Groq raises $650M for its inference cloud
Groq raised $650M on June 22 to expand inference capacity, six months after Nvidia licensed its chip IP for about $20B and hired away its founders.
Groq raised $650 million on June 22 to fund its inference cloud, led by existing investors Disruptive and Infinitum. The company did not give a new valuation; its last was $6.9 billion in September 2025. The money goes toward inference capacity, with a stated goal of 200 megawatts by 2027.
The raise closes a strange six months. Around December 2025, Nvidia paid roughly $20 billion in a non-exclusive license for Groq’s Language Processing Unit technology and hired away founder and chief executive Jonathan Ross, president Sunny Madra, and other staff, then shipped its own Nvidia Groq 3 LPX hardware in March. Doug Wightman, who started Groq with Ross at Google, is now chief executive, alongside new operating, technology, and product chiefs drawn from xAI, Meta, and Microsoft. Groq says it now runs 13 data centers across four regions, serves more than five million developers, and processes trillions of tokens a week.
Why this matters
If you use Groq for fast, cheap inference, the headline is stability: it kept its customers while losing its founders, and it is funding more capacity rather than selling for parts. But a provider that just had its leadership and core IP bought out is a dependency worth watching. Price it the way you would any single vendor, and keep an exit ramp ready. It is also another sign that inference, not training, is where the model-access business is now being fought.