SambaNova raised $1 billion at $11 billion valuation on surge in AI inference demand

SambaNova Systems, a maker of specialized AI chips, has raised $1 billion in funding at a $11 billion valuation — a validation of investor confidence in growing demand for AI inference infrastructure. SambaNova's CEO Rodrigo Lian shared details of the round in an interview with Ed Ludlow on Bloomberg Tech.

What we know about the round

SambaNova is a Palo Alto startup founded in 2017 by Stanford alumni, including professor Kunle Olukotun. The company is developing its own Reconfigurable Dataflow Unit (RDU) chip architecture — an alternative to Nvidia's graphics processors, which currently dominate neural network training and inference markets. The idea behind RDU is that data "flows" through the chip along the optimal route for a specific neural network, rather than being processed through a universal GPU architecture — the company claims this reduces latency and power consumption during large model inference.

• Round size — $1 billion in funding
• Post-round company valuation — $11 billion
• CEO — Rodrigo Lian
• Interview aired on Bloomberg Tech with Ed Ludlow

Why inference has become the main demand driver

Lian emphasized that demand for high-speed inference — the quick and cost-effective deployment of already-trained models in real products — is growing at enormous rates. While a few years ago most industry capital investment went into training large models, companies are now widely deploying AI into their services and need infrastructure capable of responding to users with minimal latency at an acceptable cost per token.

This shift in demand is linked to the fact that instead of one-off queries to a chatbot, businesses increasingly deploy AI agents and copilots that call models repeatedly within a single task — from document analysis to workflow automation. Each such call requires model response in fractions of a second, and this is what makes inference cost and speed, not just model quality, the deciding factor when choosing infrastructure.

Who is competing for the inference market

This shift in focus — from training to inference — has become one of the major themes of 2026 for the entire chip industry. Besides Nvidia, which still holds the dominant share of the AI accelerator market thanks to its CUDA ecosystem, hyperscalers are bringing their own chips to market — Google's TPU, Amazon's Trainium and Inferentia, as well as chip startups Groq and Cerebras competing with SambaNova for enterprise clients needing an alternative to expensive GPU clusters. For such clients, the key metrics are not peak performance on paper, but the real cost per token of model response and latency under high load.

What this means

SambaNova's large round against the backdrop of the overall boom in AI infrastructure investment shows: the market is increasingly betting not only on developers of models themselves, but on companies building the "hardware" for their cheap and fast execution — this is increasingly what defines the real economics of scaling AI in products, not just the quality of neural networks themselves.