Prof. Kunle Olukotun (Stanford Electrical Engineering and Computer Science, pioneer of multi-core processors), and Prof. Christopher Ré (Stanford Computer Science, expert in data systems and machine learning). The company was built to rethink AI compute from the ground up with a Reconfigurable Dataflow Architecture (RDA) implemented as the Reconfigurable Dataflow Unit (RDU), an alternative to GPUs that minimizes data movement and enables software-defined hardware reconfiguration between workloads without respinning silicon.
SambaNova monetizes as a full-stack platform: the SN40L RDU chip (5nm TSMC, 3-tier memory with 12TB DDR, HBM and on-chip SRAM supporting 1M+ token context), DataScale rack-scale systems and SambaRack on-prem/cloud appliances, and SambaNova Suite software including SambaFlow (PyTorch/JAX compiler), Samba-1 composition-of-experts generative AI platform, and DataScale Dataflow-as-a-Service. Customers span U.S. national labs (Argonne National Laboratory AI Testbed), sovereign clouds (UAE government ministry via G42/Inception partnership), and enterprises deploying private SambaCloud. Backed by SoftBank Vision Fund 2, Temasek, GIC, Intel Capital, GV (Google Ventures), Walden International, BlackRock and Redline Capital, the company hit $5.1B valuation after its $676M Series D in April 2021 and estimates ~$100M revenue.
SambaNova is the $5.1B anti-Nvidia bet on Reconfigurable Dataflow — a full-stack AI hardware + enterprise software platform (SN40L RDU + DataScale + SambaNova Suite/Samba-1) targeting trillion-parameter training and sovereign inference. With $1.1B raised led by SoftBank Vision Fund 2's $676M Series D (Apr 2021), strategic backing from Intel Capital, GV, BlackRock, Temasek and GIC, and landmark deployments at Argonne National Laboratory and UAE ministries, SambaNova offers dataflow reconfigurability and terabyte-scale memory that GPUs cannot match for large-context inference — but faces Nvidia's CUDA ecosystem, capital intensity, and concentration in lumpy government/enterprise deals on ~$100M revenue.
Browse: Home | Blog | About | All Startups A-Z | View full profile on StartupWiki