SambaNova Raises $1B in Series F First Close at $11B Valuation

Introduction

SambaNova Systems announced on July 8, 2026, that it has completed the first close of its Series F funding round. The round raised $1 billion at an $11 billion post-money valuation. General Atlantic led the round.

This is a first close only. SambaNova and its investors expect a second close in the coming weeks, so the round is not yet finalized. Even so, the $1 billion first tranche is one of the larger single fundraises among AI inference chip startups in 2026, and it places SambaNova among a small group of companies attempting to build alternatives to Nvidia for AI inference workloads.

SambaNova, founded in 2017 and headquartered in Palo Alto, California, designs custom AI chips built around a Reconfigurable Dataflow Unit (RDU) architecture rather than the GPU designs used by Nvidia. The new capital arrives as the company prepares to scale production of its next-generation SN50 chip and expand deployments with enterprise, sovereign, and neo-cloud customers.

Feature Overview

Reconfigurable Dataflow Architecture

SambaNova's core technology differentiator is the RDU, a tile-based dataflow chip design rather than a general-purpose GPU. Each RDU combines processing circuits with a three-tiered memory hierarchy: high-speed SRAM for active computation, HBM (high-bandwidth memory) for model weights and key-value cache, and DRAM for inactive or less frequently used networks. This layered memory approach is intended to let a single system hold larger models, and more distinct models simultaneously, than a GPU-only setup with a flatter memory structure.

The company's current flagship product is the SN40L, launched in September 2023, alongside its next-generation SN50 chip, which SambaNova unveiled in February 2026. SN50 chips are shipped in a 16-chip appliance called the SambaRack. The system is air-cooled and draws approximately 20 watts per chip, a relatively low power figure that SambaNova positions as an advantage for data center deployment density and cooling costs.

Rather than positioning RDUs as a full replacement for Nvidia GPUs, SambaNova frames its chips as complementary hardware in what it calls a disaggregated inference approach. In this model, Nvidia GPUs handle the prefill stage of inference — processing the initial input prompt — while SambaNova RDUs handle the decode stage, which generates output tokens one at a time and is typically more memory-bandwidth-bound. This split allows enterprises to mix hardware types within a single inference pipeline rather than choosing one vendor exclusively.

Funding Details and Investor Base

Beyond lead investor General Atlantic, the Series F first close included Seligman Ventures, T. Rowe Price Associates Inc., and Capital Group as significant participants. Additional investors named in the announcement include BlackRock-managed funds, Intel Capital, the Qatar Investment Authority (QIA), Vista Equity Partners, and JPMorgan Chase & Co. Several of these — Intel Capital and BlackRock-managed funds — have backed SambaNova in earlier rounds as well, indicating continuity in its investor base alongside new entrants such as sovereign wealth and financial-sector capital.

SambaNova CEO Rodrigo Liang said the capital is being used "to secure the supply chain," which he described as essential to fulfilling orders over the next 12 months. Beyond supply chain security, the company said funds will go toward expanding manufacturing and deployment capacity, accelerating product innovation, scaling deployments for enterprise customers, neo-clouds, sovereign AI customers, and service providers, and enhancing its chip lineup, rack design, and software stack.

Funding History

The Series F first close follows a rapid escalation in round sizes since SambaNova's founding in 2017.

The jump from Series E's $350 million in February 2026 to a $1 billion first close just five months later reflects both accelerating capital needs for chip manufacturing and heightened investor interest in inference-focused AI infrastructure.

Market Context

For enterprise AI buyers, this funding round is most relevant as a signal of SambaNova's near-term viability as an infrastructure vendor rather than as a product update. JPMorgan Chase & Co. has selected SambaNova as an inference infrastructure partner, deploying SN40/SN40L and SN50 systems for secure, on-premises AI inference on enterprise workloads. Darrin Alves, CIO of Infrastructure Platforms at JPMorgan Chase, is quoted in connection with the partnership. JPMorgan is also listed among the Series F investors, tying its role as customer and financial backer together.

SoftBank is reported as the first deployment partner for the SN50 chip, and SambaNova has cited Saudi Aramco and Intel among other customers. These relationships suggest demand concentrated among large enterprises, sovereign entities, and financial institutions that prioritize on-premises or private inference deployments over shared cloud infrastructure — a segment where data residency and security requirements can matter as much as raw throughput.

On performance, SambaNova has published claims from testing conducted with third-party benchmarking firm Artificial Analysis. According to SambaNova, a heterogeneous setup combining four Nvidia H200 GPUs with sixteen SN50 RDUs reached 763 tokens per second on a large language model benchmark at short context length (10,000 input tokens), and sustained over 450 tokens per second at longer context lengths. Separately, SambaNova claims the SN50 delivers more than three times the throughput of Nvidia's B200 GPU, with peak speed up to five times faster. These figures come from the vendor and its benchmarking partner rather than from fully independent, reproducible third-party testing, and prospective buyers should treat them as company-reported claims pending broader independent verification.

Pros and Cons

SambaNova's memory-tiered RDU architecture offers a genuine technical differentiator, particularly for workloads that require holding multiple large models in memory simultaneously. Its disaggregated inference model, splitting prefill and decode across GPU and RDU hardware, gives enterprises a path to mix vendors rather than commit fully to one architecture. The company's investor roster now includes established asset managers and a sovereign wealth fund, which may support longer-term capital stability compared with pure venture backing.

On the downside, the $1 billion figure represents only a first close, with a second close still pending as of this announcement — the ultimate size and finality of the Series F round remain unconfirmed. The company's own performance benchmarks, while conducted with a named third-party firm, have not been independently reproduced outside SambaNova-affiliated testing. SambaNova also faces direct competition from other well-funded inference-chip challengers, meaning its capital advantage may narrow over time.

Outlook

Liang said SambaNova is considering an initial public offering in 2027, most likely in the U.S. The Series F financing, once fully closed, would give the company a substantially larger balance sheet heading into any public listing process, though the timeline and structure of an IPO remain unconfirmed statements of intent rather than filed plans.

SambaNova operates within a broader trend of investors backing Nvidia inference-chip challengers, a group that also includes Cerebras, known for its wafer-scale engine architecture, and Groq, known for its deterministic low-latency architecture. Each company pursues a distinct hardware approach, and SambaNova's three-tiered memory architecture is its primary technical bet against both GPU incumbents and other specialized challengers. Whether this bet translates into durable market share will depend heavily on execution against the 12-month order fulfillment window Liang referenced, as well as on independent validation of its performance claims.

Conclusion

SambaNova's $1 billion Series F first close at an $11 billion valuation confirms continued investor confidence in AI inference infrastructure as a distinct market from GPU-centric training hardware. The round strengthens the company's supply chain position and expands its enterprise customer relationships, including with JPMorgan Chase and SoftBank. Enterprise buyers evaluating on-premises or hybrid inference infrastructure should weigh SambaNova's architectural differentiation against the fact that its headline performance figures remain vendor-reported and that the financing round itself is not yet fully closed. This is a story primarily relevant to infrastructure decision-makers, data center operators, and AI investors tracking the inference hardware market rather than to end users of consumer AI products.

Editor's Verdict

SambaNova Raises $1B in Series F First Close at $11B Valuation is a workable proposition that fills a clear gap, even if it doesn't fundamentally change the landscape.

The strongest case for paying attention is distinctive three-tiered memory architecture (SRAM/HBM/DRAM) tailored for multi-model, high-capacity inference, which raises the bar for what readers should now expect from peers in this space. Reinforcing that, established enterprise and sovereign customer relationships, including JPMorgan Chase, SoftBank, Saudi Aramco, and Intel adds practical value rather than just headline appeal. The broader signal worth registering is straightforward: the $1 billion raised is only a first close; a second close is still pending, so the full Series F size is not yet finalized. On the other side of the ledger, the $1 billion round is only a first close; the round is not yet fully closed and its final size remains unconfirmed is a real constraint, not a marketing footnote, and it should factor into any serious decision. Layered on top of that, headline performance claims (tokens/second, throughput versus Nvidia B200) are vendor-reported and tested with a partner firm, not independently verified by outside benchmarking narrows the set of teams for whom this is an obvious yes.

For AI industry watchers, strategy teams, and decision-makers tracking platform shifts, the smart move is to track its trajectory and revisit once the rough edges are filed down. For everyone else, the safer posture is to monitor coverage and revisit once the use cases that matter to your team are demonstrated in the wild.