SambaNova raises $1 billion at an $11 billion valuation, with JPMorgan as proof

SambaNova Systems closed the first tranche of a $1 billion Series F round on July 8, landing an $11 billion post-money valuation less than five months after its $350 million Series E. General Atlantic led the round, with Seligman Ventures, T. Rowe Price, Capital Group, BlackRock-managed funds, Intel Capital, Qatar Investment Authority, and Vista Equity Partners all participating. The company says more investors are expected to join a second close in the coming weeks.

The valuation jump matters less than who the money is chasing. SambaNova is one of a small group of companies betting that AI inference, not training, is where the real hardware differentiation opens up, and the funding round arrived paired with a customer announcement that makes that bet concrete.

First close of $1B financing at an $11B valuation, led by @GeneralAtlantic.@jpmorgan deploying our RDUs (SN40 + SN50) for secure, on-prem AI inference.

So proud of our team and everyone who helped get us here xf0x9fxa6xbe

Read the press release: https://t.co/VRXyNsY2yl pic.twitter.com/9zqq0TKQip

— SambaNova (@SambaNovaAI) July 8, 2026

What is actually new

SambaNova builds its own chips around a “reconfigurable dataflow” architecture rather than licensing Nvidia’s GPU stack, and sells full systems, the SN40 and SN50, aimed at running inference on-premises rather than through a cloud API. JPMorgan Chase is deploying those systems for what the bank frames as secure, on-prem AI inference, a notable choice for an institution that could otherwise rent GPU capacity from any major cloud.

That customer detail is the story underneath the funding number. SambaNova is not pitching itself as a cheaper GPU alternative for training frontier models. It is pitching itself as the hardware layer for organizations that want AI inference running inside their own walls, for compliance, latency, or data-sovereignty reasons that a shared cloud instance cannot fully satisfy.

What this means for infrastructure buyers

Regulated industries, banks, healthcare systems, government contractors, are the natural buyers for on-prem inference hardware, and JPMorgan choosing SambaNova over a build-it-yourself GPU cluster or a cloud provider’s on-prem appliance is a signal worth watching for any enterprise weighing the same trade-off. The pitch is narrower than “cheaper than Nvidia,” and that narrowness is likely why it is working: a specialized architecture for a specific compliance-driven buyer is an easier sell than a general-purpose alternative competing on raw performance.

It also adds a data point to a trend we have tracked across the AI capital stack: money is flowing into infrastructure companies that let buyers avoid dependency on any single hyperscaler or model provider. Crusoe’s reported $3 billion round and Together AI’s $800 million raise are both bets on the same underlying anxiety: that being fully dependent on one lab’s infrastructure is itself a business risk worth paying to avoid.

What to watch

Watch for the second close and who joins it, since a wider investor base would suggest this is becoming a category rather than a single company’s story. Watch too for whether other regulated-industry buyers follow JPMorgan’s lead, since one large bank choosing on-prem inference hardware is a data point, and three or four would be a procurement trend.