SambaNova Closes $1 Billion Series F as JPMorgan Bets on On-Prem AI Chips

The AI infrastructure story has long been told as a cloud story — hyperscalers, GPU clusters, Nvidia's dominance. SambaNova's $1 billion Series F, and the JPMorgan deal that anchors it, suggest that framing is aging fast.

The Round and What It Signals

SambaNova Systems has closed a Series F at a $1 billion raise, pushing its valuation to $11 billion. General Atlantic led the round, joined by existing and new institutional investors. The capital is earmarked for deployment and continued development of SambaNova's SN40 and SN50 AI accelerator chips, both targeting enterprise inference workloads.

The size of the round matters less than the timing and the customer attached to it. A billion-dollar raise at this valuation is a statement that institutional money believes the GPU monoculture in enterprise AI has a credible challenger — and that challenger has found the right vertical to prove it.

JPMorgan and the On-Prem Argument

JPMorgan has agreed to deploy SambaNova's SN40 and SN50 chips for on-premises enterprise AI inference. That last word — on-premises — is the operative one. JPMorgan isn't running these workloads through a public cloud. It's bringing the hardware inside its own walls.

For a financial institution operating under layers of data residency, compliance, and security requirements, that distinction isn't cosmetic. Public cloud GPU access, however scalable, introduces surface area that regulated industries have to justify to auditors and regulators. On-prem inference hardware sidesteps that conversation entirely.

SambaNova has positioned its systems as a direct alternative to leading GPU vendors for large-scale inference, with financial services and other regulated industries as the primary addressable market. JPMorgan's commitment validates that positioning in the most concrete way available: a production deployment at one of the most scrutinized financial institutions on earth.

The Non-Nvidia Wedge

The deal and the funding round together underscore something that's been building quietly: demand for non-Nvidia AI hardware is real, and it's concentrating in exactly the environments where Nvidia's cloud-first ecosystem is hardest to operate — high-security, compliance-heavy, data-sovereign deployments.

This isn't the first time alternative AI chip vendors have claimed traction. What makes SambaNova's position distinct is the specificity of the use case. Inference — running trained models at scale in production — is where enterprise compute budgets are increasingly landing. Training runs are one-time or periodic; inference is continuous, and its cost compounds daily. A chip optimized for that workload, deployed inside a bank's own data center, addresses a genuine operational pain point rather than a theoretical one.

General Atlantic's decision to lead the Series F reflects a calculation that this wedge is durable. Regulated industries — finance, healthcare, defense-adjacent — represent enormous inference budgets, and they're structurally incentivized to diversify away from any single vendor, cloud provider, or chip architecture.

What Comes Next

The SN40 and SN50 chips now have a marquee reference customer and fresh capital to fund further deployment. The immediate question for SambaNova is whether JPMorgan becomes the first of many regulated-industry deployments or remains a high-profile exception.

The broader shift this round represents is harder to reverse than any single deal. Enterprise buyers in high-security environments have been watching for proof that alternatives to Nvidia and public cloud GPU access could perform at scale in production. A signed deployment at JPMorgan, backed by $1 billion in institutional capital, is as close to that proof as the market has seen. The center of gravity in AI infrastructure — where the chips live, who controls them, and what architecture runs inference — is moving. SambaNova just moved it a little further.