Nvidia's $20B Groq Deal Is Under Senate and FTC Fire — Is It an Acquisition in Disguise?

On Christmas Eve 2025, Nvidia announced what it called a “non-exclusive technology licensing agreement” with Groq — the AI inference startup whose Language Processing Unit chips had become the fastest publicly available inference hardware on the market. The price: $20 billion, the largest deal in Nvidia’s history. Jonathan Ross, Groq’s founder and CEO, and many of its senior engineers were simultaneously joining Nvidia. Groq’s office doors would stay open; some core operations would continue. It was technically not an acquisition.

Four months later, that technical distinction is at the center of one of the most consequential antitrust fights in the history of the semiconductor industry.

What Nvidia Bought — and What It Didn’t Admit It Bought

To understand why the deal is generating so much regulatory heat, you have to understand what Groq actually built. Unlike Nvidia’s general-purpose GPUs — which are enormously powerful but relatively expensive and power-hungry for pure inference workloads — Groq’s LPU (Language Processing Unit) was architected from the ground up for the specific task of running trained AI models quickly and efficiently at the moment of user interaction.

The LPU delivered inference speeds that, for many model sizes and architectures, outperformed Nvidia’s H100 and H200 by a factor of three to five on throughput-per-watt. For large-scale AI application providers — companies spending tens of millions of dollars monthly on inference compute — that efficiency gap represented a meaningful cost advantage. By late 2025, OpenAI was reportedly evaluating Groq’s chips as a more economical alternative to Nvidia for certain categories of inference workloads.

Those talks between OpenAI and Groq were, by multiple accounts, shut down within weeks of Nvidia’s licensing deal announcement. Jonathan Ross and the engineers who understood the LPU architecture most deeply were now Nvidia employees.

The Senate Investigation

On March 20, 2026, Senators Elizabeth Warren (D-MA) and Richard Blumenthal (D-CT) sent a letter to Nvidia CEO Jensen Huang asking pointed questions about the deal’s structure and intent. The senators gave Nvidia until April 3 to respond, and their language was blunt: “By licensing its technology and hiring its most important employees, NVIDIA has effectively acquired Groq in all but name.”

The letter outlined the senators’ concern that the transaction was deliberately engineered to evade the Hart-Scott-Rodino (HSR) Act’s notification requirements, which mandate that acquisitions above a certain transaction value be reported to the FTC and DOJ for antitrust review. By structuring the deal as a license rather than a direct acquisition of assets or stock, Nvidia avoided triggering that review.

“NVIDIA already controlled approximately 90% of the market for GPUs — the high-end chips used to develop and deploy AI,” the senators wrote. “Groq specializes in developing inference chips that are more energy-efficient than NVIDIA’s.” The senators argued that the combination stifles competition at exactly the moment when viable alternatives to Nvidia’s GPU monopoly were beginning to emerge.

Subsequently, Senators Warren, Wyden (D-OR), and Blumenthal wrote to federal regulators calling for a broader investigation into what they termed “reverse acqui-hire” deals — a pattern they argued was emerging across the AI industry, in which dominant incumbents neutralize potential competitors by buying their technology and talent without triggering acquisition review thresholds.

The FTC’s Posture

The Federal Trade Commission has been monitoring the transaction. Sources close to the process have indicated that the FTC’s technology enforcement team is examining whether the Nvidia-Groq arrangement, despite its nominal structure as a license, should be reclassified as an acquisition under the agency’s substantive authority.

The key legal test is functional: if Nvidia obtained the practical equivalent of ownership — exclusive or near-exclusive control over the technology, and the human capital necessary to develop and extend it — antitrust regulators have historically argued that form should not dictate substance. The non-exclusive nature of the license provides Nvidia with some cover, since Groq technically retains the right to license the LPU technology to others. But if the employees who could execute such a license have all joined Nvidia, the non-exclusivity may be more form than substance.

If the FTC reclassifies the deal as a reportable acquisition, Nvidia could face a range of remedies including forced divestiture of the license, an unwinding order requiring the return of Groq’s technical staff, or substantial civil penalties. The agency could also use the investigation to establish precedent for how similar deals are evaluated in the future.

Nvidia’s Defense

Nvidia has maintained that the deal is a legitimate licensing agreement, emphasizing its non-exclusive nature and disputing the characterization that it eliminated meaningful competition from Groq. Company spokespeople have noted that Groq as a legal entity continues to operate, that its LPU technology remains licensable, and that Nvidia’s acquisition of talent is standard practice in a competitive market for specialized engineering skills.

Jensen Huang, in public appearances since the deal was announced, has framed Nvidia’s inference strategy as an expansion of its platform — arguing that by incorporating LPU design insights into its roadmap, Nvidia is improving outcomes for customers, not restricting competition.

The company also has a structural argument: the AI chip market is more competitive today than it was two years ago, with AMD’s MI400 series gaining ground, Google’s custom TPUs expanding, Amazon’s Inferentia handling large fractions of AWS AI inference, and a new wave of European and South Korean startups building specialized silicon. In that context, Nvidia argues, characterizing the LPU acquisition as a competition-ending move overstates its market impact.

The Broader Stakes

The Nvidia-Groq situation is not an isolated case. It sits at the intersection of several colliding forces in 2026: an AI industry where a handful of companies have achieved extraordinary market concentration; a political environment in which AI policy has become a first-tier issue; and an antitrust enforcement apparatus that has been asked to apply 20th-century frameworks to 21st-century market structures.

The outcome of the FTC’s review will have implications beyond Nvidia. If regulators establish that talent-plus-license deals above a certain value require HSR notification, it will significantly constrain how dominant AI companies can acquire nascent competition. That would be a fundamental change to the deal environment in Silicon Valley, where “acqui-hire” arrangements — buying a company primarily for its people while nominally preserving the entity — have long been a standard tool for absorbing potential rivals.

For Groq’s former competitors and customers, the message is already clear regardless of how regulators ultimately rule: building a sufficiently differentiated AI infrastructure product now carries the risk of being absorbed by the companies that benefit most from its non-existence as a standalone enterprise. Whether that dynamic constitutes an antitrust violation is a question the FTC and potentially the courts will have to answer.

The hearing is ongoing. Nvidia’s response to the Senate letter, and the FTC’s next move, are expected in the coming weeks.