Groq, one of the loudest challengers to Nvidia in the AI inference chip market, announced a $350M round on 17 August 2026. Its valuation fell to $3.5B — half of the $6.9B it commanded in September 2025. This is not routine funding: the company is dropping its own silicon and pivoting to a neocloud model built on Nvidia GPUs.
Key takeaways
- Round: $350M, led by Disruptive, with planned Nvidia participation.
- Valuation $3.5B — down from $6.9B in September 2025 (down round).
- Groq drops its own LPU chips and becomes a cloud operator on Nvidia GPUs.
- 13 data centers, more than 6 million developers and companies using its services.
- Plan to scale from 54 MW to over 200 MW by 2027.
From rival to Nvidia customer
Groq built its brand on the LPU (language processing unit) — custom chips optimized for fast inference of large language models, meant to be a cheaper and faster alternative to Nvidia GPUs. That chapter is closing.
In December 2025 Nvidia hired Groq founder and CEO Jonathan Ross and part of the top team as part of a licensing deal. New chairman and CEO Alex Davis took over, and Groq went from rival to Nvidia customer, building services on its GPUs.
We are building Groq into the world's leading AI inference cloud.
Alex Davis, chairman and CEO, Groq.
A down round despite scale
The new round is smaller and priced lower than the last — in June 2026 Groq raised $650M. Despite the drop, the company reports real operating scale: 13 data centers across North America, Europe, the Middle East and Asia Pacific, more than 6 million developers and companies, and a plan to grow capacity from 54 MW to over 200 MW by 2027.
This is the neocloud model — a GPU infrastructure provider for inference and training, competing with players like CoreWeave.
| Date | Round | Valuation |
|---|---|---|
| Sep 2025 | — | $6.9B |
| Jun 2026 | $650M | — |
| Aug 2026 | $350M | $3.5B |
Why it matters
Groq's pivot shows how hard it is to compete with Nvidia on its own turf — even for a company that built a recognizable, alternative chip.
The founder's move to Nvidia and the switch to its GPUs signal that value is shifting from designing silicon toward operating compute. A down round alongside growing operational scale suggests the market now prices AI infrastructure providers more cautiously than a year ago, despite sustained inference demand.
What's next?
- Scaling capacity from 54 MW to over 200 MW planned by 2027 per the company.
- Nvidia's announced participation will complete Groq's shift from rival to partner-customer.
- The $3.5B valuation will be a reference point for future neocloud rounds.
