0

Non-exclusive agreement transfers key talent and technology while Groq remains independent

In sum – what we know:

• Deal structure: Nvidia and Groq announced a non-exclusive licensing agreement valued at approximately $20 billion according to media reports.
• Talent transfer: Groq founder Jonathan Ross and President Sunny Madra are moving to Nvidia alongside key engineers, bringing specialized expertise in inference chip design and compiler optimization.
• Operational status: Groq will continue operating as an independent company, with Simon Edwards stepping in as the new CEO to oversee the GroqCloud platform going forward.

Nvidia and chip startup Groq have announced a non-exclusive licensing agreement reportedly worth around $20 billion. The announcement turned heads immediately — not just because of the price tag, which clocks in at nearly three times Groq’s $6.9 billion valuation from just months earlier, but because of what’s actually changing hands. On paper, it’s a licensing deal, but in practice, with Groq’s founder, president, and top engineers heading to Nvidia, the deal looks a lot more like an acquisition wearing a different outfit.

A hybrid deal

This transaction doesn’t necessarily fit neatly into any standard category. The non-exclusive licensing structure means Groq can still license its technology elsewhere, even as Nvidia gains access — a setup that looks nothing like a traditional acquisition.

The $20 billion figure, drawn mostly from CNBC and Reuters reporting since neither company has disclosed official numbers, represents a hefty premium. Groq had raised $750 million at a $6.9 billion valuation in September 2025, so the implied near-3x jump in value stands out.

What’s taking shape here is something of a hybrid arrangement. Nvidia walks away with intellectual property rights and a major talent infusion, while Groq continues to exist as its own entity. Bank of America analysts summed it up as “surprising, strategic, expensive, offensive, defensive, complementary.”

Acqui-hire

The talent moving to Nvidia might end up mattering more than the technology itself. Jonathan Ross, who founded Groq in 2016 after helping build Google’s Tensor Processing Unit, is bringing his deep expertise to Nvidia. President Sunny Madra and a group of key engineers are making the move alongside him.

The knowledge these people carry is highly specialized. Groq’s team designed their hardware, software, and compiler as a unified system, with particular expertise in static scheduling compiler technology — an approach where every computational cycle gets mapped out ahead of time with zero runtime flexibility. That’s a fundamentally different design philosophy than what drives Nvidia’s GPUs, and it’s not something you can just recreate from scratch.

The arrangement has sparked some criticism, though, particularly around the employees staying behind. Leadership and select engineers get the Nvidia offer, while everyone else remains with a restructured Groq under new CEO Simon Edwards. That split raises real questions about what the future holds for those who weren’t part of the move.

The inference war: Buying the competition?

Nvidia’s strategic rationale comes into sharper focus when you consider what’s happening in inference. Training remains Nvidia’s stronghold, but inference, where trained models actually run and generate outputs, is shaping up to be a more contested space. AMD and startups like Cerebras have been gaining ground, and both Groq and AMD recently announced inference-focused projects in the Middle East. That’s not to mention the first-party chips that the likes of Google, Amazon, and others are designing.

Groq’s SRAM-based architecture presents a particular challenge to Nvidia’s playbook. Built specifically for real-time, latency-critical applications like voice agents and autonomous systems, Groq’s chips shine exactly where Nvidia’s HBM-based GPUs struggle to keep up. 

Locking in access to this technology, and the team that created it, lets Nvidia accomplish several things at once. It takes a potential competitor off the board before Groq could scale into a real independent threat. It adds complementary capabilities for latency-sensitive workloads without leaning entirely on costly HBM solutions. And it brings in engineering talent that would have taken years to cultivate internally, if it could be done at all.

Conclusions

Supply chain considerations factor in here too. Nvidia’s January 2025 Form 10-K disclosed that the company had “paid premiums, provided deposits, and entered into long-term supply agreements” to lock in future capacity. HBM memory costs aren’t coming down anytime soon, and specialized inference capability gives Nvidia a way to reduce its exposure on certain workloads.

There’s been speculation that Google may have been eyeing Groq — a move that would have bolstered Google’s TPU efforts and created a much stronger Nvidia challenger. Whether that interest was real or rumored, the possibility alone may have pushed Nvidia to move faster.

The deal obviously looks like Nvidia neutralizing competition before it could grow into something more threatening. The non-exclusive licensing framework and Groq’s continued independence provide regulatory and legal cover, but the practical outcome, like top talent departing and rival technology secured, resembles acquisition far more than partnership. For Nvidia, this hybrid structure may deliver the best of both worlds. For everyone else in the AI chip market, it’s another concentration of power in an industry where one player already dominates.