Groq raises $650 million to rebuild after the Nvidia deal

Published: 2026-06-24T09:27:39.000Z

Most $650 million rounds are about growth, but Groq's latest is stranger, a reinvention of a company effectively gutted by a deal of its own making. The AI chip and cloud firm raised $650 million just six months after Nvidia paid roughly $20 billion to license its technology and hire away its founders.

Most $650 million funding rounds are about growth, but Groq's latest is something stranger, a reinvention of a company that was effectively gutted by a deal of its own making. The AI chip and cloud infrastructure firm has confirmed it raised $650 million, led by existing backers Disruptive and the hedge fund Infinitum, to expand its data center capacity and complete a pivot from chip maker to AI computing provider. What makes the raise unusual is the context, since it lands just six months after Nvidia paid roughly $20 billion to license Groq's core technology and hire away its founder and much of its senior team. The investors funding this round are effectively backing a second company built from what was left behind. The backstory is one of the odder episodes the sector has produced. In December 2025, Nvidia struck what was technically a non exclusive licensing agreement rather than an acquisition, paying around $20 billion for Groq's inference technology while simultaneously hiring founder and chief executive Jonathan Ross, president Sunny Madra and a chunk of the engineering bench. Nvidia then folded that intellectual property into its own LPX inference hardware, which it shipped in March 2026. The deal was a windfall for Groq's venture backers, who got paid out in cash, and those same investors are now being asked to reinvest in what some are calling Groq 2.0. Disruptive and Infinitum agreed to backstop the entire round if other shareholders declined their share, which is partly why the money was close to guaranteed. What Groq is selling now is the business that the Nvidia deal did not absorb, namely its inference cloud. Groq designed a chip called the LPU, short for language processing unit, that is purpose built for AI inference, the processing that happens after a user prompts a model rather than the heavier work of training it. Inference is currently a far larger and faster growing need than training, and Groq operates a cloud platform that lets developers and enterprises run those workloads. The company says that platform now processes trillions of tokens weekly for more than five million developers across 13 data centers on multiple continents. The fresh capital will fund a scale up toward 200 megawatts of capacity by the end of 2027, with some of that power drawn, ironically, from Nvidia's new LPX systems that incorporate Groq's own designs. The company has also rebuilt its leadership from the ground up. With Ross and Madra gone, finance chief Simone Edwards stepped up, and Groq has named Alan Rice, formerly of xAI and Meta, as chief operating officer, with a new chief technology officer and chief product officer joining shortly. The challenge ahead is steep, since Groq must convince customers that GroqCloud is a serious inference provider while competing against Cerebras, SambaNova and the hyper-scalers, all without the founding team that built its reputation. The regional read is unusually direct. Groq has deep ties to the Gulf, having previously struck a major agreement with Saudi Arabia's HUMAIN to build inference infrastructure in the kingdom, and its data center network already spans the Middle East. As Saudi Arabia and the UAE pour capital into becoming global AI compute hubs, a well funded inference specialist with existing regional roots is exactly the kind of partner the Gulf's sovereign AI ambitions depend on.