Nvidia just tried to sidestep antitrust review with a $20 billion licensing deal for AI chip startup Groq, and Warren and Blumenthal aren't buying it.

The Summary

  • Nvidia structured a $20 billion Groq deal as a licensing agreement, possibly to dodge merger review requirements
  • Senators Warren and Blumenthal are investigating whether the deal illegally consolidates Nvidia's dominance in AI compute
  • The probe signals regulators may start scrutinizing "creative" deal structures that achieve acquisition outcomes without triggering antitrust thresholds

The Signal

Nvidia controls roughly 80% of the AI accelerator market. Groq built custom chips specifically designed to run inference workloads faster and cheaper than Nvidia's GPUs, making them one of the few credible challengers to Nvidia's inference dominance. A $20 billion licensing deal that gives Nvidia effective control over Groq's technology without technically acquiring the company looks like monopoly maintenance with extra steps.

The licensing structure matters. Traditional acquisitions over $119 million trigger Hart-Scott-Rodino Act filing requirements, giving the FTC and DOJ time to review competitive impacts. Licensing deals typically fly under that radar. But if the licensing terms give Nvidia exclusive rights to Groq's chip designs, manufacturing processes, or roadmap direction, it achieves the same market concentration as an outright purchase. Warren and Blumenthal are essentially asking: does calling it a licensing deal make it legal, or just clever?

This probe arrives as the AI infrastructure layer solidifies. The companies that control compute in 2026 will shape which agents get built, who can afford to run them, and what economic models survive the agent economy buildout. Nvidia already holds that position for training workloads. If they lock down inference through deals like Groq, they control both ends of the AI supply chain. No competitor emerges, no price competition materializes, and the cost of running agents stays artificially high.

The timing also matters. We're past the "let innovation bloom" phase of AI development. Regulators now see consolidation plays clearly. Expect more scrutiny on talent acquisitions, exclusive cloud partnerships, and any deal structure that concentrates power in AI compute.

The Implication

If you're building in the agent economy, watch whether this deal gets blocked or restructured. Nvidia's pricing power directly impacts your cost structure for running agents at scale. A competitive inference chip market means lower costs and more deployment options. A Nvidia-dominated one means you're a price taker.

For investors, this signals a new era of antitrust enforcement in AI infrastructure. Creative deal structures won't necessarily save you from regulatory review. The FTC is learning to look past labels and examine actual competitive effects.

Source: Bloomberg Tech