The Department of Justice is probing a licensing arrangement between Nvidia and Groq, one of the small handful of startups building inference chips designed to compete with Nvidia's own accelerators. The probe puts the dominant supplier of AI hardware back under antitrust scrutiny at a moment when its chips underpin nearly every frontier training run and most large-scale inference deployments in the industry.
The terms of the licensing deal at issue have not been disclosed publicly. What is known is that DOJ investigators are examining whether the arrangement between Nvidia and Groq raises competition concerns — the kind of question that typically turns on whether a dominant firm has used a commercial agreement to blunt a would-be rival rather than accelerate it.
Groq occupies a specific niche in the AI hardware market. Its language processing units are built for inference — the run-time serving of already-trained models — and the company has pitched its architecture as faster and cheaper than running the same workloads on Nvidia GPUs. That positioning is precisely the sort of competitive threat antitrust reviewers tend to focus on when evaluating deals between market leaders and challengers.
Key facts
- 01The Department of Justice is probing a licensing deal between Nvidia and AI chip startup Groq.
- 02Groq builds inference-focused accelerators positioned as an alternative to Nvidia GPUs.
- 03The probe adds to a widening set of antitrust reviews targeting Nvidia's grip on the AI hardware stack.
Nvidia has not commented publicly on the reported probe, and Groq has not detailed the licensing terms under review. The DOJ has likewise not confirmed the investigation on the record.
This is not the first antitrust question Nvidia has faced. Regulators in the United States and Europe have opened inquiries over the past two years into how the company allocates supply of its most sought-after accelerators, how it structures pricing for large cloud customers, and how its CUDA software stack interacts with competing hardware. Each individual thread has moved slowly. Taken together, they represent the most sustained regulatory attention the company has drawn in its history.
The inference-chip market is one of the few segments where Nvidia faces credible technical challengers. Groq, Cerebras, SambaNova, and a growing set of hyperscaler-designed silicon — Google's TPUs, Amazon's Trainium and Inferentia, Microsoft's Maia — are all trying to peel off workloads where Nvidia's general-purpose GPUs are overkill or overpriced. Any commercial arrangement between Nvidia and one of those challengers is going to draw regulatory eyes by default.
Licensing deals in semiconductors typically cover patents, interconnect standards, software interfaces, or reference designs. Without visibility into which of those the Nvidia-Groq agreement touches, it is difficult to say what specific theory of harm the DOJ is testing. The range of possibilities runs from mundane — a routine cross-license that regulators wave through — to structural, if investigators conclude the terms constrain how aggressively Groq can compete on the workloads where it overlaps with Nvidia.
Groq has raised substantial private capital on the strength of its inference performance claims and has been pursuing enterprise and API customers looking for alternatives to Nvidia-hosted inference. A DOJ finding that its commercial relationship with Nvidia limits that pursuit would matter well beyond the two companies. It would establish a template for how regulators view every subsequent deal between the incumbent and a would-be alternative.
The probe also lands as Nvidia's revenue and share of the AI compute stack continue to expand. The company's data-center business has driven successive quarters of record results, and its accelerators remain the default choice for frontier labs training and serving large models. The larger that footprint grows, the lower the bar becomes for antitrust reviewers to treat any bilateral deal as potentially foreclosing competition.
The immediate market question is not whether the DOJ files a case — probes at this stage frequently close without action — but whether the review reshapes how Nvidia negotiates with the next tier of chip startups. A more cautious Nvidia, one that structures its commercial deals to preempt antitrust review, is a different competitive counterparty than the one Groq and its peers have been dealing with. For inference-chip challengers hoping to carve share out of a market Nvidia currently dominates, that shift in negotiating posture may end up mattering more than the outcome of this specific investigation.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
