Groq Raises 650M: AI Chipmaker Reboots After Nvidia

These deals follow a pattern. A larger player pays investors a hefty IP licensing fee while pulling out the critical talent, leaving the original company intact on paper but hollowed out in practice. Groq is now testing whether what remains can stand on its own.

Ross is a notable loss. He came from Google, where he helped create the Tensor Processing Unit, Google AI chip, and he carried real weight in the AI chip world. He launched Groq a decade ago with another Google engineer, Doug Wightman. Wightman stayed on after the Nvidia deal and stepped up as CEO.

How Groq technology actually works

Groq built a chip it calls a language processing unit, or LPU. Unlike a general-purpose GPU, the LPU is designed specifically for inference, the stage where a trained AI model generates outputs rather than learns. Speed and predictable latency are the whole game in inference, and that is the niche the LPU was built to win.

Groq sold the LPU two ways: as part of a cloud service that developers tap remotely, and as on-premises hardware clusters that companies run inside their own facilities.

Here is the catch. With Nvidia now holding the IP for LPUs, the GPU giant moved fast to compete on Groq own turf. At its GTC event in March, Nvidia announced its own hardware cluster, branded the Nvidia Groq 3 LPX inference hardware system. In other words, Groq signature product now has a version sold by the most powerful company in AI hardware.

For any company that licensed its core IP to a rival, the urgent question is no longer whether the technology is good. It is whether the business around that technology can outrun a competitor that now owns the same blueprint.

The pivot: betting the company on neocloud

Groq response is to lean into its neocloud business rather than try to out-manufacture Nvidia. Neocloud is shorthand for a new wave of cloud providers built specifically to serve AI workloads, as opposed to the general-purpose hyperscalers.

That business was previously run by Madra, after Groq acquired his AI data analytics company, Definitive Intelligence, in 2024. By Groq own account it has grown into a substantial operation.

• 13 data centers across North America, Europe, the Middle East, and APAC.
• More than five million developers served.
• Thousands of AI companies as customers.
• Trillions of tokens processed each week.

Those are the company numbers, not independently audited figures, so treat them as Groq framing of its own scale. Still, the strategic logic is clear: if you can no longer own the chip exclusively, own the service layer that runs on it.

Rebuilding the C-suite from scratch

A pivot of this size needs leadership, and Groq has been hiring replacements for the executives Nvidia took. The new bench is built largely from cloud and enterprise software veterans rather than pure chip designers, a tell about where the company thinks its future lies.

Schuller and Malhotra arrive as a pair. They worked together at Apprenda, an enterprise cloud software company Schuller founded, then co-founded Nuvalence, a software-engineering firm acquired by EY in 2024. That is a software-and-cloud pedigree, which fits a company now selling inference as a service.

What happens over the next 24 to 72 hours

Expect the immediate aftermath to play out in framing and signals rather than hard product news. In the day or two after a raise like this, a few things tend to surface.

1. Watch for whether Groq or its investors put a number on the new valuation. The silence so far is itself a data point, and any leak comparing it to the prior 6.9 billion mark will set the narrative.
2. Look for customer and developer reaction to the neocloud pitch, since the five-million-developer claim is the foundation of the entire strategy.
3. Anticipate competitive positioning against Nvidia inference offerings, now that both companies can build around the same LPU IP.

Beyond the immediate window, the real test is slower. Whether Groq can succeed after almost selling itself depends on how competitive its inference cloud stays now that the key hardware IP is shared with Nvidia. Inference is an area of tremendous demand and heavy VC investment, but it is also drawing intense innovation and new competitors.

There is at least one encouraging precedent. Scale AI CEO Jason Droege told Forbes that business rebounded after Meta did a 14.3 billion not-acqui-hire roughly a year ago, with the company on track for 1 billion in revenue. Survival after these deals is possible. For Groq, the 650 million just raised buys the runway to find out.

Source: TechCrunch