SambaNova is closing in on a fresh funding round of up to $1 billion, and the number that matters most is not the check size. It is the valuation, roughly $10 billion, about five times what investors said the company was worth just four months ago.
Intel CEO Lip-Bu Tan let the news slip at a public event, confirming that SambaNova, the Palo Alto based AI chip maker Intel has both invested in and partnered with, is in talks to raise between $800 million and $1 billion. The Information first reported the target valuation, and outlets including Reuters and Yahoo Finance have since confirmed the roughly $10 billion figure Intel's chief referenced. That's not a modest markup. This is a company that was valued somewhere between $2.2 billion and $4.8 billion in February, depending on how you count the terms of that round, suddenly worth two to four times as much just months later.
You don't see valuations move that fast in most industries. In AI infrastructure right now, you do.
SambaNova's last round closed in February, when it raised $350 million co-led by Vista Equity Partners and Cambium Capital. Intel Capital came in too, with a commitment of up to $150 million, and the round also drew BlackRock, T. Rowe Price, Battery Ventures, GV and Qatar's sovereign fund QIA. That money went toward SambaNova's SN50 chip, an expansion of its SambaCloud platform, and deeper software integrations for enterprise customers. Five months later, the company is apparently worth enough to raise nearly three times as much capital at a valuation multiple that would make most late-stage investors blink.
SambaNova doesn't build GPUs. It builds something called a Reconfigurable Dataflow Unit, a chip architecture that layers SRAM, HBM and DRAM together so a single system can hold far larger models in memory than a typical GPU cluster can. The pitch to customers is that inference, the work of actually running a trained model rather than training it, doesn't need Nvidia's general purpose horsepower. It needs hardware built for the specific math of serving answers fast and cheap at scale.
That argument has gone from niche to central in about a year. In December, Nvidia validated it in the most expensive way possible, agreeing to pay roughly $20 billion to license Groq's inference technology and bring Groq founder Jonathan Ross in house, according to reporting from CNBC and Axios on the deal. Groq had been valued at $6.9 billion three months earlier. Cerebras, another Nvidia challenger built around a dinner plate sized chip packed with 900,000 cores, has filed to go public. SambaNova's own jump toward a $10 billion valuation lands right in the middle of that pattern.
None of this means Nvidia is losing its grip on AI hardware. It still dominates training, and it just wrote the largest check in its history to make sure it doesn't get boxed out of inference either. What it does mean is that investors no longer treat an alternative to Nvidia as a losing bet. They're pricing it as one of the more attractive ones on the table.
SambaNova's customer list looks different from Groq's or Cerebras's. The company has leaned hard into government and national lab deployments rather than chasing the consumer facing speed demos that made Groq's name. Argonne National Laboratory runs a SambaNova cluster to help train AuroraGPT, its foundation model for scientific research across biology, chemistry and climate modeling, and Oak Ridge National Laboratory has deployed the hardware too. SambaNova also markets a sovereign AI pitch to governments that want to run open source models without sending sensitive data to an outside cloud. That's a narrower customer base than a general purpose inference platform, but it's also one with fewer rivals fighting for the same accounts, and contracts that don't evaporate the moment a flashier chip ships.
Frankly, a fivefold valuation jump says less about SambaNova's fundamentals changing overnight and more about how scarce it's become to find a real, working alternative to Nvidia at any size. There are only so many companies with functioning silicon, paying customers and the balance sheet to keep building at the pace this market now demands. SambaNova has spent nine years getting here since its 2017 founding, through a $5.1 billion valuation on SoftBank Vision Fund money in 2021, a rough stretch in which Intel reportedly explored buying the company outright for around $1.6 billion before those talks fell apart, and now this.
The round hadn't closed as of early July, and terms can still shift before anyone signs a check. But the direction is clear enough. Money chasing alternatives to Nvidia isn't slowing down, and SambaNova just became the latest and largest proof of how much of it is out there.
Also read: Anthropic Leases a Hudson Square Tower Even as It Rations Its Own AI Model • China Will Debut the World's First AI Agent Phone at Its Shanghai AI Summit • A Former Tesla Optimus Engineer Is Betting Europe Can Build Its Own Robots
