The DOJ's Nvidia-Groq Probe and the Acqui-Hire Question

The Justice Department is investigating whether Nvidia designed its deal with the AI chip startup Groq to slip past antitrust review, according to reporting on September 10, 2026. The mechanics are the whole story. Nvidia did not buy Groq. It took a non-exclusive license to Groq's chip technology, reported at somewhere between 17 and 20 billion dollars, and hired several of the startup's executives, including founder Jonathan Ross. The company stayed independent on paper. The question the DOJ is asking is whether that structure was a genuine partnership or an acquisition wearing a costume.

NVIDIA

Microsoft

Google

AWS

Why the structure, not the size, is the issue

Under US law, deals above a certain size normally trigger a Hart-Scott-Rodino filing, a mandatory pre-merger notice that gives regulators a window to examine an acquisition before it closes. An outright purchase of Groq would almost certainly have crossed that threshold. A license plus a round of hiring might not, because neither a licensing agreement nor an employment offer is, by itself, a merger.

That is the loophole the DOJ is probing. If a company can obtain the substance of an acquisition, the technology and the people who built it, through contracts that individually escape review, then the notice regime that antitrust enforcement depends on stops working. Nvidia's public framing is that the deal is competition working as intended: "a prime example of the American system working as designed to promote innovation, reward entrepreneurs, and benefit consumers." The DOJ's implied counter is that the design in question may be the deal's, not the system's.

This is a pattern, not a one-off

The reason the case matters far beyond one chip deal is that the license-and-hire structure has become a standard move across AI. Nvidia itself used a close variant with Poolside: a reported 6 billion dollar license payment alongside the hiring of 109 staff. And the template predates Nvidia's chip deals. In March 2024 Microsoft licensed Inflection AI's models for a reported 620 million dollars, paid a further 30 million, and absorbed the founders and roughly seventy staff. That June, Amazon hired Adept's founders and about two-thirds of its employees while licensing its technology. In August 2024 Google licensed Character.AI's models and hired around 21 percent of its people.

Regulators have a name for this, in fact several: stealth mergers, reverse acqui-hires, pseudo-acquisitions. The FTC signaled it would examine the structure, the UK's Competition and Markets Authority opened a formal investigation into Microsoft's Inflection hires, the European Commission said it would monitor such deals, and US senators wrote to the FTC and DOJ arguing the transactions "function as de facto mergers." What has been missing is a decisive enforcement outcome. The Nvidia-Groq probe is the clearest attempt yet to convert that concern into a case.

What a ruling could and could not do

It is worth being precise about the range of outcomes, because coverage tends to skip to the dramatic one. The probe could lead to a fine or a negotiated remedy, but reporting suggests it is unlikely to unwind the Nvidia-Groq arrangement itself. The more consequential effect would be precedent. If the DOJ establishes that license-and-hire deals of a certain shape require merger-style disclosure and review, the next such deal gets slower, more expensive and more visible, and some deals that would have happened quietly do not happen at all.

There is a genuine tension underneath, and neither side is obviously right. License-and-hire deals can be a legitimate and fast way for a struggling startup's technology and team to find a home that keeps them productive, which is a real benefit to innovation and to employees. They can also be a way for the largest incumbents to absorb their most promising challengers without ever facing the scrutiny a purchase would draw, which is a real harm to competition. The same transaction can be both, and telling them apart is exactly the hard, fact-specific work an investigation exists to do.

The concentration backdrop

For anyone building on AI, the probe is a reminder of how few hands the critical layers of the stack sit in. When the dominant supplier of AI accelerators can pull the technology and talent of a rival inference-chip startup into its orbit through a contract, the set of independent alternatives at the hardware layer narrows, whatever the deal is called. That concentration is precisely why the discipline of staying portable matters higher up the stack, keeping the model layer and the tools built on it able to move across providers rather than fused to whichever vendor consolidates next. It is the same reasoning behind model-agnostic platforms like Metir, which treat no single provider as permanent. The antitrust question about who owns the chips and the competitive question about who owns your workflow are two views of one problem: how much optionality is left when a market consolidates.

The Nvidia-Groq probe will take time, and its direct result may be modest. Its lasting importance is that it forces a question the AI industry has spent two years avoiding out loud: when does buying the technology and hiring the team stop being a partnership and start being a merger that someone was supposed to review.

Sources:

• DOJ Probes Nvidia's License Deal With Groq on Antitrust Concerns (Bloomberg, Sept 10, 2026)
• DOJ investigates Nvidia's deal with Groq (Axios, Sept 10, 2026)
• The Justice Department is investigating whether Nvidia structured its Groq deal to avoid antitrust review (The Next Web)
• Amazon's deal with AI startup Adept faces FTC scrutiny (CNBC, July 16, 2024)
• FTC Eyes Reverse Acquihires in AI Sector (American Action Forum)
• Global Antitrust Enforcers Tackle AI Stealth Mergers and Acqui-Hires (Mogin Law LLP)

Image credits

• Hero: Nvidia Endeavor headquarters, Santa Clara. Wikimedia Commons, licensed CC BY-SA 4.0.
• In-body: Nvidia sign on Scott Boulevard, Santa Clara. Wikimedia Commons, licensed CC BY 2.0.