Call It a License. Antitrust Will Ask What Actually Changed.

The label is the beginning of the analysis

Reuters reported early Thursday that DOJ is examining whether Nvidia's arrangement with AI-chip startup Groq was structured to avoid antitrust scrutiny. Reuters attributed the underlying report to The New York Times and said DOJ sent Nvidia a formal request for information after the arrangement was announced. Nvidia defended the transaction as an example of the American innovation system working as designed. Groq and DOJ did not comment to Reuters. That is the present procedural posture: a reported investigation, a company defense, and no public agency finding.

The arrangement was unusual enough to invite a closer look. On December 24, 2025, Groq publicly described a non-exclusive license of its inference technology to Nvidia. Groq's founder Jonathan Ross, president Sunny Madra, and other members of the team were to join Nvidia. Groq said it would continue operating independently under a new chief executive and that its cloud business would continue. The companies did not disclose financial terms at the time. The new report describes the arrangement as worth $17 billion.

None of those facts, standing alone, answers the antitrust question. A non-exclusive license can preserve competition because the licensor remains free to use and license the technology. Hiring executives and engineers can be ordinary labor-market movement. A startup can remain a meaningful competitor after leadership changes. Payment for technology can reward invention and accelerate deployment. Antitrust is not a rule against creative contracting or successful companies buying useful inputs.

But the form does not end the inquiry either. If a package transfers the technology, the people who know how to develop it, access to supporting assets, and the practical ability to determine how the capability is commercialized, the remaining corporate shell may tell less about competition than the package itself. The point is not that this happened here. The point is that the answer requires facts the phrase ‘non-exclusive license’ does not contain.

Notification and competitive harm are different questions

Lawyers should resist collapsing every concern into Hart-Scott-Rodino filing analysis. The HSR Act creates a premerger notification and waiting-period system for certain acquisitions of voting securities or assets that meet the statutory and regulatory requirements. Whether a particular bundle triggers a filing obligation is a technical question about the transaction, the parties, thresholds, exemptions, and what is being acquired. It deserves its own careful answer.

The broader competitive question is different. The DOJ and FTC Merger Guidelines explain how the agencies assess whether a transaction may substantially lessen competition or tend to create a monopoly. The Guidelines are enforcement guidance, not a statute and not a judicial holding. They nevertheless make the agencies' analytical focus clear: market realities, competitive effects, control, entry, coordination, and the loss of actual or potential rivalry matter more than the vocabulary chosen for a press release.

That distinction is critical because a lawyer can be right that a conventional stock acquisition did not occur and still be unprepared for questions about the economic substance of a set of related agreements. Conversely, agency interest does not transform an unconventional commercial arrangement into an unlawful merger. Investigation is the process for learning which description is more accurate. It should not be reported as a conclusion.

The FTC signaled this concern before the Nvidia-Groq report. In January 2024, it used its statutory study authority to demand information about major investments and partnerships involving AI developers and cloud providers. The agency specifically sought agreements, strategic rationales, governance rights, practical implications, competitive-impact analyses, and information about access to key AI inputs. That list is useful because it shows why a single noun such as partnership, investment, license, or acquisition is analytically thin.

The real asset may be the team that can make the technology work

Technology transactions strain legal categories because the valuable asset is rarely one thing. A patent may be useless without tacit engineering knowledge. Source code may be reproducible but the deployment discipline is not. A license may provide formal permission while the hiring plan supplies the people who can turn permission into a product. Customer relationships, road maps, specialized hardware, data access, and the credibility to recruit the next technical team can matter as much as title to an asset.

That is why the remaining company's competitive capacity deserves close attention. After the deal, does it retain the engineers, capital, leadership, rights, infrastructure, and customer access required to improve the product? Can it license the technology to rivals in practice, not merely in theory? Do transition services, exclusivity around future developments, incentive payments, or restrictions on personnel narrow that freedom? Does the buyer obtain information or influence that changes the smaller company's decisions? These questions do not presume an answer. They identify the evidence needed to find one.

The counterfactual also matters. Without the arrangement, was Groq positioned to become a stronger inference competitor, likely to seek a different buyer, dependent on new capital, or headed toward commercial difficulty? A transaction that supplies funding and distribution may strengthen an independent rival. The same transaction may weaken rivalry if it removes the people and capabilities most likely to challenge the buyer. Antitrust analysis is difficult precisely because both stories can sound plausible before the documents, witnesses, and economics are tested.

This is where deal lawyers should be skeptical of tidy narratives from either direction. ‘The company still exists’ is not proof that its competitive significance survived. ‘The founder left’ is not proof that it disappeared. Corporate continuity and competitive continuity are related, but they are not identical.

The file should explain the transaction before an investigator asks

Unconventional structures require more documentation, not less. Counsel should map the agreements as one transaction architecture: licensed rights, acquired or accessed assets, employee movement, payments, governance, information rights, restrictions, continuing dependencies, and post-closing commercial relationships. Separate documents signed by separate entities on separate dates may still describe one economic event. A board presentation or executive message that talks about eliminating a competitor while the operative agreement emphasizes independence will predictably attract attention.

The analysis should also preserve the procompetitive case in ordinary business language. What problem could the parties solve together that they could not solve separately? Why was non-exclusivity necessary? What resources remain with the smaller company? What protects its ability to serve customers and work with other partners? Which alternatives were evaluated, and why was this structure chosen? A legal memo written after a subpoena cannot substitute for contemporaneous evidence of how businesspeople understood the deal.

Public communications deserve the same discipline. Calling an arrangement a partnership may be commercially convenient, but it can sound evasive if most of the valuable team and technology moved in one direction. Calling it an acquisition may be equally misleading if the startup retains meaningful assets, people, customers, and freedom to compete. The best description is specific: say what was licensed, who was hired, what remains independent, and which facts have not been disclosed.

This is not merely advice for billion-dollar AI companies. Professional-services firms increasingly assemble technology, talent, vendor relationships, and intellectual property through contracts that do not resemble a traditional acquisition. Lawyers advising those arrangements should learn the larger lesson. A transaction is not protected by the fact that no single document looks like the transaction an investigator is trying to understand.

The reported DOJ inquiry does not establish that Nvidia, Groq, their executives, or their advisers violated antitrust or premerger-notification law. The relevant agreements and most financial terms are not public. This article analyzes the questions raised by the reported structure and is not legal advice about any particular transaction.

Substance is not a slogan

It is easy to say that regulators look to substance over form. The harder work is defining substance without simply treating every large payment, important hire, or strategic relationship as a disguised acquisition. That requires precision about what moved, what stayed, which competitive options disappeared, which new ones became possible, and what the market would likely have looked like without the deal.

The Nvidia-Groq investigation may end without an enforcement action. It may produce a filing dispute, a competitive-effects case, a settlement, or no public resolution at all. Until the facts are developed, confident conclusions are advocacy. The useful lesson arrives earlier: if the commercial objective is to obtain technology and the people capable of advancing it, counsel should assume that an investigator will examine the whole package. The title on the contract will be one fact among many.