DOJ Probes Nvidia-Groq Deal: A Landmark Test for AI Reverse Acquihires

Share

The U.S. Department of Justice (DOJ) opened a formal probe in September 2026 into Nvidia’s December 2025 deal with Groq, a reverse acquihire valued at $17-20 billion. The investigation could create the first antitrust precedent on the legal qualification of a technology license combined with a mass transfer of senior personnel.

๐Ÿ”‘ Key takeaways

  • Initial announcement: December 24, 2025 (non-exclusive license + founder recruitment)
  • Deal value: between $17 billion (Reuters, Axios) and $20 billion (The Register, IntuitionLabs)
  • Current HSR threshold: $119.5 million; potential fine: ~$50,000/day of violation
  • Key personnel transferred: Jonathan Ross, Sunny Madra, and senior engineers
  • Implications: Microsoft-Inflection, Google-Character.AI, and Meta-Scale AI exposed retroactively

A reverse acquihire disguised as a technology license

On December 24, 2025, Nvidia announced it had secured a non-exclusive license to Groq’s LPU (Language Processing Unit) inference chips while recruiting founder Jonathan Ross, president Sunny Madra, and several senior executives. Groq Inc. formally remained an independent entity, retaining the right to use its own technology. On March 9, 2026 โ€” less than three months after signing โ€” Nvidia unveiled at GTC a new inference processor integrating Groq’s intellectual property, demonstrating the know-how transfer was already effective.

This deal illustrates a broader trend across the AI market: facing antitrust scrutiny, tech giants are bypassing traditional M&A by using hybrid legal structures. The playbook: sign an IP license, transfer the brains, and let the target entity survive formally as an operational shell.

Groq 3 LPU: 45ร— the bandwidth of an H100 GPU

Groq built an architecture based on on-die SRAM (fast static memory), eliminating the need for external HBM (High Bandwidth Memory). The Groq 3 LPX, in full production since August 2026, packs 500 MB of on-chip SRAM and delivers an internal memory bandwidth of approximately 150 TB/s โ€” roughly 45ร— that of an Nvidia H100 GPU. Nvidia claims 35ร— more throughput per megawatt than HBM-based inference at equivalent model size, a decisive edge for real-time workloads (conversational agents, autonomous driving, search).

The DOJ and the Hart-Scott-Rodino (HSR) Act

On September 10, 2026, the New York Times and Reuters confirmed the opening of a formal DOJ probe. The core question: should the arrangement have triggered pre-merger notification under the Hart-Scott-Rodino (HSR) Act, whose threshold stands at $119.5 million? Nvidia argues that, as a license rather than a stock acquisition, no filing was required. The DOJ is examining whether the combined transfer of know-how and human capital constitutes an “asset acquisition” under the Clayton Act.

“By licensing its technology and hiring its most important employees, NVIDIA has effectively acquired Groq in all but name.”

Elizabeth Warren and Richard Blumenthal, U.S. Senators

Academic doctrine supports the asset-acquisition qualification. David T. Wong, in a Yale Law Journal comment, argues that reverse acquihires must be reclassified, since the precious asset is the human capital and patent access. The University of Chicago Law Review reached a similar conclusion.

If the DOJ concludes the deal should have been notified, Nvidia faces civil penalties of approximately $50,000 per day of violation, accruing from the closing date. However, according to sources close to the matter, the agency is unlikely to unwind the already-integrated deal: a financial sanction and a legal precedent would suffice.

A precedent with massive sector-wide implications

A DOJ ruling against Nvidia would ripple across other AI reverse acquihires that all sidestepped HSR filing:

AcquirerTargetStructureYear
NvidiaGroqLicense + founder recruitment2025
MicrosoftInflection AITeam recruitment + license2024
GoogleCharacter.AI / WindsurfInvestment + recruitment2024
MetaScale AIMinority stake + CEO recruitment2025

The Meta-Scale AI deal, valued at $14 billion, is the largest transaction of this kind to date and would immediately become vulnerable to retroactive review. China’s SAMR already opened an investigation in December 2024 into Nvidia’s 2020 Mellanox acquisition, while the European Commission unconditionally cleared the Run:ai deal ($700M) in late 2024.

Groq Inc. after the founders’ departure

Despite the brain drain, Groq Inc. remains operational. In June 2026, it raised $650 million to rebuild its AI inference cloud business, now operates 13 data centers across four continents, and claims more than 5 million developers on its platform. It is even among the first customers of its own Groq 3 LPX chips and signed a partnership with Humain, the Saudi state-backed AI firm, to triple deployments in the Kingdom.

Before the deal, Groq had carried out layoffs and struggled to secure compute resources to scale LPU production. The inference cloud was projected to generate more than $40M in 2025 revenue, with total sales forecast at $500M. Simon Edwards, appointed CEO after the founders’ departure, has confirmed the continuation of an independent strategy focused on sovereign models and white-label deployments.


Conclusion: the end of undeclared reverse acquihires?

The DOJ probe marks the first time a U.S. regulator moves from stated concerns to a formal inquiry on the legal structure of reverse acquihires. Nvidia, through Jensen Huang, responded that the deal is a “prime example of the American system working as designed,” but the “asset acquisition” qualification now looks hard to dismiss: human capital, know-how, and patents are all transferred de facto, regardless of formal legal status.

If the DOJ sanctions Nvidia, the precedent will force the entire sector to rework its structures and retroactively notify past transactions โ€” a legal earthquake whose waves will extend well beyond semiconductors. Conversely, a dismissal would officially enshrine a new standard of contractual engineering that every acquirer would seek to emulate.

Sources

This article is published for informational and educational purposes only. It does not constitute investment advice. Do your own research (DYOR) before making any decision.

Disclaimer: this content is for information purposes only and is not financial advice. Cryptocurrencies are highly volatile: you may lose all of your capital. Always do your own research. Legal notice
Telemac
Telemachttp://cryptoinfo.ch
Passionnรฉ de nouvelles technologies, jโ€™explore lโ€™univers de la blockchain et des cryptomonnaies pour partager lโ€™actualitรฉ et les innovations du secteur.

Read More

Items