The DOJ Is Testing Whether a License Is Really a Merger
Nvidia licensed Groq's technology for roughly $20 billion and hired its founder without buying the company. The Justice Department has now sent a formal information demand asking whether that was the point.
On September 10, Bloomberg reported that the Justice Department is investigating whether Nvidia structured its roughly $20 billion licensing agreement with Groq to avoid antitrust review. The DOJ opened the inquiry shortly after the deal was announced in December and has since sent Nvidia a formal demand for information.
The deal itself is a now-familiar shape. Nvidia took a non-exclusive license to Groq's chip technology and hired several of its executives, including founder Jonathan Ross. It did not buy the company. Groq still exists, still has shareholders, and on paper nothing changed hands but intellectual-property rights and employment contracts.
That structure is the entire subject of the investigation.
Why the structure exists
Under Hart-Scott-Rodino, an acquisition above a dollar threshold must be notified to the agencies and cleared before it closes. A license is not an acquisition. Hiring people is not an acquisition. Do both at once, at sufficient scale, and you can obtain most of what a merger would have delivered — the technology, the team that knows how to build it, the roadmap — without triggering the filing that would have put the deal in front of a reviewer.
The pattern has a name now: the reverse acquihire. It has moved more than $100 billion of AI-era capital without mandatory merger review. Microsoft and Inflection ran it. Amazon and Adept ran it. Google and Character.AI ran it. Nvidia and Groq is the largest single instance.
Until now, every regulatory response to this playbook has stopped at concern. Senators Warren and Blumenthal questioned the Groq arrangement in March. The FTC ran a 6(b) study on AI partnerships. Nothing escalated. This is the first time a US regulator has moved from stated concern to a formal investigation of the structure itself.
What the DOJ has to prove, and why it is hard
To act, enforcers would have to argue that the arrangement functioned as a merger even though it was not one on paper.
That is a real legal theory, not a stretch. Section 7 of the Clayton Act reaches acquisitions of "assets," and courts have long held that substance beats form. If the license transfers effective control of Groq's competitive capability, and the hiring transfers the people who could rebuild that capability elsewhere, the practical effect on the market is indistinguishable from Nvidia buying Groq and shutting the product line.
But there are two hard problems.
The first is the word non-exclusive. Groq retained the right to keep using and licensing its own technology. Nvidia's lawyers will point to that clause as proof that competition survived, and any remedy the DOJ wants has to explain why a non-exclusive grant removed a competitor from the market.
The second is timing. The deal was announced in December and has been operating for nine months. Unwinding it would mean un-hiring executives and un-transferring know-how, which is not a thing courts can order. Bloomberg's reporting notes the investigation is unlikely to unwind the arrangement. Whatever the DOJ gets here, it will not be Groq restored to independence.
So what is the investigation actually for
Two things, both more important than this deal.
Precedent on the structure. If the DOJ articulates a theory under which license-plus-hire equals acquisition, that theory applies going forward to every AI deal built on the same template. The deterrent value does not require winning in court. It requires making the structure legally uncertain enough that boards stop treating it as a free pass around HSR. That is achievable through an investigation that ends in a consent decree, a guidance document, or even a well-argued complaint that settles.
Leverage on Nvidia specifically. Nvidia is already the subject of open antitrust interest in the US, the EU, and China over bundling, allocation, and interconnect. An information demand on Groq gives the DOJ a documentary window into how Nvidia thinks about competitive threats in inference — which is the segment where Groq's low-latency architecture was one of the few credible alternatives.
The part that should worry acquirers
The specific detail the agencies keep circling is talent plus IP together.
Hiring a competitor's engineers is generally legal. Licensing a competitor's technology is generally legal. The theory under examination is that doing both, simultaneously, at a price that approximates the company's enterprise value, is a purchase with extra steps. Roughly $20 billion is close enough to "what Groq was worth" that the characterization is not absurd — and that is the number that makes this case different from the smaller instances of the pattern.
Deals below the HSR threshold were never going to attract this. Deals at twenty billion dollars, dressed as licenses, were always going to eventually.
What happens next
A formal information demand is discovery, not a charge. Nvidia will produce documents over months. The realistic outcomes are: the DOJ closes the inquiry without action, opens a broader investigation into Nvidia's inference-market conduct using what it finds, or issues guidance that makes the structure riskier without litigating this instance.
None of those change the Groq outcome. All of them change the next one. The reverse acquihire has spent two years as the cleanest way to buy a company without asking permission. It just stopped being free.