Senators Elizabeth Warren and Richard Blumenthal sent a letter to Nvidia on Friday questioning the company’s $20 billion licensing deal with AI chip startup Groq, warning the arrangement could “stifle competition” and “further entrench Nvidia’s dominance in the AI chip market,” Bloomberg reported.

The letter marks the senators’ second action targeting Nvidia’s AI ambitions this week. Earlier, Warren and Blumenthal sent a separate letter pressing for oversight of xAI’s Pentagon contracts. The two-front campaign suggests a broader Senate push to scrutinize how AI infrastructure deals are reshaping market competition.

What the Deal Actually Is

Nvidia announced the Groq deal at GTC 2026, licensing Groq’s LPU (Language Processing Unit) technology for $20 billion. LPUs are purpose-built for inference workloads, processing tokens at speeds that conventional GPUs can’t match at equivalent power consumption. The Mercury News characterized the deal as Nvidia’s “next Mellanox” move, referencing the 2019 networking acquisition that gave Nvidia control of data center interconnects.

Nvidia pushed back on the framing. According to Yahoo Finance reporting, the company clarified it did not acquire Groq outright but licensed the technology, maintaining that the arrangement preserves Groq’s independence as a company.

The distinction matters legally. An outright acquisition would trigger Hart-Scott-Rodino antitrust review thresholds. A licensing deal, even a massive one, occupies a regulatory gray zone that the senators’ letter appears designed to challenge.

Why This Matters for AI Agents

LPUs are specifically optimized for the inference workloads that power AI agent frameworks like OpenClaw. Every time an agent processes a prompt, evaluates a tool call, or generates a response, it runs inference. GPU clusters handle this today, but LPU technology promises significantly lower cost-per-token at scale.

If Nvidia locks up LPU technology through exclusive licensing, the economics of running AI agents shift. Competitors building inference hardware — AMD, Intel, startups like Cerebras and SambaNova — would face a wider gap against Nvidia’s combined GPU + LPU stack. For OpenClaw operators paying cloud inference bills, the deal’s competitive implications flow directly into their monthly costs.

Market Reaction

NVDA dropped 1.65% on Friday’s session, with Yahoo Finance data showing the stock under pressure from the Senate scrutiny. The broader AI infrastructure sector also saw weakness, with Microsoft trading at $382.69 — well below its 50-day moving average of $422 — amid its own OpenAI-related disputes.

The senators have not indicated whether they plan to introduce legislation or request formal FTC review of the Groq arrangement. For now, the letter functions as a shot across the bow: Washington is watching how AI chip deals reshape the competitive landscape, and $20 billion licensing agreements won’t fly under the radar.