The US Department of Justice (DOJ) is investigating Nvidia’s $17 billion licensing arrangement with artificial intelligence chip startup Groq over concerns that the deal may have been structured to avoid antitrust scrutiny, according to a report by The New York Times cited by Reuters. The investigation focuses on the structure of the arrangement rather than being a finding that Nvidia has violated antitrust laws.
Nvidia announced the deal with Groq in December last year. Under the arrangement, Nvidia obtained a “non-exclusive license” to Groq’s chip technology and hired several executives from the startup, including its founder Jonathan Ross. The structure allowed Nvidia to access Groq’s technology and bring key members of its leadership and engineering team into the company without formally acquiring Groq.
According to the New York Times report, the Justice Department opened its investigation shortly after Nvidia announced the arrangement in December. The agency has also sent Nvidia a formal request for information as part of its examination of the deal. The investigation is aimed at determining whether the structure of the transaction was designed to avoid the type of antitrust review that can accompany a traditional acquisition.
The arrangement has drawn attention because of the combination of technology licensing and executive hiring. Along with obtaining a non-exclusive license to Groq’s chip technology, Nvidia brought in several members of Groq’s leadership and engineering teams. Jonathan Ross, who founded Groq and previously worked on Google’s artificial intelligence chip programme, was among the executives joining Nvidia.
Nvidia has defended the arrangement. A company spokesperson said, “The Groq story is a prime example of the American system working as designed to promote innovation, reward entrepreneurs, and benefit consumers.”
Groq and the US Justice Department had not immediately responded to Reuters’ requests for comment outside regular business hours. The DOJ investigation therefore remains ongoing, and there has been no reported conclusion that Nvidia breached antitrust rules through the arrangement.
The transaction is also significant because of the technology involved. Groq specialises in AI inference, the process through which trained artificial intelligence models generate responses to user requests. Nvidia has traditionally held a dominant position in AI training hardware, while inference has become an increasingly competitive area involving established semiconductor companies and specialised AI-chip startups.
Nvidia’s agreement with Groq was structured as a licensing arrangement rather than a conventional acquisition. Earlier reports had cited a possible $20 billion acquisition of Groq’s assets, but Nvidia and Groq did not describe the transaction as an acquisition. Instead, the companies announced a $17 billion arrangement involving a non-exclusive technology licence and the movement of key Groq personnel to Nvidia.
The regulatory scrutiny comes as major technology companies increasingly use licensing agreements and talent-focused transactions to gain access to artificial intelligence technology and specialised teams without formally purchasing entire startups. Such arrangements have attracted greater attention from antitrust authorities because they can combine access to technology with the movement of important executives and researchers.
The DOJ could impose a fine if it concludes that Nvidia mishandled the arrangement, according to the New York Times report cited by Reuters. However, the report said the agency is unlikely to seek to unwind the transaction.
The investigation adds another layer of regulatory scrutiny to Nvidia as the company continues expanding its position across the AI semiconductor ecosystem. Its deal with Groq was intended to strengthen its capabilities in AI inference, while the DOJ’s review will examine whether the way the transaction was structured complied with US antitrust requirements.
 is investigating Nvidia’s $17 billion licensing arrangement with artificial intelligence chip startup...){kind=link}