Nvidia paid twenty billion dollars for a non-exclusive license to Groq's inference technology. Ninety percent of Groq's employees joined Nvidia. No antitrust filing was made. The deal structure is the antitrust strategy.
On Christmas Eve 2025, Nvidia announced the largest deal in its history — roughly twenty billion dollars for Groq, the inference chip startup. But Nvidia says it did not acquire Groq. A spokesperson told reporters that the company purchased a non-exclusive license to Groq's intellectual property and hired engineering talent. Groq remains, on paper, a separate and independent business.
Ninety percent of Groq's employees joined Nvidia, including founder and CEO Jonathan Ross. The deal valued Groq at 2.9 times the 6.9 billion dollar valuation from its Series E round three months earlier. No Hart-Scott-Rodino filing was made. No antitrust review was triggered.
Today, Senators Elizabeth Warren and Richard Blumenthal sent a letter to Jensen Huang asking for the specific terms of the deal and whether it was structured to evade scrutiny by antitrust regulators. They wrote that Nvidia effectively controls which companies can compete in AI and that the entire AI industry is held hostage to Nvidia's product decisions and priorities.
This is not the first time these senators have raised the alarm. On February 4, Warren, Blumenthal, and Ron Wyden sent a joint letter to the FTC and DOJ urging investigation of what they called reverse acqui-hire deals — naming the Nvidia-Groq, Meta-Scale AI, and Google-Windsurf transactions as de facto mergers. FTC Chair Andrew Ferguson said in a Bloomberg interview on January 16 that the agency is beginning to examine acqui-hires to make sure they are not an attempt to get around the merger review process.
The Pattern
The Nvidia-Groq deal is not an anomaly. It is the latest instance of an industry-standard technique that has moved over forty billion dollars in transactions past antitrust regulators.
Microsoft paid 650 million dollars in March 2024 for a non-exclusive license to Inflection AI's technology. It hired nearly all seventy employees, including co-founder Mustafa Suleyman. The UK's Competition and Markets Authority designated it a relevant merger situation. The FTC opened an inquiry.
Amazon paid roughly twenty-five million dollars in June 2024 to license Adept AI's technology. It hired about two-thirds of Adept's employees, including the CEO and co-founders. The FTC sent questions.
Google paid 2.7 billion dollars in August 2024 for a non-exclusive license to Character.AI's models. It hired co-founder Noam Shazeer and about thirty key researchers. The Department of Justice began investigating.
Google paid 2.4 billion dollars in July 2025 for a non-exclusive license from Windsurf after OpenAI's three-billion-dollar acquisition attempt collapsed. It hired the CEO and key R&D staff. No equity stake was taken.
Meta paid 14.8 billion dollars in June 2025 for a forty-nine percent non-voting stake in Scale AI and hired founder Alexandr Wang. Scale AI's CFO insists it was not an acqui-hire.
In every case the legal structure is identical. The acquirer pays acquisition-level prices. The acquirer hires most or all of the target's leadership and technical talent. The target remains nominally independent. No antitrust filing is made because the transaction is structured as a license, not a merger.
Bernstein analyst Stacy Rasgon wrote about the Nvidia-Groq deal: Antitrust would seem to be the primary risk here, though structuring the deal as a non-exclusive license may keep the fiction of competition alive — even as Groq's leadership and, we would presume, technical talent move over to Nvidia.
The Economics
The pricing reveals what the structure conceals. Nvidia paid 2.9 times Groq's valuation from three months earlier. Microsoft paid 650 million for a company that would have been worth far less under normal licensing terms. Google paid 2.7 billion for Character.AI access to a handful of researchers and a non-exclusive license to chatbot technology.
These are acquisition prices attached to licensing paperwork. The premium compensates the target's investors for giving up their company without triggering the legal review that an actual acquisition would require. Series E investors in Groq received roughly 2.9x their money in three months. Early investors received eleven times or more.
At GTC 2026, Huang explained why Nvidia needed Groq's technology. Low-latency premium token generation should represent about twenty-five percent of compute in an AI cluster, and GPUs alone cannot stretch the performance curve far enough for latency-sensitive inference. Groq's LPU dataflow architecture fills an architectural gap that Nvidia's own silicon cannot close. Commentators called it Nvidia's Mellanox moment — buying specialized technology to complete a platform.
That rationale makes the licensing fiction even more transparent. You do not pay twenty billion dollars and absorb ninety percent of a company's workforce for a non-exclusive license to technology you plan to use as twenty-five percent of your platform. You do that to acquire the company.
The Structural Question
The legal question is whether this structure survives regulatory scrutiny. The economic question is what it means that the technique has become standard operating procedure.
Five of the six largest AI companies have now used the same playbook. The deals total over forty billion dollars. In each case the acquiring company obtained the target's technology, leadership, and engineering talent while the target's corporate shell continued to exist as a nominal independent entity. In each case, no Hart-Scott-Rodino notification was filed.
The Hart-Scott-Rodino Act requires companies to notify the FTC and DOJ before completing mergers or acquisitions above a certain threshold. The notification triggers a waiting period during which regulators can investigate potential competitive harm. By structuring transactions as licensing agreements rather than acquisitions, the parties avoid this process entirely.
The FTC chair has said the agency is examining whether these structures circumvent the law. Senators have asked for investigations. The DOJ has opened inquiries into at least one of the deals. But no enforcement action has been taken. The deals have closed. The talent has moved. The technology has been integrated.
What is striking is not that one company found a loophole. It is that the loophole became an industry standard within eighteen months. When Microsoft structured the Inflection deal in March 2024, it was novel — creative lawyering to solve a specific problem. When Nvidia structured the Groq deal in December 2025, it was routine — the expected structure for any large AI talent acquisition.
The regulatory apparatus designed to prevent excessive market concentration in the technology industry has been routed around by a legal technique that all major participants now treat as standard. The question is not whether individual deals are anticompetitive. It is whether the technique itself — the licensing structure as an industry-wide standard for avoiding antitrust review — represents a systematic failure of the regulatory framework.
Warren and Blumenthal are asking about Nvidia. The more important question is about the forty billion dollars that has already moved through the loophole while regulators examine whether it exists.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)