SambaNova Systems

Market position

SambaNova competes in AI accelerator hardware and inference services against NVIDIA, which Sacra estimates holds roughly 85% market share, as well as AMD and Intel; cloud in-house silicon such as Google TPUs, AWS Trainium/Inferentia and Microsoft Maia; and inference-focused startups including Cerebras, described as its closest full-stack competitor and which completed an IPO in May 2026, and Groq, whose inference technology NVIDIA licensed in December 2025. Its stated differentiation is "premium inference" for the largest frontier models. The company was included in the Forbes AI 50 in 2026, and SambaNova Suite appeared on TIME's Best Inventions list in 2024.

SambaNova frames its edge as running the largest frontier models — including multi-trillion-parameter models fitted onto a single rack — at high sustained throughput and low power, enabled by dataflow execution and tiered memory rather than GPU-style architectures. The company claims the SN50 delivers three times the savings of competitive chips for agentic inference and over three times the throughput of NVIDIA's Blackwell B200 for agentic tasks, and cites independent benchmarks from SemiAnalysis and Artificial Analysis. Its Intel relationship, which includes co-development, host CPU supply and joint distribution, is another stated distinguishing factor.

Technology

The core technology is the Reconfigurable Dataflow Unit, a processor architecture using dataflow execution and a three-tier memory architecture to reduce data movement and allow hundreds of models to remain available on a single node with microsecond switching, which supports agentic workflows that route across reasoning, coding and voice models. The SN40L, introduced in 2023, added HBM for LLM inference; the fifth-generation SN50, unveiled in February 2026, is stated to provide five times more compute per accelerator and four times more network bandwidth than its predecessor and to scale to 256-accelerator clusters. Systems are packaged as SambaRack units (16 RDUs per rack) and exposed through OpenAI-compatible APIs and the SambaOrchestrator management layer. The company also demonstrated a disaggregated inference architecture for AI agents in June 2026, and with Intel is developing an architecture combining GPUs for prompt processing, RDUs for token generation and Intel Xeon 6 CPUs for agent orchestration.

Go-to-market

Distribution combines direct enterprise and government sales with channel partnerships: sovereign AI data center partners in Australia, Europe and the United Kingdom; regional cloud operators such as OVHcloud and SoftBank that resell inference capacity; SambaManaged deployments for data center operators; and joint go-to-market with Intel through its enterprise and cloud channels. A self-serve developer motion runs through SambaCloud's OpenAI-compatible APIs.

SambaNova identifies three customer types: sovereign clouds where governments fund local partners to build private national clouds; neoclouds; and enterprises building AI infrastructure for their own use. Named customers and partners include JPMorganChase, Saudi Aramco, Intel, SoftBank, OVHcloud, and Japanese firms, alongside public research institutions such as Argonne National Laboratory and Riken.

Geography

Headquarters are variously reported as Palo Alto, California and San Jose, California. Deployments and partners span the United States (Argonne National Laboratory, JPMorganChase, Vector Core Compute), Japan (Riken/Fugaku, SoftBank), Germany and the rest of Europe (OVHcloud availability zone), the United Kingdom and Australia via sovereign AI data center partners, and the Middle East via customers such as Saudi Aramco and investor Qatar Investment Authority.

History

After a $2 million seed from Celesta Capital in November 2017 and the 2018 Series A, SambaNova raised a $150 million Series B in April 2019, a $250 million Series C in February 2020 and a $676 million Series D led by SoftBank Vision Fund 2 in April 2021 that valued it above $5 billion, bringing cumulative funding to roughly $1.1 billion. It launched RDUs in 2020, began offering cloud AI services and inference platforms in the early 2020s, and introduced the SN40L in September 2023 (on-premises from November 2023). Reports later described stalled fundraising and an exploration of a sale, and Bloomberg reported December acquisition talks with Intel at roughly $1.6 billion. In February 2026 the company raised an oversubscribed Series E of over $350 million led by Vista Equity Partners with Intel Capital, unveiled the SN50 chip and announced a multi-year Intel collaboration; in July 2026 it announced the first close of a $1 billion Series F at an $11 billion valuation led by General Atlantic.

Risks & controversies

Reported risks include a period of stalled fundraising during which The Information reported the company was exploring a sale, and a December Bloomberg report of Intel acquisition talks at roughly $1.6 billion — well below the 2021 Series D valuation of $5.1 billion. Grokipedia notes that no official valuation was disclosed for the February 2026 Series E and that some reports estimated a post-money valuation of about $2.24 billion, a decline from 2021 attributed to market adjustment and competitive pressure. The company competes against NVIDIA's dominant position and depends on securing supply chain capacity, which management cited as the primary use of Series F proceeds. Sources also conflict on headquarters location (Palo Alto vs. San Jose) and on total funding raised.