Groq has closed a $650 million round to relaunch itself as an AI inference cloud, months after a Nvidia deal reshaped the company entirely.
Last December, Nvidia paid out Groq's investors and hired away the company's founder, effectively gutting the startup that had spent years positioning itself as Nvidia's loudest chip-level rival. What's left is a reconstituted Groq betting that its Language Processing Units — purpose-built chips designed specifically for AI inference — can out-perform general-purpose GPUs on the workloads that matter most. The $650 million raise is the capital needed to rebuild that case from the ground up.
The timing is pointed. The inference market has exploded as companies shift from training massive models to running them cheaply and fast — and that's exactly the gap Groq's architecture was designed to fill. If purpose-built inference silicon has a window, this is it; the question is whether a company that lost its founder and its original investor base can execute quickly enough to matter.
Groq isn't the only one making this argument — Cerebras and Tenstorrent are running similar plays — but a $650 million war chest buys time to find out whether the bet holds.