SambaNova Systems, Inc.

About SambaNova Systems, Inc.

Financials

Dated, source-cited figures. Estimates are marked.

Products by SambaNova Systems, Inc.

All products by SambaNova Systems, Inc. reviewed by our AI panel of experts.

SambaNovaAI APIs

AI inference infrastructure built on custom RDU chips

Free7.2/10

How Our Review Panel Scores SambaNova Systems, Inc.

Six independent reviewer personalities assessed SambaNova Systems, Inc.'s product on their own terms. They landed within 0.9 points of each other, an unusually strong consensus.

Averaged across 1 reviewed product in our catalog. Scores are out of 10.

Competitive Landscape

Who SambaNova Systems, Inc. competes with, and how they differ.

Cerebras8.0/10

The closest structural rival: custom AI silicon plus a first-party inference cloud, chasing the same sovereign-AI and national-lab deals. Cerebras bets on a single wafer-scale engine with all weights in on-chip SRAM, where SambaNova uses a reconfigurable dataflow unit with tiered HBM/DRAM memory to hold trillion-parameter models on far fewer nodes.

1 product reviewed on TopReviewed

Groq7.7/10

Also pairs proprietary inference silicon with a developer cloud and competes on tokens per second. Groq's LPU is a deterministic, SRAM-only inference-only design sold mainly as GroqCloud API capacity, while SambaNova trains as well as serves and leads with on-premises racks for regulated buyers such as JPMorganChase and national laboratories.

1 product reviewed on TopReviewed

NVIDIA Corporation8.2/10

The incumbent SambaNova benchmarks against, claiming roughly 3x lower total cost of ownership versus GPUs. NVIDIA's advantage is CUDA and an entrenched software ecosystem sold through every cloud, whereas SambaNova ships a vertically integrated rack and model stack that has no third-party toolchain behind it.

1 product reviewed on TopReviewed

Together AI8.0/10

Competes directly for SambaNova Cloud's token-serving business with an open-model inference and fine-tuning API, but rents NVIDIA GPU capacity rather than designing silicon. Together sells breadth of models and GPU cluster rental; SambaNova sells speed and precision on hardware it controls end to end.

1 product reviewed on TopReviewed

Fireworks AI8.0/10

Another GPU-based fast-inference API targeting the same open-weight serving workloads as SambaNova Cloud, optimising at the software layer with custom kernels and speculative decoding instead of custom chips. It has no on-premises appliance business, so it does not compete for SambaNova's enterprise and sovereign deployments.

1 product reviewed on TopReviewed

Company Timeline

Key milestones in SambaNova Systems, Inc.'s history, with sources.

1. Jul 8, 2026Funding

   Series F first close — $1B at an $11B valuation, led by General Atlantic

   Seligman Ventures, T. Rowe Price Associates, Capital Group, BlackRock funds, Intel Capital, Vista Equity Partners, Battery Ventures and the Qatar Investment Authority joined the round. Announced alongside JPMorganChase selecting SambaNova as an inference-infrastructure partner for on-premises SN40L and SN50 deployments.

   sambanova.ai
2. Feb 24, 2026Funding

   Series E — $350M+ co-led by Vista Equity Partners and Cambium Capital

   Intel Capital, Battery Ventures, Seligman Ventures, Assam Ventures, Mayfield and T. Rowe Price-managed accounts participated. The round was announced together with a multi-year Intel partnership covering Xeon-based AI cloud, system integration and co-selling. No valuation was disclosed.

   sambanova.ai
3. Feb 24, 2026Launch

   SN50 reconfigurable dataflow unit announced for agentic AI

   SambaNova said the SN50 delivers five times the compute per accelerator and four times the network bandwidth of the SN40L, scales to 256 accelerators over a multi-terabyte-per-second interconnect, and supports models above 10 trillion parameters with 10M+ token context lengths.

   sambanova.ai
4. Mar 5, 2025Partnership

   SoftBank Corp. expands SambaNova deployment for AI inference across APAC

   SoftBank Corp. widened its use of SambaNova racks to offer high-speed inference through SambaNova Cloud in Japan, serving Japanese open-source models such as Swallow alongside Meta Llama and Alibaba Qwen.

   sambanova.ai
5. Sep 10, 2024Launch

   SambaNova Cloud inference API launched

   The service ran Llama 3.1 405B at 132 tokens per second in full 16-bit precision and Llama 3.1 70B at 461 tokens per second on SN40L hardware, offered in free, developer and enterprise tiers.

   sambanova.ai
6. Feb 28, 2024Launch

   Samba-1 — one-trillion-parameter enterprise model released

   Samba-1 used a Composition of Experts architecture assembling more than 50 open-source models, including Llama 2, Mistral, Falcon and DeepSeek Coder, for private deployment on customer premises running on SN40L hardware.

   sambanova.ai
7. Sep 19, 2023Launch

   SN40L chip unveiled

   The fourth-generation reconfigurable dataflow unit, manufactured by TSMC, combined dense and sparse compute with a tiered HBM and DRAM memory system, supporting models of up to 5 trillion parameters and 256k+ sequence lengths on a single system node.

   sambanova.ai
8. Apr 13, 2021Funding

   Series D — $676M led by SoftBank Vision Fund 2 at a $5.1B valuation

   Temasek and GIC joined as new investors alongside BlackRock, Intel Capital, GV, Walden International and WRVI. The round took total funding past $1B, then the largest sum raised by an AI chip startup.

   techcrunch.com
9. Feb 25, 2020Funding

   Series C — $250M led by BlackRock

   GV, Intel Capital, Walden International, WRVI Capital and Redline Capital also participated. The round took total funding raised to more than $450M.

   venturebeat.com
10. April 2019Funding

    Series B — $150M

    The company closed a $150M Series B in April 2019, its second institutional round, a year after the Series A.

    venturebeat.com
11. Mar 15, 2018Funding

    Series A — $56M led by GV and Walden International

    Redline Capital and Atlantic Bridge Ventures also participated in the company's first institutional round, roughly four months after founding.

    techcrunch.com
12. November 2017Founded

    SambaNova Systems founded by Kunle Olukotun, Christopher Ré and Rodrigo Liang

    Stanford computer scientists Kunle Olukotun and Christopher Ré co-founded the company with former Oracle engineering executive Rodrigo Liang, who became CEO, to build a reconfigurable dataflow architecture as an alternative to GPUs for machine learning.

    techcrunch.com