The U.S. Department of Justice (DOJ) has opened a formal antitrust investigation into the licensing agreement between Nvidia and AI inference chip startup Groq, and has issued a written request for information to Nvidia. According to Bloomberg, the deal is valued at roughly $20 billion (some reports cite a figure of $17 billion). The agreement was announced in December 2025: under the banner of a "non-exclusive license," Nvidia acquired Groq's Language Processing Unit (LPU) chip technology while bringing Groq founder and CEO Jonathan Ross and Chief Operating Officer Sunny Madra into its ranks. Groq itself nominally remains an independent entity.
Senators Elizabeth Warren and Richard Blumenthal promptly opened a parallel investigation, concluding that the transaction is in fact a "de facto merger wrapped in a license," designed to sidestep federal filing review while squeezing out one of the few genuinely differentiated competitors Nvidia faces in AI chips. Nvidia responded: "The Groq case is precisely proof that the American system works well — it is designed to promote innovation, reward entrepreneurs, and benefit consumers." The DOJ declined to comment, citing the ongoing investigation.
The Licensing Cloak: The Operating Logic of Evading Filing Thresholds
To understand why this deal has set off regulatory alarm bells, one must first grasp a legal boundary: under the U.S. Hart-Scott-Rodino Act (HSR Act), mergers and acquisitions exceeding a certain value threshold must be filed in advance with the Federal Trade Commission (FTC) and the DOJ and await review. But in the traditional understanding, the trigger for that filing obligation hinges on the formal element of "equity acquisition."
The structural design of the Nvidia–Groq arrangement lands precisely outside that boundary: no equity is bought or sold, no assets are acquired — only a "non-exclusive" technology license, under which Groq could in theory license the same LPU technology to others. Yet at the same time, the company's actual brain — its founder and core management — has moved wholesale into Nvidia. What remains is a shell with a name, investors, and books that still technically function, plus an agreement that gives Nvidia full technology rights on the most threatening heterogeneous path in AI inference accelerators.
What investigators are focused on is precisely the "subjective" character of this structure: was the deal designed from the outset on the premise of circumventing HSR filing? If so, even without a completed equity transfer, it could constitute evasion of antitrust filing obligations and trigger civil penalties. However, according to sources cited by South Korea's SBS, even if the DOJ concludes there was a violation, it is more likely to opt for a fine rather than demanding that the deal be unwound — the latter would be legally extremely difficult.
Hollowing Out or Empowering: What Groq's Valuation Shift Reveals
One detail best illustrates where this deal is really heading. According to data from market research firm PitchBook, Groq was valued at roughly $7 billion before signing the agreement. Yet according to a separate Bloomberg report, shortly after the deal was announced, Groq completed a $350 million funding round — at a valuation of just $3.5 billion, halved from before the transaction.
For a startup, securing licensing revenue on the order of $17–20 billion would normally be enough to send its valuation soaring. Instead, the opposite happened: Groq's market pricing shrank by half over that period. One possible explanation: with the core talent and technological say already in Nvidia's hands, the market objectively discounted the competitiveness represented by "what remains of Groq." In its reporting, The New York Times noted that such deals often leave startups stripped of their core resources with nothing but a hollowed-out shell.
That valuation trajectory also undercuts Nvidia's "promoting innovation" defense: if the deal were a win-win, both companies should benefit — not just one side's share price and technology stockpile.
Not Just Nvidia: The Endpoint of an Industry-Wide Evasion Template
What has sent systemic shockwaves through the AI industry is that Nvidia–Groq is not an isolated case, but the latest instance of an operating template that large companies have already validated and reused at scale.
Over the past two years, tech giants have deployed more than $40 billion through a combined "license plus bulk talent acquisition" structure: Amazon absorbed the core team of AI agent company Adept, Microsoft brought Inflection AI co-founder Mustafa Suleyman under its wing, and Google completed a similar technology licensing and talent transfer arrangement with Character.ai. The core logic of every deal is the same — formally preserving the target company's legal independence while substantively completing an off-the-books acquisition. According to reports, Amazon's Adept deal has already drawn preliminary scrutiny at the FTC, and Microsoft's partnership with OpenAI was flagged by the FTC in January of this year as posing anticompetitive risks.
In February 2026, Senators Warren, Blumenthal, and Wyden jointly sent a letter to the FTC and DOJ demanding investigations into similar transactions by Nvidia, Meta, and Google, and calling for such deals to be "blocked or unwound" upon a finding of illegality. FTC Commissioner Andrew Ferguson also stated publicly in January of this year that the agency is studying such structural transactions.
The Nvidia–Groq case marks a milestone as the first time federal enforcement agencies have formally opened a case against such a transaction at the "investigation" level. It signals the end of an implicit consensus: this structured path of "bypassing HSR" long operated in a regulatory gray zone, and regulators have now decided to confront its legal boundaries head-on.
Why Nvidia Needs Groq
To understand the strategic significance of this deal for Nvidia, one must confront an uncomfortable fact: in the AI inference chip market, Nvidia is not without rivals, but Groq is among the most technically differentiated of them.
The LPU (Language Processing Unit) developed by Groq is not an improvement along the architectural path of Nvidia's GPUs, but a heterogeneous chip purpose-built for large language model inference scenarios, with a significant advantage in latency metrics. That means Groq was not trying to outrun Nvidia on the same track, but to open a different one — and that track happens to be the market frontier heating up fastest as compute demand on the inference side explodes.
For Nvidia, whose market share is already highly concentrated, eliminating a competitor capable of charting a different course at the methodological level carries more long-term value than defeating a rival of the same kind. The senators' allegations point squarely at this core logic: the goal of the deal was not technology integration, but strategic blockade.
What Regulators Can and Cannot Do
In practice, this investigation faces a fundamental constraint: the deal is already done, the key personnel have already started, and the technology license is already in effect. Even if the DOJ ultimately determines that Nvidia evaded a filing obligation, the enforcement tools available are fairly limited — a fine is the most likely outcome, but for a company with a multi-trillion-dollar market capitalization, the fine amount is little more than a calculable cost of doing business.
The greater enforcement value lies in the demonstration effect: if the DOJ ultimately finds that a "license plus talent acquisition" structure can trigger antitrust penalties, it will create a real expectation of legal costs for similar deals currently being planned in Silicon Valley. That carries more deterrent force than the final ruling in any single case.
At the same time, it forces regulators to confront the question of the HSR rules' applicability in the AI era: can a filing threshold designed for the age of industrial mergers still cover a digital economy reality in which "talent is the asset and a technology license is control"? The senators' intent may not be aimed solely at Nvidia, but at pushing Congress to systematically update the merger review framework through legislation.
Independent Assessment
The real crux of this case is not whether Nvidia violated a specific provision, but that it has precisely exposed an institutional loophole: when competitive advantage can be obtained by "buying the founders" rather than "buying the company's equity," the traditional merger review framework fails structurally.
The data showing Groq's valuation falling from $7 billion before the deal to $3.5 billion is the most honest footnote to this logic. A startup that truly holds a "non-exclusive" license and still retains its core competitiveness should not see its market value halved after signing a large licensing agreement. The market has told us, through valuation, a fact that no regulatory filing can state outright: this is not a partnership, but a structurally designed absorption.
Even if it ends in a fine, the DOJ's decision to open this case is historic — it is the first time regulators have formally acknowledged that the risk of market concentration in the AI era requires new enforcement tools. And for the AI infrastructure supply chain, the deeper question is this: as choice at the chip layer continues to converge toward a handful of companies, every AI system built on top of it in the future will inevitably operate within a passive monopoly structure.
© 2026 Winzheng.com 赢政天下 | 转载请注明来源并附原文链接