SambaNova has closed the first part of a 1 billion dollar Series F at an 11 billion dollar valuation, led by General Atlantic, and the money is aimed at one target: running AI inference on hardware that is not a Nvidia GPU. Announced on July 8, 2026, only about five months after its last large round, the raise lands alongside a marquee customer win, with JPMorganChase selecting SambaNova to power secure, on premises AI inference on its own systems. For developers weighing where inference actually runs, the story is that the alternative silicon camp just got a lot better funded, and it is chasing the workloads enterprises most want to keep in house.
The short answer
SambaNova closed the first part of a 1 billion dollar Series F at an 11 billion dollar valuation, led by General Atlantic, roughly five months after its last big raise. The company builds AI chips and systems tuned for inference rather than general purpose GPUs, and it just landed JPMorganChase as an inference infrastructure partner for secure, on premises AI. The bet is simple: a lot of enterprise inference wants to run in house, not on rented GPU clusters, and the money is there to chase it.
Most of the AI money you read about goes to two places: the model labs and Nvidia. SambaNova sits in the third place, the one betting that the chip running your inference does not have to be a GPU. On July 8 that bet got a billion dollars bigger, and it came with a customer name that makes the pitch concrete.
What SambaNova raised
SambaNova completed the first close of a 1 billion dollar Series F, valuing the company at 11 billion dollars post money. General Atlantic led, joined by Seligman Ventures, T. Rowe Price, Capital Group, funds managed by BlackRock, Intel Capital, the Qatar Investment Authority, and Vista Equity Partners, among others. Chief executive Rodrigo Liang told reporters that more investors are expected to join, with a second close likely to follow in the coming weeks.
The timing is the tell. This round arrives only about five months after SambaNova's previous large raise, which had unveiled its SN50 system alongside a strategic collaboration with Intel. Raising again, at a higher valuation, that quickly usually means demand is running ahead of what the last round could fund.
The product bet: inference, not GPUs
SambaNova does not try to beat Nvidia at everything. It builds chips and full systems tuned for inference, the phase where a trained model actually serves requests, rather than the general purpose parts used for training. Its SN40L and newer SN50 systems are designed to run large models efficiently, and the company aims them squarely at on premises and private cloud deployments, the setups where an organization wants its data and models to stay inside its own walls.
That focus matters because inference, not training, is where most production spending eventually lands. Training a model is a one time cost. Serving it runs every hour of every day, and at scale the cost per token and the location of the hardware become the whole argument. If you can run inference efficiently on gear you own, in your own building, you sidestep both the GPU rental bill and the data governance headaches of sending everything to a public cloud.
Why the JPMorgan win carries weight
Funding rounds are easy to announce. Production customers are harder, and this one is notable. JPMorganChase selected SambaNova as an inference infrastructure partner, with its systems set to power secure, on premises AI inference at the bank. A large, heavily regulated financial institution choosing non Nvidia hardware to run private inference is close to the strongest validation the alternative silicon camp could ask for.
Read it as a readiness signal. Banks do not put experimental hardware into compliance sensitive workloads. If SambaNova cleared that bar, it means the systems, the tooling, and the model support are mature enough for demanding real world use, not just benchmarks. That is the gap most Nvidia challengers fail to cross, and clearing it is worth more to the pitch than any single specification.
What developers should take from it
You do not have to switch anything today, but the landscape is shifting in a way worth tracking.
- The layer below the GPU is filling in. More funded, production ready options exist for inference than a year ago. If your workload is inference heavy, it is worth knowing what runs it besides a rented GPU.
- On premises is back on the table. Privacy, compliance, and steady per token cost are pushing serious inference back inside company walls. Dedicated inference systems are built for exactly that.
- Test the boring things. If you evaluate alternative silicon, the questions that decide it are model support, how mature the software stack is, and real cost per token at your traffic. Get those right before the architecture story.
SambaNova's raise does not dethrone anyone. Nvidia still owns training and most of the market. But the money and the JPMorgan win together say the inference layer is becoming a real contest, and for developers that means more genuine choice about where and how your models actually run.
Sources and further reading
- TechCrunch: AI chip maker SambaNova raises 1B at 11B valuation, five months after last mega round
- General Atlantic: SambaNova completes first close of 1 billion dollar financing at 11 billion valuation
- HPCwire: SambaNova completes first close of 1B financing at 11B valuation
- citybiz: AI compute firm SambaNova raises 1B Series F led by General Atlantic
Frequently asked questions
How much did SambaNova raise and who led the round?
SambaNova completed the first close of a 1 billion dollar Series F, valuing the company at 11 billion dollars post money. General Atlantic led the round, with participation from Seligman Ventures, T. Rowe Price, Capital Group, funds managed by BlackRock, Intel Capital, the Qatar Investment Authority, and Vista Equity Partners, among others. A second close is expected in the coming weeks.
What does SambaNova actually build?
SambaNova designs AI chips and full systems built for inference rather than general purpose GPUs. Its SN40L and newer SN50 systems are meant to run large models efficiently, and the company positions them for on premises and private cloud deployment where an organization wants to keep data and models inside its own walls.
Why is the JPMorgan win a big deal?
JPMorganChase selected SambaNova as an inference infrastructure partner, with its systems set to power secure, on premises AI inference at the bank. A large regulated institution choosing non Nvidia hardware for private inference is a strong signal that alternative silicon has reached production readiness for demanding, compliance heavy workloads.
What does this mean for developers deploying inference?
It means more real choice below the GPU. If your workload is inference heavy and needs to run in your own environment for privacy or compliance reasons, dedicated inference systems are becoming a credible option next to renting GPU clusters. The practical questions to test are model support, tooling maturity, and cost per token at your traffic level.