Groq spent years trying to build chips that could beat Nvidia at its own game. Now Nvidia is one of the investors funding what Groq has become instead.
Groq raised $350 million this week, led by investment firm Disruptive with Nvidia planning to participate, at a $3.5 billion valuation. That's half of what the company was worth in September, when it raised money at $6.9 billion.
From Rival to Customer
The valuation drop traces back to December, when Nvidia struck a $20 billion licensing deal that brought Groq founder and CEO Jonathan Ross, along with much of the company's top engineering talent, over to Nvidia. Groq's investors were paid out as part of that arrangement.
A company spokesperson told TechCrunch the new valuation shouldn't be read as a down round, but rather as the starting point for what they called the "post-Nvidia-licensing-deal version of Groq."
That version looks nothing like the original. Groq built its own AI chips, called LPUs, specifically to compete with Nvidia on inference, the compute-intensive work of running AI models in real time rather than training them. After losing its star team, the company shifted into operating data centers built on Nvidia's own hardware instead.
Building the Neocloud
Groq kicked off that pivot with a $650 million raise in June. The company now runs 13 data centers spanning North America, Europe, the Middle East and Asia Pacific, serving more than 6 million developers and enterprises.
The plan is to scale fast. Groq intends to grow its capacity from 54 megawatts to more than 200 megawatts by 2027, and says the new funding will go toward serving customers who need medium and large clusters of Nvidia-powered compute for both training and inference work.
"Inference will without a doubt become the largest and most critical layer of AI infrastructure," said Alex Davis, Groq's chairman and CEO of Disruptive.
That bet isn't unique to Groq. It's the same bet CoreWeave, Lambda and Nebius are all making.
A Familiar Nvidia Pattern
Whether that bet pays off is still an open question. CoreWeave, the most established of the group, posted strong second-quarter revenue growth and landed major contracts with Meta and Anthropic. But investors remain uneasy about its heavy debt load, steep capital spending, and exposure to hardware that loses value fast.
Groq's own financials aren't public yet, so it's too early to know which camp it falls into.
What is clear is the shape of the relationship. Nvidia supplies the GPUs that power CoreWeave, Lambda and Nebius, and it's investing billions of dollars directly into several of those same companies as they race to build capacity. Groq's new round puts it inside that same loop, buying Nvidia hardware with money that partly comes from Nvidia itself.
That structure has drawn scrutiny elsewhere in the AI industry too, most recently around Nvidia's financing arrangements for OpenAI's data center buildout. The pattern is the same each time: the chip supplier becomes a lender and investor to its own customers, which makes demand for its hardware look stronger than it might if that money weren't circulating back through the system.
None of that makes Groq's bet wrong. Inference demand is genuinely growing as more companies move AI from pilot projects into production. But it does mean Groq's success is now tied more tightly to Nvidia's fortunes than it was when the two were rivals.
What This Means for Miami
Groq's target scale, jumping from 54 megawatts toward more than 200, dwarfs the AI infrastructure projects currently under construction in South Florida, where facilities like Metrobloks' Sweetwater data center are measured in the tens of megawatts rather than hundreds.
That gap matters for how Miami's investors and enterprise buyers should think about the neocloud category. The economics of running a Groq-scale operation, and the risk that comes with it, are fundamentally different from the smaller, regionally focused inference infrastructure being built closer to home.
Miami companies evaluating neocloud partners for their own AI workloads should be asking the same questions investors are asking about CoreWeave: not just how much capacity a provider has, but how much debt and hardware depreciation risk is sitting behind it.