Warren, Blumenthal Open Antitrust Probe Into Nvidia’s $20 Billion Groq Deal
Senators allege licensing structure was designed to evade merger review while consolidating Nvidia's 80% market dominance in AI accelerators.
U.S. Senators Elizabeth Warren and Richard Blumenthal launched a formal antitrust investigation into Nvidia’s $20 billion licensing agreement with AI chip startup Groq, alleging the deal circumvents merger review requirements while illegally consolidating Nvidia’s control over AI computing infrastructure. The probe, announced Thursday night in a letter to Nvidia CEO Jensen Huang, marks the first congressional enforcement action targeting licensing-as-acquisition strategies that bypass traditional Federal Trade Commission scrutiny.
How the Deal Evaded Review
The December 2025 transaction brought Groq CEO Jonathan Ross and approximately 90% of the startup’s employees to Nvidia while granting the chipmaker a nonexclusive license to Groq’s inference accelerator technology. Critically, Groq did not file the agreement for Antitrust review under Hart-Scott-Rodino Act requirements, which mandate scrutiny for most acquisitions, according to Mercury News reporting.
The structure exploits a regulatory gap: by avoiding a formal acquisition while transferring key personnel and intellectual property, the deal sidesteps the thresholds that would trigger mandatory FTC review. Bernstein analyst Stacy Rasgon noted the arrangement “may keep the fiction of competition alive while effectively neutralizing a rival,” per Tom’s Hardware.
“We are concerned that this takeover could stifle competition, further entrenching Nvidia’s dominance in the AI chip industry and ceding our technological leadership to China.”
— Senators Elizabeth Warren and Richard Blumenthal
The senators’ letter frames the deal as regulatory arbitrage designed to consolidate Nvidia’s position across both AI training and inference markets. Nvidia commands 81% of data center chip revenue according to CNN Business citing International Data Corporation research, while the company faces more competition in inference — where Groq, Advanced Micro Devices, and Cerebras Systems have challenged its H100 and H200 accelerators.
The Reverse Acqui-Hire Phenomenon
Nvidia’s approach follows a pattern established by Amazon, Microsoft, and Google, all of which have reached licensing and hiring deals with AI startups that avoided antitrust filing requirements. FTC Chair Andrew Ferguson signaled enforcement intent in January 2026, stating the agency is “beginning to examine these acqui-hires to make sure they are not an attempt to get around” merger review processes, according to Bloomberg.
The Warren-Blumenthal letter frames the issue in national security terms, arguing that Nvidia’s infrastructure control creates single-point dependency across the U.S. AI ecosystem. “Because its technology is essential for advanced AI development, Nvidia effectively controls which companies can compete in AI, and the entire AI industry is held hostage to Nvidia’s product decisions and priorities,” the senators wrote, according to Mercury News.
Technical Lock-In and Market Power
Nvidia’s dominance extends beyond chip market share into software infrastructure. The company’s CUDA programming framework and NVLink interconnect architecture have become de facto standards for AI development, creating switching costs that reinforce hardware lock-in. The Groq acquisition extends this control into inference workloads, where Groq’s language processing units offered an alternative architecture optimized for model deployment rather than training.
Groq’s inference accelerators use a deterministic architecture that eliminates cache hierarchies, achieving predictable latency for real-time AI applications. Nvidia’s acquisition of this technology and talent base removes a differentiated competitor in the inference market, where workload economics differ substantially from training — inference requires lower precision computation at higher throughput, creating opportunities for specialized chip designs outside Nvidia’s GPU roadmap.
When announcing the deal in December 2025, Jensen Huang emphasized the licensing structure: “While we are adding talented employees to our ranks and licensing Groq’s IP, we are not acquiring Groq as a company,” he told CNBC. Groq’s cloud business continues to operate independently, though the departure of its engineering leadership and most technical staff raises questions about the unit’s viability as a competitive force.
Precedent and Enforcement
The investigation arrives as Congress and regulators confront gaps in antitrust frameworks designed for asset-heavy mergers rather than talent-and-IP transfers. The American Action Forum analysis highlights how licensing structures fall outside Hart-Scott-Rodino notification thresholds, which focus on voting securities and asset transfers rather than human capital mobility.
- Does a $20 billion licensing agreement that transfers 90% of a startup’s workforce constitute a de facto acquisition requiring HSR filing?
- Can antitrust law effectively police competitive harm when market power consolidates through talent raids rather than equity purchases?
- Should inference and training accelerator markets be analyzed separately given distinct technical requirements and customer bases?
- Does national security concern over single-vendor AI infrastructure dependency warrant sector-specific merger standards?
For Nvidia, the investigation adds regulatory uncertainty at a moment when the company faces both competitive pressure in inference markets and geopolitical scrutiny over AI chip exports to China. Shares traded down 1.31% to $176.22 on Thursday as news of the probe circulated.
What to Watch
The Warren-Blumenthal letter demands documentation on deal structure, competitive impact analysis, and internal communications regarding regulatory strategy. Response timelines have not been disclosed. Separately, monitor whether the FTC opens a parallel investigation — Ferguson’s January comments suggest the agency views acqui-hire structures as enforcement priorities, but no formal action has been announced.
Broader implications center on whether licensing-as-acquisition becomes a viable consolidation strategy across the AI infrastructure stack. If the deal survives scrutiny, expect similar structures from hyperscalers acquiring foundational model startups or infrastructure providers. If regulators successfully challenge the arrangement, it would establish precedent for treating talent-and-IP transfers as functional mergers subject to antitrust review — a significant expansion of enforcement scope in labor-intensive technology sectors.
For competitors, the investigation creates a narrow window to establish differentiated inference offerings before Nvidia integrates Groq’s technology into its product roadmap. AMD, Cerebras, and cloud providers building custom silicon face pressure to demonstrate inference economics and technical differentiation that justify switching costs away from Nvidia’s ecosystem. The outcome will determine whether the AI accelerator market remains contestable or calcifies around a single dominant architecture.