By ByteBot
On Christmas Eve, while most people were preparing for the holidays, Nvidia quietly announced a $20 billion deal that might eliminate one of its last serious competitors in AI inference. The company framed it as a “non-exclusive licensing agreement” with chip startup Groq—but when you hire the founder, president, and key engineers from a competitor, is that really just licensing?
The Deal That Walks Like an Acquisition
Nvidia is paying $20 billion for what it calls a “non-exclusive license” to Groq’s inference technology. For context, Groq was valued at $6.9 billion just three months earlier. Why pay nearly triple for licensing rights when you could acquire the whole company for less?
The answer becomes clearer when you look at who’s joining Nvidia: Jonathan Ross, Groq’s founder and the inventor of Google’s Tensor Processing Unit; Sunny Madra, Groq’s president; and a significant portion of the engineering team. Groq will continue as an “independent company” under a new CEO, but here’s the question—can a chip company compete without its founding technical genius and core team?
Bernstein analyst Stacy Rasgon put it bluntly: “Antitrust would seem to be the primary risk here, though structuring the deal as a non-exclusive license may keep the fiction of competition alive.” That phrase—”fiction of competition”—cuts to the heart of what’s happening. This deal is structured to look like licensing while functionally eliminating a competitor.
The Microsoft-Inflection Playbook
This isn’t an isolated case. In March 2024, Microsoft hired almost all of Inflection’s team while paying $650 million for “licensing rights.” The FTC opened an investigation. UK regulators ruled it was a merger, just structured differently. The European Commission agreed it constituted a “concentration” under merger law.
In response, the FTC created new rules in February 2025 requiring disclosure for “talent-based concentrations”—deals where hiring key personnel and licensing IP effectively transfers control without a formal acquisition. The Nvidia-Groq deal follows the exact same pattern regulators flagged as problematic.
The timing raises eyebrows too. As Rasgon observed, “They’re so big now that they can do a $20 billion deal on Christmas Eve with no press release and nobody bats an eye.” A $20 billion transaction that potentially eliminates a competitor should spark intense scrutiny, not be treated as routine holiday news.
Why Inference Matters
Nvidia has a near-monopoly on AI training chips—the hardware used to build AI models. But the larger opportunity is AI inference, the process of running those models in production. The inference market hit $106 billion in 2025 and is projected to reach $255 billion by 2030, with 80% annual growth.
Here’s Nvidia’s problem: they face much more competition in inference than in training. AMD’s Instinct MI300X competes directly. Cerebras built radical wafer-scale processors. And Groq developed LPU (Language Processing Unit) technology that delivered inference up to 18 times faster than traditional GPUs while using one-tenth the energy.
With Groq’s technical leadership now joining Nvidia, that competitive pressure just decreased significantly. AMD and Cerebras remain as alternatives, but the field is narrowing.
The Talent That Matters
Jonathan Ross isn’t just another chip designer. He invented Google’s TPU as a 20% project and designed its core architecture. After leaving Google, he founded Groq in 2016 with a mission to “eliminate artificial scarcity in AI compute.” The company’s LPU delivered on that promise—750 tokens per second on Llama 2 7B, powering AI applications for over 2 million developers.
Now Ross will be building Nvidia’s next-generation inference chips instead of competing against them. This is the real story: Nvidia isn’t just buying technology, it’s acquiring one of the few people who has successfully challenged their chip dominance—twice.
What Happens Next?
Regulators could investigate, following the Microsoft-Inflection precedent. The new FTC rules that took effect in February 2025 were specifically designed to catch deals like this. Multiple regulatory bodies have already established that hiring a company’s key talent while licensing their technology constitutes a merger, even without a formal acquisition.
But enforcement is uncertain. Big Tech companies have deep resources and political influence. The deal’s structure—non-exclusive licensing with the target company continuing operations—provides plausible deniability. And Nvidia’s scale makes even a $20 billion transaction feel routine.
The stakes extend beyond this single deal. If companies can eliminate competitors by hiring their talent and licensing their IP without merger scrutiny, we’ll see more consolidation, less innovation, and developers locked into monopolistic ecosystems.
The Bottom Line
This deal walks like an acquisition, talks like an acquisition, and eliminates a competitor like an acquisition—but it’s labeled “licensing” to dodge antitrust review. Groq might continue operating on paper, but without Jonathan Ross and its core engineering team, it’s difficult to see how it remains a meaningful competitor.
Nvidia’s strategy is clear: dominate AI training chips (achieved), dominate AI inference chips (in progress). Each competitor that disappears moves that goal closer to reality. For developers building on AI infrastructure, the choice of chip providers is shrinking. For the AI industry, the question is whether anyone will challenge this consolidation before it’s too late.
