Groq has raised $350 million as it continues a major change in direction: from building AI chips to operating the cloud and data center infrastructure that companies use to run AI workloads.
The round, led by investment firm Disruptive with planned participation from Nvidia, values Groq at $3.5 billion. That figure is below the $6.9 billion valuation Groq reached last September, before Nvidia hired founder and CEO Jonathan Ross and other senior talent as part of a licensing deal.
From AI chipmaker to neocloud provider
Groq originally focused on its own chips, called LPUs, or language processing units. The company positioned those chips around inference, the compute used to run AI workloads in real time.
That made Groq part of the broader race to supply infrastructure for artificial intelligence. Its original strategy centered on competing with Nvidia in a specific and increasingly important part of AI computing.
But the company’s direction changed after it lost its star team. Groq shifted from being a pure AI chipmaker into a cloud and data center provider that operates Nvidia systems. In practical terms, the remaining Groq business has become an Nvidia customer rather than only a challenger.
The new financing follows a $650 million round in June that was intended to begin that pivot. Groq now says it will use the fresh funds to serve “those seeking usage of medium and larger sized clusters of Nvidia accelerated computing for training and inference.”
The valuation reset
The $350 million round gives Groq a $3.5 billion valuation. That is significantly lower than the $6.9 billion valuation reported last September.
A Groq spokesperson told TechCrunch the company does not view the change as a down round. Instead, the spokesperson described it as a new valuation for the “post-Nvidia-lincensing-deal version of Groq.”
That distinction matters because the company being valued now is not exactly the same company investors valued last September. Groq’s business model, leadership picture, and infrastructure role have all shifted after the Nvidia licensing deal and the departure of Jonathan Ross and other top talent.
The company is now presenting itself less as a stand-alone AI chip contender and more as a provider of AI infrastructure services. Its path depends on delivering access to powerful GPUs and data center capacity for developers, enterprises, and AI-native companies.
What Groq is building now
Groq says it currently operates 13 data centers across North America, Europe, the Middle East, and Asia Pacific. It also says it serves more than 6 million developers, enterprises, and AI-native companies.
The company intends to scale from 54 megawatts to more than 200 megawatts by 2027. That expansion reflects the infrastructure-heavy nature of the neocloud business. To serve training and inference demand, Groq needs large amounts of computing capacity and the facilities to run it.
Alex Davis, Groq’s chairman and CEO of Disruptive, framed the strategy around inference. “We are building Groq into the world’s leading AI inference cloud,” Davis said in a statement. “Inference will without a doubt become the largest and most critical layer of AI infrastructure.”
That focus is consistent with Groq’s original emphasis on inference, even though the way the company plans to serve the market has changed. Instead of only pushing its own LPUs, Groq is now operating Nvidia systems as part of a cloud and data center business.
The neocloud question
The demand side of the story is clear from the source: enterprises are scaling AI workloads, and inference is in high demand. Groq’s fundraising and expansion plans are built around that market need.
The harder question is whether neoclouds can produce strong enough returns after making the large investments required to build capacity. The business requires expensive infrastructure, and that creates pressure to turn customer demand into durable financial results.
CoreWeave offers one example of the opportunity and the concern. It reported strong second-quarter revenue growth and recently landed major contracts, including with Meta and Anthropic. At the same time, investors have remained concerned about high capital expenditures, heavy reliance on debt, exposure to rapidly depreciating hardware, and whether growth can become free cash flow.
Groq’s financials remain private. That makes it difficult to assess how its pivot is performing financially. What is visible is its deeper position inside Nvidia’s AI infrastructure ecosystem.
That position is not unusual among neoclouds. Nvidia supplies the GPUs powering clouds from CoreWeave, Lambda, and Nebius, while also investing billions into some of those companies as they race to build more capacity.
Why the raise matters
The $350 million raise gives Groq more capital for a strategy that now depends on scale, capacity, and access to Nvidia accelerated computing. It also marks another step in the company’s attempt to redefine itself after a major change in its original chip-focused plan.
Groq is still tied to inference as a central theme. What has changed is the route. The company began with its own LPUs and a direct challenge around real-time AI compute. It is now pursuing that market through neocloud services, data centers, and Nvidia systems.
For customers seeking larger clusters for training and inference, Groq wants to be part of the infrastructure layer that makes AI workloads run. For investors, the question is whether that infrastructure layer can justify the cost of building it.