SambaNova Systems has raised $1 billion in the first close of a Series F round at an $11 billion valuation, led by General Atlantic, with a second close expected in the coming weeks as more investors join. The round arrives just five months after the Palo Alto-based company’s $350 million Series E in February, which came bundled with the unveiling of its SN50 chip and included Intel as a co-investor, an unusually short gap between mega-rounds even by 2026’s inflated AI-funding standards.
SambaNova was founded in 2017 by Rodrigo Liang, a former Oracle executive who ran the SPARC processor and ASIC business, alongside two Stanford professors: Kunle Olukotun, sometimes called the father of the multi-core processor, and Christopher Ré, a MacArthur Fellow known for his data-systems research. Rather than adapting an existing GPU architecture the way most AI chip startups do, the founding team built what they call a Reconfigurable Dataflow Unit from scratch, a chip that reshapes its own circuitry for each specific neural network workload instead of running every model through the same fixed pipeline. That from-scratch approach is also why SambaNova’s path to this $11 billion valuation was not a straight line: eight months before this round closed, in October 2025, the company was reportedly exploring a sale after struggling to raise fresh capital at its then-roughly-$4 billion mark, a reversal of fortune that makes the jump to $11 billion in under a year one of the sharper swings in this AI-funding cycle.
The investor list signals how far SambaNova has moved from a niche chip-design bet into something institutional allocators now treat as core AI infrastructure. Alongside General Atlantic, the round includes Qatar Investment Authority, BlackRock, T. Rowe Price Associates, Capital Group, Vista Equity Partners, Battery Ventures, and several smaller specialist funds. That’s a materially different investor base than the one backing a typical Series F chip startup, closer to the sovereign-wealth-and-asset-manager mix now flowing into Anthropic and OpenAI than to a traditional semiconductor venture round.
SambaNova’s specific bet is inference, not training. Training a large model happens once and is dominated by Nvidia’s highest-end GPUs; inference, the actual process of running a trained model to generate responses in production, happens constantly, at far larger cumulative scale, and is where custom silicon can plausibly undercut Nvidia on cost per query rather than trying to out-engineer it on raw training throughput. That’s the pitch behind SambaNova’s SN40L and SN50 systems, and it’s also the pitch behind rivals like Groq and Cerebras, all racing to establish themselves as credible inference alternatives before Nvidia’s own inference-optimized offerings close the gap. Those rivals are raising at similar speed: Groq closed $750 million in September 2025 at a $6.9 billion valuation and separately secured a $1.5 billion Saudi Arabian commitment for chip deliveries, while Cerebras went public and now carries a roughly $49 billion market capitalization on $510 million of annual revenue, backed by a December 2025 agreement in which OpenAI committed to buying 750 megawatts of Cerebras inference capacity. SambaNova’s $11 billion sits between those two data points, still a fraction of Cerebras’s public-market value but confirmation that investors see at least three viable non-Nvidia inference challengers worth funding simultaneously, not just one eventual winner.
The round’s most concrete validation came in the same announcement rather than the fundraise itself: JPMorgan Chase named SambaNova as an inference-infrastructure partner, with SN40L and SN50 systems set to power secure, on-premises AI inference inside the bank. On-premises deployment matters specifically for financial institutions, which face data-residency and audit requirements that push against routing every AI inference call through a third-party cloud API, exactly the kind of regulated-industry use case that determines whether an inference-chip challenger becomes a durable business or stays a well-funded science project.
The broader pattern is one of AI capital diversifying away from a single point of failure. Nvidia’s dominance in training silicon is not seriously contested by anyone in this round of funding, but every major AI lab and every large regulated enterprise has an incentive to avoid depending on one vendor for the inference layer that will eventually carry the bulk of AI compute spend. SambaNova’s $11 billion valuation, still a fraction of Nvidia’s market capitalization but a serious number for a company outside the big three cloud providers, reflects investors betting that inference infrastructure ends up more fragmented than training infrastructure has been.
For Philippine companies, the direct relevance is narrow today, none of this hardware is being deployed locally, and SambaNova has no announced Philippine presence, but the trend line matters more than the specific vendor. The country’s BPO and financial-services sectors are exactly the kind of regulated, cost-sensitive, high-volume-inference environments this new generation of chip alternatives is built for, and BSP’s own recent push toward standardized, auditable digital-banking infrastructure creates the same on-premises and data-residency pressures that just won SambaNova its JPMorgan deal. If inference-chip competition genuinely drives down the cost of running AI at scale over the next two to three years, Philippine banks and BPOs stand to benefit from cheaper AI deployment options regardless of which specific vendor wins, the same way cloud-computing price wars a decade ago eventually benefited every downstream buyer, not just the initial customers who signed first.
Whether SambaNova specifically becomes one of the winners, or gets squeezed out once Nvidia and the hyperscalers respond in kind, is a two-to-three-year question, one made sharper by how close the company came to a distressed sale just months before this raise. What the JPMorgan deal establishes now is that a regulated financial institution with far more resources than any Philippine bank was willing to bet its own AI infrastructure on a non-Nvidia challenger, a signal worth tracking for any Philippine CTO currently assuming that a single dominant vendor is the only realistic choice for enterprise AI deployment going forward.
Share this article