AI inference company Groq has raised $350 million, at a $3.5 billion valuation.

The company develops its own SRAM-heavy inference chips, which it offers from 13 data centers across North America, Europe, the Middle East, and Asia Pacific. The round was led by Disruptive, with "planned participation" from Nvidia.

The fundraising comes despite Nvidia licensing all of the company's technology on Christmas Eve last year, in a $20bn deal that also saw the GPU giant hire founder and then-CEO Jonathan Ross, along with president Sunny Madra and other members of the Groq team.

Nvidia plans to launch its own Groq-based hardware later this year, and then develop more advanced chips based on Groq, but with an expanded team and utilizing Nvidia IP, including NV Link.

The departure of its leadership and the creation of a major rival has impacted Groq's valuation. In September before the deal, Groq raised $650m in a round again led by Disruptive, but at a $6.9bn valuation.

“We are building Groq into the world’s leading AI inference cloud," Alex Davis, Disruptive CEO and Groq executive chairman, said after this week's round.

"We look forward to continuing our partnership with Nvidia at such an important juncture for the ecosystem. Inference will without a doubt become the largest and most critical layer of AI infrastructure. Our team has unmatched experience operating LPUs at scale and delivering the performance, efficiency, and reliability that the next generation of AI demands. We will be focused on supporting the most important model makers to further contribute to the global growth of innovation and American ingenuity.”

Groq remains an Nvidia Cloud Partner (NCP), certified to design, deploy, and operate Nvidia accelerated computing to Nvidia’s reference architecture and operational standards.

The company said that it hopes to scale from 54MW of compute to 200MW+ in 2027.