SambaNova SystemsThe Story
AI chip company SambaNova Systems has raised $1 billion at an $11 billion post-money valuation in the first close of its Series F round, led by General Atlantic. Seligman Ventures, T. Rowe Price Associates and Capital Group made what the company described as significant investments. The financing was announced on 8 July 2026. The round is not finished. SambaNova has not disclosed the target size of the full Series F, how much General Atlantic put in, how the rest divides between participants, whether any portion is secondary, or what governance came with the lead cheque. Co-founder and chief executive Rodrigo Liang said at the time that more investors would join within weeks and that a second close was likely to complete. No such completion has been announced since. The company has disclosed no revenue figure alongside the valuation. New and existing investors listed in the first close include A&E Investment, Assam Ventures, Battery Ventures, funds managed by BlackRock, Cambium Capital, Intel Capital, Kabila Capital, QFO Capital, the Qatar Investment Authority, Vista Equity Partners and Volantis. The raise lands five months after a $350 million Series E led by Vista Equity Partners and Cambium Capital in February 2026, which accompanied the unveiling of the SN50 chip. Before that, Bloomberg reported in December 2025 that Intel was close to acquiring SambaNova for roughly $1.6 billion, and reported in January that the company was seeking up to $500 million after those talks stalled. SambaNova was valued at $5.1 billion at its Series D in 2021. Market trackers put total funding at about $2.5 billion across seven rounds. Alongside the financing, SambaNova said JPMorganChase had selected it as an inference-infrastructure partner, deploying SN40L and SN50 systems for on-premises inference. Liang said the money would go towards scaling the business and securing supply chain and materials for the next twelve months.
Why It Matters
Demand for AI inference is running ahead of the hardware available to serve it, and the shortage is not evenly distributed. Frontier labs and large clouds hold the allocation. A bank, an oil company or a government department that wants to run a large model on its own data, inside its own building, has had fewer options and has often been told to use somebody else's cloud instead. SambaNova, founded in 2017 and based in Palo Alto, sells to exactly that buyer. Its reconfigurable dataflow processors are built to hold very large models in a single rack, which Liang describes as premium inference: the biggest models, run fast, without splitting them across a sprawl of machines. The company sells complete systems rather than chips alone, on-premises or through its own cloud, and names sovereign cloud programmes, neoclouds and enterprises running their own workloads as its three customer types. That is a capital-heavy way to make money. Fabrication, memory supply and systems integration all have to be paid for long before a customer takes delivery, and the company was explicit about where the new money goes: securing the supply chain and buying materials to fulfil orders over the coming twelve months. A billion dollars raised partly to pay for inventory is a statement about the business model as much as about demand. What has not been disclosed is the size of any of it. SambaNova has published no revenue figure, no order book and no contract value for the JPMorganChase deployment. The SN50, announced in February, is still due to start shipping in the second half of this year.
The Strategic Read
The market assumption changing behind this investment is that inference, not training, is where an architecture other than Nvidia's can win, and that regulated buyers who cannot put their models in a public cloud will pay a premium for silicon they can install themselves. The previous generation of Nvidia challengers went at training. Graphcore, Habana and others built accelerators for the model-building workload, discovered that CUDA and supply allocation mattered more than any benchmark, and were mostly absorbed or repositioned. SambaNova came close to that ending. Seven months ago it was reportedly being bought by Intel for about $1.6 billion, which for a company that had raised well over a billion dollars would have returned very little to anyone. What has moved since is not obviously the product. The SN50 was unveiled in February and has not started shipping. What moved is the price of inference capacity, the arrival of buyers with names, and the willingness of growth investors to underwrite an alternative supplier while Nvidia allocation stays tight. The company went from a reported $1.6 billion takeover mark in December to a reported $2.2 billion at the February Series E to $11 billion in July. That is roughly five times in five months and seven times the Intel figure in seven, on a product cycle that did not move at anything like that speed. The repricing is being done by capital, not by shipments. For that mark to hold, the demand has to be reproducible rather than anecdotal. JPMorganChase is a strong reference, and a bank building private inference infrastructure is a genuine signal to its own industry. It is also one contract of undisclosed size, in a customer base that includes Saudi Aramco and sovereign programmes, which are lumpy, politically exposed and slow to repeat. No revenue figure has been given at any point in this sequence. The structural risk is the shape of the round itself. An $11 billion post-money on a first close is a mark set before the round is full. If the second close takes longer than the few weeks indicated, or arrives on different terms, that becomes the story. Meanwhile Intel sits on the register as investor, co-development partner and former suitor at one-seventh of the current price, which is a comfortable position for Intel and a complicated one for everybody else on the cap table.
For daily, sharp analysis of the biggest moves in the Indian business and startup ecosystem, follow StartupFox.
