Groq's real story isn't another August funding round. It's the cleaner, stranger pivot that followed Nvidia's $20 billion licensing deal.
Groq raised $650 million in June to build out its AI inference cloud, and that verified number tells you more than the unconfirmed $350 million figure attached to this draft. According to Groq's June 22 announcement, the round was led by Disruptive and Infinitum, with the money aimed at expanding a global cloud business that runs AI models after they've been trained. That's inference. It's also where a lot of AI spending is moving now that companies have to serve real users, not just train bigger models for demos.
The company Groq is today is not the company investors valued at $6.9 billion last September. In December, Groq announced a non-exclusive licensing agreement with Nvidia for its inference technology. Founder Jonathan Ross, president Sunny Madra and other Groq team members moved to Nvidia as part of the deal. TechCrunch and Axios both reported the value around $20 billion, while Nvidia said it was not buying Groq as a company.
That distinction matters. Groq kept operating, but the center of gravity moved. The startup that spent years telling customers its LPU chips could beat GPUs on inference speed now has to prove it can run a cloud business after its best-known founder and much of its hardware story shifted into Nvidia's orbit.
That's a hard reset.
Groq is raising $650 million to fund its pivot from AI chip maker to inference neocloud provider, following a $20 billion LPU licensing deal with Nvidia that paid out shareholders and handed the founding team to its former competitor. Existing investors Disruptive and Infinitum are backstopping the full round. The raise signals durable investor... - Groq raises 650 million for AI inference neocloud - AI chip startup licensing technology to Nvidia deal
Groq says it now operates 13 data centers across North America, Europe, the Middle East and Asia Pacific, serving more than five million developers and thousands of AI companies that process trillions of tokens each week. It also says it expects to scale toward 200 megawatts by the end of 2027. Those are not small-company numbers. They are data-center numbers, and data centers eat capital quickly.
Groq is selling capacity now #
The old Groq story was easy to understand. It built LPUs, language processing units designed for fast inference, and pitched them as a sharper tool than general-purpose GPUs for real-time AI responses. Forbes reported last year that Groq's 2023 sales were still tiny beside Nvidia's, but the speed demos gave the company a real place in the AI infrastructure conversation. You didn't have to believe Groq would topple Nvidia to see why customers cared.
The new Groq story is less romantic. It's probably more expensive too: the company is now leaning into GroqCloud, its inference cloud for developers and enterprise customers, and in plain terms, it wants to rent AI computing capacity at scale. CoreWeave built a public company around a related idea, using Nvidia GPUs to serve customers that didn't want to build their own clusters. Nebius is chasing the same market from a different starting point.
Groq's twist is that it still has its own inference identity, even after Nvidia licensed the technology. Groq's June announcement says the new capital will support its latest inference technology, including Nvidia's LPX system that incorporates Groq's technology. That's a neat line for a press release. The business reality is sharper: Groq now has to make money in a market where Nvidia is supplier, partner and gravitational force all at once.
Here's the thing. You can call that partnership, but you shouldn't miss the dependency.
Groq has rebuilt the management bench around that task. The company says Adam Winter is chief executive and Matt Eng is chief financial officer, with Alan Rice joining as chief operating officer after roles at xAI and Meta data centers. Sinclair Schuller is coming in as chief technology officer, and Rakesh Malhotra as chief product officer after earlier work at Microsoft and Nuvalence. That mix points to operations, not chip-show bravado.
The valuation question has not gone away #
The draft's claimed August raise at a $3.5 billion valuation could not be verified through live search, and Groq's own newsroom does not show that announcement. The verified financing is the June $650 million round. Groq also did not disclose a new valuation in that June announcement, while TechCrunch noted the company had last been valued at $6.9 billion after its September $750 million raise.
Nvidia announced a long-term strategic partnership with Ilya Sutskever's Safe Superintelligence Inc. on July 27, 2026, taking an equity stake and committing to supply SSI with Vera Rubin compute systems. The deal gives Nvidia rare access to SSI's research to inform future chip design, extending its circular financing model to its most structurally... - Nvidia investment in Safe Superintelligence - Ilya Sutskever's research lab partnership
That makes the article's original down-round argument too shaky to publish. There may well be a lower implied value inside later private paperwork, but you don't print that without a source you can stand behind. A reader deserves the clean version: Groq raised new money after Nvidia licensed its technology and hired key leaders, but the company did not publicly price the new round.
What you can say is enough. Groq is no longer mainly asking the market to believe in a chip challenger. It's asking customers to believe it can run inference infrastructure at global scale, with 13 data centers today and a 200-megawatt target for 2027. That's a different test. Faster tokens helped Groq get noticed. Power, customers and uptime will decide whether the remaining company matters.
Also read: Amazon Is Cutting the Spines Off Rare Books to Train Its AI Models, Researchers Show How Microsoft Copilot Can Turn One Login Into $247,500 and Micron Launches a $250 Million Fund to Bankroll the AI Startups Buying Its Chips