Intel-Backed SambaNova Raises $1B, Powers JPMorgan AI Inference

What happened

SambaNova closed the first tranche of a $1 billion Series F round led by General Atlantic, with participation from Intel Capital, Vista Equity Partners, JPMorgan, and others. The funding follows a $350 million raise in February and values the company at $11 billion. The company announced JPMorgan as a customer deploying its SN40 and SN50 systems for on-premises AI inference. SambaNova also released benchmark results showing its heterogeneous platform-combining four Nvidia H200 GPUs for prefill operations with 16 SN50 accelerators for decode operations-achieving 763 tokens per second at short context lengths and over 450 tokens per second at longer contexts. This performance significantly exceeds GPU-only inference platforms and demonstrates how customers can extend the lifespan of existing GPU fleets by pairing them with SambaNova's air-cooled accelerators.

Why it matters

For Intel, this development validates its strategic investment in SambaNova and offers a competitive alternative in the AI inference market dominated by Nvidia. Intel announced a collaboration with SambaNova in February, and the startup's success provides Intel with exposure to the rapidly growing inference market without relying solely on its own accelerator products. SambaNova's Reconfigurable Dataflow Units (RDUs) use a specialized architecture optimized for inference workloads, featuring a three-tier memory design that allows rapid switching between hundreds of AI models. The company's ability to attract a major enterprise customer like JPMorgan and demonstrate superior performance in independent benchmarks strengthens Intel's position in the AI infrastructure ecosystem. The startup's air-cooled systems also address a practical challenge for datacenter operators, as they can be deployed in existing facilities without requiring liquid cooling infrastructure-a constraint affecting Nvidia's latest Rubin GPUs.

Bigger picture

The AI inference market is experiencing rapid evolution as the industry shifts focus from model training to deployment at scale. Disaggregating prefill and decode operations has emerged as a key strategy for reducing costs and improving performance, particularly for long-running AI agents like code assistants. Nvidia pioneered this approach with its NVL72 rack systems and later demonstrated it with Groq-based LPX racks at GTC. Now competitors including AMD, AWS, Cerebras, and SambaNova are all introducing heterogeneous inference platforms. SambaNova's approach positions its accelerators as complementary to rather than purely competitive with Nvidia GPUs, potentially appealing to enterprises with existing GPU investments. The company's $11 billion valuation reflects investor confidence in the inference market's growth trajectory, especially as enterprises move AI workloads into production. The banking sector's adoption of on-premises AI inference, demonstrated by JPMorgan's deployment, signals growing enterprise demand for secure, high-performance AI infrastructure that can handle sensitive workloads without relying on cloud providers.

What to watch

Monitor whether SambaNova can scale production of its fifth-generation chips and meet demand from enterprise customers. The company plans to demonstrate configurations with 128 and eventually 256 accelerators to show sustained high token generation rates at scale-a capability that GPU-only systems have historically struggled to achieve. Watch for additional enterprise customer wins, particularly in regulated industries like financial services where on-premises inference is critical. Intel's ongoing collaboration with SambaNova, including the Vector Core Compute deployment with TogetherAI as a large-scale customer, will indicate whether the partnership can gain meaningful market share against Nvidia. Also track whether other chip makers and cloud providers announce similar heterogeneous inference platforms, as the industry appears to be converging on disaggregated architectures. Finally, observe how SambaNova uses its $1 billion in fresh capital to expand manufacturing capacity and whether it can maintain its performance advantages as Nvidia and competitors release next-generation products.