Groq's $350M raise tests whether neoclouds can turn Nvidia hardware into cash

Groq just raised $350M at a $3.5B valuation, roughly half the $6.9B mark it carried in September 2025, and the company wants the headline to be about its future, not its markdown. The round, led by Disruptive with planned participation from Nvidia, funds a neocloud business that didn't exist a year ago: a fleet of data centers renting out AI compute capacity to other companies, built on Nvidia's chips rather than Groq's own.

A Groq spokesperson told TechCrunch the new figure isn't a down round but a "new valuation for the post-Nvidia-licensing-deal version of Groq." That licensing deal is why the pivot exists. Groq became an Nvidia customer in the same transaction. The original thesis was custom AI chips called Language Processing Units, or LPUs, built to compete with Nvidia on inference, the step where a trained model returns an answer to a prompt. That business was effectively wound down. What remains is a company operating Nvidia systems at scale: 13 data centers across North America, Europe, the Middle East, and Asia Pacific, already serving more than 6 million developers and enterprises.

The June $650M round kicked off the neocloud buildout. The fresh $350M is meant to push capacity from 54 megawatts today to more than 200 megawatts by 2027, enough compute to train and run large AI models for paying customers.

Neoclouds, AI infrastructure companies renting out GPU and inference capacity to other businesses, sit in the same economic position as older data-center operators, with one sharp difference. Their inventory is Nvidia hardware that loses value much faster than the steel-and-concrete of a traditional data center. Capital goes in up front, the depreciation clock starts immediately, and cash has to come back before the next chip generation resets the math.

CoreWeave, the only public neocloud comp in the bundle, already shows what the unit-economics risk looks like in public. Per TechCrunch, its Q2 results carried strong revenue growth and Meta and Anthropic contracts, but investors pressed on capex, debt, fast-depreciating hardware, and free-cash-flow conversion. The market is rewarding revenue and punishing balance sheets. Groq's path to a similar profile is what Disruptive and Nvidia are underwriting with this round.

Bloomberg independently confirmed the $3.5B figure on the same day. Nvidia's participation is described as "planned" because the terms aren't closed at announcement, a flag worth holding if the round reshapes before close. Groq's prior round history runs through Tracxn's funding tracker.

The test is concrete. If a public neocloud, CoreWeave being the first read, shows durable free-cash-flow conversion despite the capex load, Groq's $3.5B reset looks less like a haircut and more like a wholesale repricing. If the next chip cycle forces another round of depreciation write-downs before cash arrives, the neocloud thesis itself gets repriced alongside it.

Groq says its 200-megawatt build starts immediately. The market gets its first independent read in CoreWeave's next earnings cycle. Groq's own read comes in stages as the capacity comes online through 2027. The data to confirm or kill the bet is already public, in another company's quarterly report.