Groq confirmed a $650M funding round on Monday, six months after Nvidia signed a $20B non-exclusive licensing agreement for its chip technology and hired away founder and CEO Jonathan Ross, president Sunny Madra, and a chunk of the engineering team. The new round follows a $750M raise in September that valued the AI chipmaker at $6.9B. Groq did not disclose the new valuation.
The December deal with Nvidia was structured as a licensing arrangement rather than an acquisition, which let investors profit handsomely while the buyer absorbed the talent and IP it wanted. What was left at Groq was the brand, the cloud business, and the open question of what the company does next when its rival now owns the rights to its core hardware.
Ross, who came from Google and helped create the Tensor Processing Unit, co-founded Groq with fellow Google engineer Doug Wightman a decade ago. Wightman stayed on after the Nvidia deal and stepped into the CEO seat. Groq's flagship was a chip it called the language processing unit, an inference-focused part sold through a cloud service or as an on-prem hardware cluster.
Key facts
- 01Groq closed a $650M round on Monday, roughly six months after Nvidia's $20B non-exclusive licensing deal stripped out its CEO and core IP.
- 02The company was last valued at $6.9B following a $750M round in September; the new valuation was not disclosed.
- 03Groq's neocloud now spans 13 data centers across four regions and serves over 5M developers, processing trillions of tokens each week.
- 04Nvidia announced its own Groq 3 LPX inference hardware system at GTC in March, built on the IP it licensed from Groq.
- 05New COO Alan Rice joins from xAI and Meta; Sinclair Schuller and Rakesh Malhotra come on as CTO and CPO after selling Nuvalence to EY in 2024.
With Nvidia now holding the LPU IP, the GPU giant announced its own hardware cluster — the Nvidia Groq 3 LPX inference system — at its GTC event in March. That left Groq competing directly against a much larger company selling effectively the same silicon design under the same brand prefix.
Groq's response is to lean on its neocloud business, which Madra had been running after Groq acquired his AI data analytics startup Definitive Intelligence in 2024. That operation now spans 13 data centers across North America, Europe, the Middle East, and APAC, serves over five million developers and thousands of AI companies, and processes trillions of tokens each week, the company said.
The executive bench is being rebuilt around that pivot. Alan Rice joined as COO from xAI and Meta, following a career in the U.S. Navy. Sinclair Schuller came in as CTO and Rakesh Malhotra as CPO; the two previously worked together at Apprenda, the enterprise cloud company Schuller founded, then co-founded Nuvalence, a software-engineering firm EY acquired in 2024. Malhotra spent roughly a decade on Microsoft's cloud products before that.
Inference is the part of the AI stack drawing the most current investment, on the bet that model training spend eventually plateaus while query volume keeps compounding. That demand is real, but so is the competition: hyperscalers, Nvidia itself, and a long list of inference-specialized startups are all chasing the same workloads, often with newer silicon and lower per-token pricing.
Groq is not the first company to try to keep going after a big-IP-license-and-poach. Scale AI's CEO Jason Droege has said business has rebounded after Meta's $14.3B not-acqui-hire about a year ago, and that the company is tracking to $1B in revenue. The template exists; the execution is what matters.
The harder question for Groq is differentiation. When the cloud underneath your service runs hardware whose design your largest competitor now also ships, the moat has to come from software, latency, pricing, geographic coverage, or developer experience — not the chip itself. Five million developers and 13 data centers is a real footprint, but it's a footprint Nvidia's own cloud partners can match.
What Groq has bought with $650M is time and a management team that has done enterprise software before. The strategic bet is that an independent inference cloud, focused on developers and token throughput rather than chip sales, can outrun a hardware vendor that has to balance the same product against every other customer it supplies. It's a thinner business than the original Groq pitch, but it's a business — and in the current market for inference capacity, that's enough to keep playing.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
