Groq’s December agreement with Nvidia was presented as an unusual AI-chip alliance, not a conventional takeover. Nvidia licensed Groq’s inference technology, brought founder and CEO Jonathan Ross, president Sunny Madra and roughly 150 to 200 engineers into its ranks, and said Groq would remain independent. Jensen Huang, Nvidia’s chief executive, told employees the company was “not acquiring Groq as a company,” even as it planned to integrate Groq’s low-latency processors into Nvidia’s AI-factory architecture.¹

That distinction is now at the heart of a Delaware Court of Chancery lawsuit filed Oct. 2 by former Groq engineers Joshua Rubin and Benjamin Serebrin, who retained stock after leaving the startup. They contend that the roughly $20 billion arrangement effectively sold core company assets while bypassing the shareholder vote required under Delaware law.

Their complaint casts the structure as a squeeze-out: $17 billion for what it says was labeled a non-exclusive licence, plus $3 billion in Nvidia restricted stock units for employees who moved with the technology. The former engineers allege Groq’s board was conflicted, failed to test or maximize the value of what Nvidia acquired, and accepted a “lowball” outcome that cost stockholders billions.

Groq rejects that account outright. A spokesperson said its Nvidia licensing agreement delivered “exceptional value for Groq, our investors, and our employees,” called the lawsuit “meritless,” and said the company would fight it vigorously while continuing to build its inference cloud.¹

The clash turns on a deceptively simple question: was this a licence with a major talent transfer, as Nvidia and Groq maintain, or a sale in all but name, as the plaintiffs argue? The answer could determine whether Groq’s shareholders were participants in the deal—or merely spectators to it.¹