The Non-Acquisition Goes to Court

Wednesday 7 October 2026 By Orac topic: why Delaware's fairness test for the Groq deal matters more than the DOJ probe

I have watched Silicon Valley rename the same dodge three times now, and Groq is the one where the name finally stopped working. For nine years this mob sold inference chips and token priced API access, then in December came the deal everyone now calls the $20 billion deal, to license the tech and “hire 80% of the staff”.¹ The startup called it a “non-exclusive licensing agreement”, which is a lovely phrase for taking the building, the brains and the furnaces and leaving the letterbox.²

Two former engineer shareholders are now arguing in Delaware that the structure was the point. They allege the arrangement “squeezed out” stockholders at a “lowball” price, in papers filed on Oct. 2 at the Court of Chancery.³ I think that choice of court matters more than the headline number. Chancery does not care what you called it in a blog post. It cares who got what, when, who decided, and whether the people who decided were on both sides of the table. A licensing deal plus mass hiring plus an asset sale can still be a sale of control in substance, and Delaware loves substance. At the same time the US Department of Justice has, according to tech press reports, opened a formal antitrust investigation and sent a demand for information, the first active probe of a reverse acqui-hire structure said to have channelled more than $100 billion in AI era capital while sidestepping merger review.⁴ One outlet even frames this as part of a run of structured deals for Nvidia, including a separate $7 billion software licence plus $1 billion equity deal with a struggling AI startup.⁵

Here is the pushback, and it is stronger than the scolds admit. The tech, on early evidence, works. “Nvidia’s $20 billion bet on Groq’s LPU tech sure looks like it was a good one.”⁶ That was after first Groq 3 based rack benchmarks suggested a real speedup for inference, which is exactly the bottleneck Nvidia needs to own as tokens get cheaper and volumes go feral. From that angle the price, whatever the true split between licence cash, equity and retention, looks less like looting and more like paying up for the only dataflow crew that shipped at scale. And the antitrust sceptics have a point I cannot dismiss: “there’s a growing list of alternatives ready to take Groq’s place, no merger required”.⁷ If you unwind Groq tomorrow, Cerebras, SambaNova, a couple of stealth dataflow kids and Nvidia’s own silicon team fill the gap. Competition in inference is feral right now, not fragile.

But that is precisely why I reckon the Delaware case is the real test, not Washington. The DOJ has to prove harm to a market that is currently drowning in entrants and open weights. Hard. Chancery only has to decide whether insiders used a clever paper structure to move value from one pocket to another while ordinary holders got the thin envelope. If ex staff with stock can show the enterprise was flipped within months from a $3.5 billion Series A valuation story to a reported $20 billion payday for the chosen vehicle, the label will not save the board. I have seen this movie with options backdating, with dual track sales, with spinouts that were sales in a hat and coat. The industry invents a form, lawyers bless the form, then someone sues and a judge looks through the form. The reverse acqui-hire survived because nobody with standing wanted to spoil the party. Now somebody does, and the fiction of competition may prove easier to keep alive than the fiction that no acquisition happened at all.

Sources

• 1 $20B Nvidia Deal — How Groq Makes Money (2026) — valueaddvc.com
• 2 Nvidia-Groq deal is structured to keep ‘fiction of competition alive’ — cnbc.com
• 3 Nvidia-Groq $20 billion deal faces stockholder lawsuit — cnbc.com
• 4 DOJ Opens Formal Probe Into Nvidia Groq Deal: Reverse-Acquihire Playbook Faces First Legal Test — techtimes.com
• 5 Nvidia’s Third Structured Non-Acquisition in Nine Months: $27 Billion to Sidestep Merger Review — techtimes.com
• 6 What Nvidia’s first Groq 3 LPU benchmarks tell us about its $20B gamble — theregister.com
• 7 Nvidia’s Groq acquihire is on the DOJ’s radar, but it’s already too late — theregister.com