SambaNova Systems, a Palo Alto-based startup designing custom AI chips, has raised $1 billion in a late-stage funding round that values the company at $11 billion. The round, led by General Atlantic, signals growing investor confidence that the AI semiconductor market is large enough to support serious challengers to Nvidia's dominance.

The funding brings SambaNova's total capital raised to more than $4.5 billion since its founding in 2017 by former Stanford University researchers. The company designs purpose-built chips optimized for AI inference — the process of running trained models to generate outputs — rather than the training phase that has historically consumed the majority of AI computing resources.

That strategic focus on inference is increasingly looking prescient. As AI models become widely deployed across enterprise applications, healthcare systems, financial services, and consumer products, the demand for efficient inference computing is growing far faster than the demand for training compute. Analysts estimate that inference workloads will account for more than 60% of total AI chip spending by 2028, up from roughly 40% today.

"Training a model is a one-time cost. Running that model millions of times a day is where the real economics kick in," said Rodrigo Liang, SambaNova's co-founder and chief executive. "Our architecture was built from the ground up for that reality."

SambaNova's chips use a reconfigurable dataflow architecture that the company claims delivers significantly better performance per watt than Nvidia's GPUs for inference workloads. The company's latest SN40L chip, released earlier this year, features 1,020 billion transistors and supports models with up to 5 trillion parameters — a scale that would be impractical on conventional GPU clusters without extensive sharding and networking overhead.

The company has quietly built a customer base that includes several Fortune 500 companies, national laboratories, and government agencies. Its technology powers AI workloads at the U.S. Department of Energy and has been adopted by financial institutions for real-time fraud detection and risk modeling.

The investment comes amid a broader boom in AI chip startups. Cerebras, which makes wafer-scale processors, filed for an IPO earlier this year. Groq, founded by a former Google engineer, has gained attention for its ultra-fast inference speeds. And in China, companies like Huawei and Cambricon are developing domestic alternatives in response to U.S. export controls.

For General Atlantic, the bet on SambaNova is part of a broader thesis that the AI infrastructure market is entering a second phase. "The first wave was about training," said Tanzeen Syed, a managing director at the firm. "The second wave is about deployment at scale, and that requires fundamentally different hardware."

SambaNova said it will use the new capital to accelerate production of its next-generation chips, expand its sales team, and build out a cloud-based inference service that will compete directly with offerings from Amazon Web Services and Google Cloud.

Image source: i.ibb.co