Nvidia's $20 billion Groq deal draws DOJ antitrust probe

Nvidia's $20 billion Groq deal draws DOJ antitrust probe

Nvidia spent $20 billion late last year on a deal with Groq structured as licensing plus hiring rather than an outright acquisition: it licensed Groq's AI accelerator technology and hired away key members of its engineering team, while leaving Groq's own inference-as-a-service business technically intact. This week The New York Times reported that the US Department of Justice has opened an antitrust probe into the arrangement.

Nvidia defended the deal in a statement provided to The Register and other outlets, calling it "a prime example of the American system working as designed to promote innovation, reward entrepreneurs, and benefit consumers."

What Nvidia actually bought was mature silicon and the talent to keep developing it. Groq built SRAM-heavy dataflow accelerators that pushed large language model inference to hundreds and now thousands of tokens a second, something GPU-based systems had struggled to match on their own. At GTC in March, Nvidia unveiled LPX racks powered by 256 Groq-3 accelerators; The Register understands these are lightly modified versions of Groq's existing Groq-2 chip design, which would explain how new silicon could ship so quickly. Nvidia CEO Jensen Huang said pairing Groq-3 with Nvidia's Vera Rubin GPU racks would deliver optimal performance across the full range of inference workloads.

Several US senators have argued that deals engineered to avoid regulatory scrutiny should get that scrutiny anyway, though the article does not name them. The central antitrust question is whether the deal harmed competition, and The Register says that will be hard to prove given how many rivals are already building comparable disaggregated compute systems: Cerebras with AWS and AMD, SambaNova with Intel, and d-Matrix combining its in-memory compute platform with Nvidia GPUs.

Even if the DOJ forced Nvidia to unwind the deal, The Register argues the practical effect would be limited. Nvidia had already been opening its hardware ecosystem before the Groq deal: it contributed its MGX rack designs to the Open Compute Project in late 2024, and in mid-2025 it opened its NVLink high-speed interconnect technology to the industry through NVLink Fusion licensing. That combination already lets other chipmakers slot compatible designs into Nvidia's racks, so unwinding the Groq acquihire would likely just turn Groq into one more hardware partner in Nvidia's ecosystem rather than restore it as an independent rival.

Key facts

  • Nvidia spent $20 billion late last year on a licensing-and-hiring deal with Groq that left Groq's own inference-as-a-service business technically intact.
  • The New York Times reported this week that the US Department of Justice has opened an antitrust probe into the deal.
  • At GTC in March, Nvidia unveiled LPX racks powered by 256 Groq-3 accelerators, which The Register understands are lightly modified versions of Groq's existing Groq-2 chips.
  • Nvidia CEO Jensen Huang said combining Groq-3 with Vera Rubin GPU racks would deliver optimal performance across the full range of inference workloads.
  • Rivals including Cerebras (with AWS and AMD), SambaNova (with Intel), and d-Matrix are building similar disaggregated compute systems, so unwinding the deal would not remove competing alternatives.

Why it matters

The case tests how far antitrust regulators will reach into "acquihire" deals built to avoid formal merger review in the AI chip market. Nvidia's $20 billion deal absorbed Groq's key engineers and licensed its technology without buying the company outright, and a DOJ probe, per The New York Times, is an early sign that structuring a deal this way does not automatically put it outside antitrust reach.

Who it affects

Nvidia and Groq directly; AI infrastructure rivals building comparable disaggregated inference hardware, including Cerebras with AWS and AMD, SambaNova with Intel, and d-Matrix; customers of Nvidia's LPX racks, which pair 256 Groq-3 accelerators with Vera Rubin GPUs; and the US senators and regulators weighing whether hiring-and-licensing deals of this kind deserve the same scrutiny as a merger.

How to use it

For other AI hardware makers weighing a similar licensing-and-hiring structure, the lesson is that avoiding a formal merger does not guarantee avoiding antitrust review. For anyone tracking Nvidia's broader strategy, the relevant detail is that its MGX rack designs, donated to the Open Compute Project in late 2024, and its NVLink Fusion licensing, opened to the industry in mid-2025, already let outside chipmakers plug compatible hardware into Nvidia's racks. That is the basis for The Register's argument that the antitrust outcome will not change much on the ground.

How solid is it

The account rests on a single Register piece citing an NYT report of the DOJ probe; the DOJ itself is not quoted or cited directly, and the article does not name the senators it says have raised concerns about deals structured this way, nor give a timeline for the probe or say whether it has produced any subpoenas or formal action. The claim that Groq's LPX chips are lightly modified Groq-2 designs is presented as the outlet's own understanding, not a confirmed technical detail. The piece is as much analysis as reporting: its conclusion that the DOJ probe is already too late to matter is The Register's own assessment, not an established fact.

Risks and caveats

The exact date the Nvidia-Groq deal was signed is unknown beyond "late last year," and no valuation is given for Groq as a standalone company, only the $20 billion Nvidia paid. Whether the DOJ probe proceeds to any formal action, and on what timeline, is not stated. Nvidia's own statement defending the deal is the only company comment quoted in the piece, so its framing of the arrangement's motives is one-sided.

“The Groq story is a prime example of the American system working as designed to promote innovation, reward entrepreneurs, and benefit consumers. The law is designed to encourage America's startup ecosystem and promote the fundamental rights of inventors and workers to pursue their dreams.”

— Nvidia, in a statement provided to The Register