NVIDIA-backed Groq raises $350M at $3.5B as AI inference race accelerates. August 18, 2026 * Groq raised $350 million at a $3.5 billion valuation, half its September 2025 mark. * Nvidia is set to invest in Groq, months after hiring away its founder and engineering team. * The round brings Groq's funding to $1 billion since June 2026, as inference demand grows. In September 2025, Groq was an independent chip company valued at $6.9 billion, built to challenge Nvidia's grip on AI hardware. As of August 2026, it is worth half that amount, and the investing company is Nvidia itself. Groq has closed a $350 million Series A at a $3.5 billion valuation, led by Disruptive, with Nvidia expected to participate. It follows the $650 million Groq raised in June 2026, bringing its recent fundraising to $1 billion in roughly two months. "We are building Groq into the world's leading AI inference cloud. Inference will without a doubt become the largest and most critical layer of AI infrastructure," said Alex Davis, Groq's executive chairman and founder of Disruptive. This $350 million round, and the valuation cut that comes with it, is the market pricing in what was lost and what remains. On one side: 13 data centres, more than six million developers, and a plan to scale capacity from 54MW to more than 200MW by 2027. On the other hand, not the founder, and not the team that built the technology in the first place. A company rebuilt around the deal that broke it. Jonathan Ross, a former Google engineer who helped build the company's Tensor Processing Unit, founded Groq in 2016. For nearly a decade, it positioned its Language Processing Units as a faster, cheaper alternative to Nvidia's GPUs. That bet looked vindicated in September 2025, when Groq raised $750 million at a $6.9 billion valuation, led by Disruptive, with BlackRock, Neuberger Berman, and Deutsche Telekom Capital Partners joining in. In December 2025, Nvidia agreed to pay a reported $20 billion to license Groq's inference technology and hire Ross, president Sunny Madra, and roughly 90% of the engineering team that built it. Groq kept operating, minus the people who invented its core product. Inference is becoming the bigger AI trend. Groq's pitch now rests on a single argument: that running AI models in production will eventually dwarf the scale and spending of training them. It is not alone in making that case, and it is no longer the best-funded name making it. That leaves Groq competing less on technology, since its edge now sits inside Nvidia, and more on operational scale and its existing developer base. It remains an Nvidia Cloud Partner, certified to run Nvidia's accelerated computing under Nvidia's own reference architecture. That makes the relationship less a rivalry than a supply agreement with an unusually large investor attached, a pattern Tech Funding News has tracked across other chip challengers Nvidia has absorbed rather than fought. The unresolved question is whether Groq can rebuild a defensible position after licensing away the technology that made it distinctive. The answer will show whether Nvidia's capital turns Groq into a genuine platform, or simply a well-capitalised, dependent supplier of the compute Nvidia itself sells.

Nvidia paid $20B for Groq's brain. The body just raised a Series A. August 18, 2026 · 05:11 UTC · News Tl;dr. Groq, the company that spent ten years building LPU inference chips to unseat Nvidia, closed a $350 million round on August 17 that it is calling, with a straight face, a Series A. The round values Groq at $3.5 billion, roughly half the $6.9 billion investors paid last September, and it funds the completion of a total pivot: Groq no longer sells chips. It is now a "neocloud" that rents out inference capacity running on Nvidia hardware, certified as an NVIDIA Cloud Partner on August 12. The lead investor is Disruptive; Nvidia itself has planned participation in the round. That would be the same Nvidia that paid about $20 billion in December for Groq's chip technology and most of its senior leadership. A Series A, ten years in. Groq was founded in 2016 by Jonathan Ross, one of the creators of Google's TPU. Its pitch was the Language Processing Unit: deterministic, SRAM-heavy silicon that served tokens faster than GPUs and made "Groq speed" a real brand among developers. By September 2025 that story was worth a $750 million raise at a $6.9 billion valuation. Eleven months later the same company is closing a "Series A," the round name startups use when they have a product and a pulse. The label is doing deliberate work: it marks the old Groq as over and reboots the cap table narrative around a new business. Per TechCrunch, the round was led by Disruptive, the Dallas firm whose founder Alex Davis now sits as Groq's executive chairman, with Nvidia planning to participate. Combined with June's $650 million raise, Groq has pulled in about $1 billion in two months to fund the rebuild. How to sell a company without selling it. To read this round you need December's deal. On December 24, 2025, Nvidia agreed to pay about $20 billion in cash for Groq's chip assets, the largest deal in Nvidia's history, nearly triple the Mellanox record. Groq framed it as a "non-exclusive licensing agreement." Ross, president Sunny Madra, and other senior leaders joined Nvidia; Groq continued as an "independent company." The structure matters because a merger gets reviewed and a licensing deal mostly does not. It is a bit like selling your house, handing over the keys, and moving out, while the paperwork describes a non-exclusive arrangement to share the kitchen: everything a sale accomplishes has been accomplished, but nothing that triggers a sale's scrutiny has technically occurred. One analyst told CNBC at the time that the deal was structured to keep the "fiction of competition alive." Lawmakers noticed too; the deal drew congressional attention within weeks. What was left behind was GroqCloud, 13 data centers, and a developer base, minus the founder, the president, and the roadmap. Today's round is the market pricing that remainder: $3.5 billion, half of September. From LPU to landlord. The June round started the rebuild: a re-staffed executive bench (COO Alan Rice from xAI and Meta, CTO Sinclair Schuller, CPO Rakesh Malhotra, per TechCrunch's June reporting) and a new identity as an inference neocloud. The August 12 NVIDIA Cloud Partner certification made the hardware direction official: Groq's growth capacity is Nvidia accelerated computing, deployed in Groq's own data centers, sold by the token. The scale, per Groq's own releases: 13 data centers across North America, Europe, the Middle East, and Asia Pacific, more than 6 million developers on GroqCloud, trillions of tokens served weekly, and 54 megawatts of capacity today with a target of more than 200 megawatts in 2027. The LPUs have not vanished; Groq still calls itself "the only team in the world with hands-on experience operating LPUs in production at scale." But the fleet it is raising money to build is green, not orange. Nvidia gets paid three times. Follow Nvidia's position through this arc. It bought the LPU technology and the team that could have commoditized inference, for $20 billion. It now sells Groq the GPUs for a 200-megawatt buildout. And it holds planned equity in the reborn company, bought at half of last year's price. Whichever way inference economics break, Nvidia collects: on the IP, on the hardware invoice, and on the equity. That is not a conspiracy, just excellent positioning, but it is worth naming that the "Nvidia challenger" category lost its most credible silicon entrant and gained another customer. For builders, the practical read is mixed. GroqCloud keeps running, and a funded, expanding host with 6 million developers is better than a zombie. But the thing that made Groq interesting was differentiated silicon with real latency numbers behind it. A neocloud running the same Nvidia racks as CoreWeave, Lambda, and a dozen sovereign clouds competes on price, ops, and contracts instead. The down round is investors saying that difference was worth about $3.4 billion. The caveats. Groq's financials are private; no revenue figures accompany the round. Nvidia's participation is described as "planned" in Groq's own release, not closed. The 200-megawatt figure is a 2027 target, not capacity that exists. And the valuation comparison carries an asterisk in Groq's favor: the September 2025 investors were pricing a chip company whose technology later fetched $20 billion from Nvidia, so "half price" describes the leftover business, not a simple markdown of the same asset. Key takeaways. * Groq closed a $350 million round on August 17 at a $3.5 billion valuation, led by Disruptive with planned participation from Nvidia, and branded it a Series A ten years after founding. * The valuation is roughly half the $6.9 billion Groq commanded in September 2025, before Nvidia's $20 billion December deal took its chip assets, founder Jonathan Ross, and senior leadership. * Groq no longer sells chips: it is now an inference neocloud, certified as an NVIDIA Cloud Partner on August 12, deploying Nvidia hardware in its 13 data centers. * The buildout targets growth from 54 megawatts today to more than 200 megawatts in 2027, serving 6 million developers and trillions of tokens weekly on GroqCloud. * Nvidia ends up positioned on every side of the trade: it owns the LPU technology, supplies the GPUs for the expansion, and plans to hold equity in the new Groq. * The December deal's "non-exclusive licensing" structure, which an analyst said kept the "fiction of competition alive," avoided merger review while removing Nvidia's most credible inference-silicon rival. AI Groq Nvidia Inference Funding GPUs Neocloud Hardware AI Bacon Weekly Liked this? Get smarter about AI, weekly. The model launches that matter, the local-LLM tricks worth stealing, and the tools actually shipping. Hand-curated into one email a week, cut down to signal. No hype, ever. Frontier models Local LLMs Builder tools Zero hype Free and curated. One tap to unsubscribe, always. Error: Domain verification failed (missing Origin header).