SambaNova raised $1B to keep AI inference off the cloud
SambaNova raised $1B at an $11B valuation for on-premises AI inference chips, with JPMorgan as a customer. Why owning the inference layer is now a bank-grade bet.
The market just put a billion dollars behind the idea that you shouldn't run all your AI on someone else's cloud. SambaNova raised $1 billion at an $11 billion valuation — a Series F first close led by General Atlantic — to build chips that serve large language models on-premises. The tell isn't the number. It's the customer: JPMorgan Chase signed on as an "inference-infrastructure partner" to run Claude-class workloads inside its own walls.
What actually happened
Per TechCrunch, SambaNova closed the round on July 8, roughly five months after a $350M Series E. The company makes inference chips — its SN40L today and the next-gen SN50, unveiled in February and shipping in the second half of 2026 — built specifically to run trained models in production, not to train them. JPMorgan will deploy both to "power secure, on-premises AI inference," per the company.
SambaNova's CEO said the quiet part out loud, per CNBC: "It sends a message to the banking industry that it's time not to completely depend on cloud services. These banks want heterogeneous infrastructure." Translation — the most compliance-obsessed buyers on earth are deciding that renting 100% of their AI from a hyperscaler is a risk, not a convenience.
Why it matters for your business
You are not going to buy a rack of AI accelerators. That's not the point. The point is that the largest, most risk-averse enterprises are now paying real money to keep the inference layer — the part that actually answers your prompts and touches your data — on infrastructure they control. When JPMorgan does something for compliance reasons, it's usually a preview of what everyone else gets told to do two years later.
The portable version of this for an operator: don't wire your business to a single cloud AI endpoint you can't move. Route inference behind an interface you own, so the model and the box underneath it are swappable — cloud today, a cheaper provider tomorrow, on-prem the day a client's contract demands it. The moat isn't the chip. It's not being trapped when the bill, the latency, or the data-residency clause changes on you.
Key takeaways
- SambaNova raised $1B at an $11B valuation (Series F first close, led by General Atlantic) on July 8, 2026
- Its SN40L and SN50 chips are built for on-premises inference — serving models in production, not training them
- JPMorgan Chase signed as an "inference-infrastructure partner" to run secure AI inside its own walls
- The signal for operators: keep the inference layer swappable and, where it matters, on infrastructure you control
Depending on one cloud AI endpoint for work that touches real data? We build vendor-agnostic systems where the model and the infrastructure underneath it are swappable — so a price hike or a data-residency clause is a config change, not a rebuild. See what we build or tell us your stack.
Sources: TechCrunch, CNBC.
- #ai-inference
- #sambanova
- #on-premise
- #vendor-lock-in
- #portability
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Squidbleed: a 29-year-old Squid proxy bug leaks HTTP creds
Squidbleed (CVE-2026-47729) is a Heartbleed-style memory leak in Squid proxy that survived 29 years. Here's who's exposed and what to do.
Read itBaseten's $1.5B says inference is the layer to own
Baseten raised $1.5B at up to $13B for AI inference infrastructure. Why the boring serving layer — not the model — decides your cost and uptime.
Read it