SambaNova Systems, the AI chip startup that bet early on an alternative to NVIDIA’s architecture, has raised $1 billion in a funding round that values the company at $11 billion, according to Reuters. New and existing investors participated in the round, which SambaNova said will fund the expansion of production of its RDU, or Reconfigurable Dataflow Unit, chips.
SambaNova builds chips and complete systems designed to run AI models rather than train them. Its architecture moves data through a configurable grid of processing elements, an approach known as dataflow, instead of the GPU design that NVIDIA has dominated. The company argues that for running models in production, its design uses memory and power more efficiently than a general-purpose GPU, which matters to enterprises paying for every hour of inference.
The funding round is the largest in the company’s history. SambaNova had raised about $676 million through its 2021 Series D, which valued it at $5.1 billion, from investors including SoftBank, Temasek and BlackRock. The new round more than doubles that valuation and gives the company the balance sheet to build out manufacturing capacity and sales teams at a moment when demand for AI infrastructure is running ahead of supply.
The timing reflects the shape of the AI market. NVIDIA dominates the training of large models, the process of teaching a neural network, but the economics of running those models, known as inference, are still being decided. As models get cheaper to run and enterprises deploy them in their own data centers, the market for inference hardware is growing faster than the market for training hardware, and challengers see an opening. Chip designers in the SambaNova camp argue the moment is the equivalent of the x86 era in computing: a standard architecture that opens a market beyond the incumbent.
The field is crowded. Groq has built its own inference chips around a streaming architecture, Cerebras sells wafer-scale processors, and a wave of startups including Etched, Positron and MatX are designing chips for specific model types. The cloud giants are also in the game: Google has its TPU line, Amazon makes its Trainium and Inferentia chips, and Microsoft has been developing its Maia accelerators. SambaNova’s differentiation is its full stack, a combination of chips, software and turnkey systems that the company says lets customers deploy models without hiring a team of hardware engineers.
The competitive stakes are real. NVIDIA’s advantage is not just the chip but the CUDA software ecosystem, which has locked in developers and made switching costly. Challengers must offer not only better hardware economics but also a software story that developers will accept. SambaNova has tried to meet that test by supporting the standard model formats and frameworks that enterprises already use, and by selling systems that promise faster deployment than a build-your-own approach.
The round is also a statement about the inference market’s trajectory. Investors who sat out earlier chip funding waves are now competing for positions in companies that can plausibly claim a share of the money flowing into AI data centers. Analysts said the $11 billion valuation, while far below NVIDIA’s, puts SambaNova in the top tier of AI chip startups and gives it the resources to survive a market where the cost of manufacturing is enormous.
SambaNova’s commercial record has been a mix of promise and patience. The company has won deployments with large enterprises, and it says its systems are running production workloads for customers in industries from telecommunications to financial services, though it has disclosed little about the scale of those deployments. The new funding will let it expand the sales force that has been a bottleneck for many chip startups, which routinely discover that winning a customer takes longer than building the silicon.
The round also signals something about the market for AI infrastructure. Capital is still flowing to hardware companies even as some investors worry about a bubble in AI spending, and the ability to raise $1 billion in a single round suggests that conviction has not cracked. The question is whether the spending on chips translates into profitable businesses, and SambaNova’s valuation will be judged against the revenue it can show in the next few years.
The round also deepens the company’s bench. SambaNova said it will use part of the proceeds to hire engineers and expand its software team, which customers say has been a deciding factor in deployment decisions. The company has also said it will invest in the tooling that lets developers move models onto its hardware without rewriting them, the kind of work that determines whether a chip architecture survives contact with a real customer.
The risks have not gone away. Model architectures change quickly, and a chip optimized for today’s models can look wrong in two years. NVIDIA is not standing still, and its next-generation platforms will arrive with the full weight of its ecosystem behind them. SambaNova’s customers will judge it on total cost of ownership, reliability and the speed of deployment, not on benchmark charts. With a $1 billion war chest, the company now has the cash to build chips, and the clock to prove they matter.
