Groq Raises $350M to Finish Turning Into a Neocloud
Groq raised $350 million at a $3.5 billion valuation, led by Disruptive with Nvidia planning to participate, per TechCrunch on August 17. TechCrunch frames it as a Series A, which sounds absurd for a company founded in 2016 until you accept that this is effectively a new company. In December 2025 Nvidia took founder-CEO Jonathan Ross and key chip talent through a $20 billion licensing deal. In June what remained raised $650 million to start the pivot. Today's Groq has stopped building its own LPU chips and runs an inference and training cloud on Nvidia GPUs across 13 data centers in North America, Europe, the Middle East and Asia Pacific, 54 megawatts now, targeting over 200 by 2027, serving more than 6 million developers.
A spokesperson insists $3.5 billion is not a down round from the prior $6.9 billion, it is "establishing a new valuation for the post-Nvidia-licensing-deal version of Groq." Translated: roughly half the old valuation walked out the door with Ross and the silicon IP, and investors are now pricing what stayed, which is capacity, contracts and a customer base.
For anyone building agents the lesson is about the stack, not this one company. Groq's original pitch was differentiated silicon making inference an order of magnitude faster. That differentiation got absorbed directly into Nvidia, and the business that survives is megawatts of interchangeable GPU cloud. Agent workloads are about to run on infrastructure where the moat is power contracts and utilization, not architecture. Two neocloud rounds in two months says investors believe demand will fill every rack anyway.
techcrunch.com/2026/08/17/groq-raises-350m-to-fuel-its-pivot-from-ai-chips-to-neocloud/
← Back to all articles
A spokesperson insists $3.5 billion is not a down round from the prior $6.9 billion, it is "establishing a new valuation for the post-Nvidia-licensing-deal version of Groq." Translated: roughly half the old valuation walked out the door with Ross and the silicon IP, and investors are now pricing what stayed, which is capacity, contracts and a customer base.
For anyone building agents the lesson is about the stack, not this one company. Groq's original pitch was differentiated silicon making inference an order of magnitude faster. That differentiation got absorbed directly into Nvidia, and the business that survives is megawatts of interchangeable GPU cloud. Agent workloads are about to run on infrastructure where the moat is power contracts and utilization, not architecture. Two neocloud rounds in two months says investors believe demand will fill every rack anyway.
techcrunch.com/2026/08/17/groq-raises-350m-to-fuel-its-pivot-from-ai-chips-to-neocloud/
Comments