The Department of Justice has opened an antitrust investigation into Nvidia's roughly $20 billion arrangement with AI chip startup Groq, questioning whether the deal was deliberately structured as a license to sidestep merger notification rules. The probe, which began shortly after Groq and Nvidia announced the agreement in December 2025, centers on a fundamental legal question: Is this transaction actually an acquisition disguised as a licensing agreement?

What Exactly Did Nvidia and Groq Agree To?

In late December 2025, Nvidia and Groq announced what they officially called a non-exclusive licensing agreement covering Groq's inference chip technology. However, reporting from CNBC described the deal as Nvidia acquiring most of Groq's assets for roughly $20 billion, nearly three times Groq's most recent private valuation of $6.9 billion. This discrepancy between how the companies characterized the deal and how the media reported it is precisely what triggered regulatory scrutiny.

As part of the arrangement, Groq co-founder and CEO Jonathan Ross, who co-invented Google's Tensor Processing Unit (TPU), and Chief Operating Officer Sunny Madra, along with several senior engineers, joined Nvidia. Groq's Chief Financial Officer Simon Edwards became the new CEO, and the company continued operating GroqCloud, its commercial inference API service, as an independent entity.

Why Is the DOJ Concerned About How This Deal Was Structured?

The core regulatory concern involves the Hart-Scott-Rodino (HSR) Act, a federal law requiring companies to notify the government before completing acquisitions above a certain financial threshold. In late 2025, that threshold was $126.4 million. Nvidia and Groq did not file an HSR notification before closing their deal. The question regulators are asking: Did Nvidia deliberately label the transaction a "license" to avoid triggering this filing requirement?

Licensing agreements do not automatically require HSR notification, but regulators can investigate whether an arrangement is economically equivalent to an acquisition despite its legal label. The Federal Trade Commission (FTC) and Department of Justice (DOJ) began scrutinizing similar AI industry structures in 2024, questioning whether arrangements that transfer talent, intellectual property, and competitive assets should require antitrust review even without a formal corporate transaction. The Nvidia-Groq deal represents a scaled-up version of that same concern.

Senators Elizabeth Warren and Richard Blumenthal raised identical concerns in a March 2026 letter to Nvidia, written after the DOJ probe had already begun. The investigation became public on September 9, 2026, when The New York Times first reported the probe; Bloomberg confirmed the investigation the following day.

What Are the Key Factors Regulators Will Examine?

• Technology Transfer: Nvidia gained access to Groq's LPU (Language Processing Unit) intellectual property under broad terms, though the license is non-exclusive, meaning Groq can license the technology to other companies.
• Talent Migration: The departure of Groq's co-founder and COO to Nvidia represents a significant loss of leadership, though Groq remains independent under new management.
• Market Competition: Groq lost key competitive assets and leadership to its largest potential customer, potentially reducing competitive pressure in the inference chip market.
• Deal Characterization Mismatch: The gap between CNBC's reporting of an asset purchase and the companies' description of a license creates ambiguity about the transaction's true economic substance.

What Outcomes Are Most Likely?

DOJ officials have told Bloomberg they do not currently expect to seek an unwinding of the transaction. Reversing a licensing arrangement and talent migration after more than eight months would be operationally unprecedented. Instead, regulators are likely to pursue one of several alternative paths.

The most probable outcome is that the investigation closes without enforcement action. If the DOJ does pursue penalties, a monetary fine for HSR non-compliance is considered more likely than structural remedies. Such fines accrue per day and would represent a fraction of one percent of a single quarter's revenue for Nvidia, which reported approximately $96 billion in quarterly revenue for the period ended July 2026.

Less likely but more strategically significant outcomes include behavioral remedies such as compulsory licensing terms or restrictions on exclusivity in future products, or constraints on how Nvidia can integrate Groq's talent and intellectual property. A hold-separate order, which would limit integration of the Groq team's work, would represent the most disruptive scenario for Nvidia's inference roadmap.

Why Did Nvidia Pay $20 Billion for Groq in the First Place?

The investigation reveals something equally important as the legal question: Nvidia's management considered Groq's inference technology threatening enough to warrant a $20 billion pre-emptive acquisition. Nvidia's dominance in AI training through its graphics processing units (GPUs) is uncontested, but inference, the process of running trained models in production at scale, operates under different competitive dynamics.

Inference rewards latency (speed of response), power efficiency, and cost per token over raw computational power. Groq's deterministic single-stream LPU architecture is specifically optimized for latency. Independent benchmark data, including assessments from Artificial Analysis, showed Groq's LPU achieving faster time-to-first-token latency and per-query output speed than Nvidia's H100 GPU on certain inference workloads before the December 2025 deal. The LPU trades aggregate batched throughput for per-user latency, the metric that matters most in real-time consumer and enterprise applications.

By securing Groq's intellectual property and leadership team, Nvidia simultaneously absorbed a credible inference rival's technical expertise and extended its inference roadmap beyond its current GPU and NVLink rack-scale system portfolio. Even if regulatory penalties and compliance friction increase the deal's total cost, the strategic calculus remains largely intact for Nvidia's long-term competitive positioning.

How to Monitor This Investigation as an Investor or Industry Observer

• Watch for Formal Charges: Monitor whether the DOJ issues a formal complaint or enforcement action. The absence of such action within the next 6 to 12 months would suggest the investigation is closing without penalties.
• Track Behavioral Remedy Announcements: If Nvidia announces new licensing restrictions, exclusivity limitations, or constraints on Groq team integration, this signals regulatory settlement negotiations.
• Follow Groq's Competitive Performance: Observe whether GroqCloud maintains technological parity with Nvidia's inference offerings or falls behind, which would indicate the degree to which Nvidia's acquisition of Groq's talent and IP has affected competitive dynamics.
• Assess Inference Market Developments: Monitor whether competitors like Etched, Tenstorrent, Cerebras, and AMD's inference-optimized MI series gain market share, which could accelerate if Nvidia faces integration constraints on Groq technology.

The probe is a regulatory headline rather than an existential threat to Nvidia, but it confirms that the company's own management judged the inference market competitive enough to justify a $20 billion response. The investigation will determine whether that response violated antitrust law in its execution.

" }