SambaNova closed a $1 billion funding round this week that values the AI chip startup at $11 billion, according to CNBC. General Atlantic led the round, with Seligman Ventures, T. Rowe Price and Capital Group also participating. The money is going toward one specific bottleneck: building enough server racks to meet customer demand for inference hardware.
“Inference has broken everything open,” co-founder and CEO Rodrigo Liang told CNBC at the Raise AI summit in Paris, framing the company’s standalone status as an advantage for moving fast across sectors. “The capital allows us to really accelerate the deployments of the racks that customers really want,” he said.
The round follows more than $350 million SambaNova raised earlier this year from investors including Intel, with which it also announced a partnership. Going from a $350 million raise to a $1 billion round in the same year signals that demand, not just investor appetite, has accelerated.
SambaNova’s pitch rests on a distinction that matters to the chip market’s structure. Nvidia built its dominance on GPUs optimized for training, the compute-heavy process of building a model from scratch. SambaNova sells its latest chip, the SN50, as part of a server unit built for inference: running an already-trained model in production, quickly and at lower cost per query. As AI agents multiply and every one of them needs to call a model repeatedly, inference workloads are growing faster than training workloads at most enterprises. That is the wedge Nvidia’s challengers are aiming at, since inference is a market defined by cost and latency rather than raw training throughput, giving alternative architectures room to compete on economics rather than beating Nvidia’s training performance outright.
SambaNova is also selling into a segment Nvidia has been slower to prioritize: on-premise deployment. JPMorgan Chase said this week it will deploy SambaNova’s systems for on-prem inference across what the bank called its “demanding enterprise AI workloads.” Liang told CNBC that SambaNova will act as the bank’s inference provider, arguing that keeping models and data inside a company’s own firewall matters most for industries like banking where data sensitivity is high.
Public markets have already priced in enthusiasm for the chip sector broadly. The PHLX semiconductor index is up roughly 80% this year, and private funding is following the same trend. Rebellions, a South Korean chip startup, is preparing an IPO on the Kospi exchange for the first or second quarter of 2027, CEO Sunghyun Park told CNBC. Nvidia itself struck a licensing deal with inference chip startup Groq last year, an acknowledgment that inference-specific architectures are not going away.
Liang said SambaNova is strongly considering an IPO in 2027, most likely in the United States. Pairing that timeline with Rebellions’ points to 2027 becoming the year inference-chip challengers test whether public investors will value them independently of Nvidia’s multiple. For enterprise buyers evaluating inference infrastructure now, SambaNova’s JPMorgan deal is the reference case to watch: if on-premise inference performance holds up at bank scale, it becomes the strongest argument for choosing a non-Nvidia stack for production AI workloads in 2027 budget cycles.
Reporting based on CNBC (Arjun Kharpal), published July 8, 2026.
