SambaNova has closed a $1 billion Series F at an $11 billion valuation, betting that inference, the phase where AI models actually answer your queries in production, is where the real hardware market gets decided. General Atlantic led the round, with BlackRock, Intel Capital, T. Rowe Price, and the Qatar Investment Authority participating. The company’s SN40 and SN50 chips are now deployed inside JPMorgan Chase’s own data centers as its primary AI inference infrastructure<\/a>, giving SambaNova its most credible enterprise reference customer to date.<\/p>\n The JPMorgan deployment is the number that matters here, not the valuation. Any chip challenger can raise a billion dollars in this environment, but landing an on-premises inference contract with a bank whose regulators audit every vendor decision is a different kind of proof. If your infrastructure team is currently in a GPU procurement queue, waiting on Nvidia H100 or H200 allocations, this round signals that SambaNova now has the capital to actually fill orders at enterprise scale, which makes them a credible evaluation option rather than a speculative one.<\/p>\n SambaNova’s specific architectural bet deserves scrutiny before you shortlist them. Their chips are purpose-built for inference workloads rather than the general-purpose training that made Nvidia’s CUDA ecosystem dominant. That specialization is genuinely a double-edged position: you get better cost-per-token economics on steady-state query traffic, but you give up the flexibility of a platform your data science team already knows how to configure. The Intel Capital equity stake deepens the software integration story, but Nvidia’s CUDA lock-in, the enormous library of code that only runs on Nvidia hardware, is an accumulated switching cost that a funding round cannot dissolve.<\/p>\n The falsification condition for SambaNova’s thesis is straightforward: if Nvidia accelerates its own inference-optimized product line, specifically the B200 and whatever follows, while expanding supply, the cost-per-token advantage SambaNova is selling compresses fast. What the JPMorgan deal actually reframes for your team is the vendor diversification question you probably already have open. The relevant budget decision isn’t whether to replace Nvidia wholesale; it’s whether your next inference capacity expansion should include a competitive bid, and right now you have a named Wall Street reference customer to benchmark against.<\/p>\nWhat this means for your business<\/h2>\n