Groq Raises $350M to Expand Its AI Inference Cloud

Groq has secured another $350 million as the former AI-chip challenger accelerates a dramatic transformation into an AI infrastructure and inference-cloud company.

The Series A round, announced on August 17, was led by Disruptive, with planned participation from Nvidia, and values the new-look Groq at $3.5 billion. Combined with the $650 million Groq raised in June, the company has brought in $1 billion in fresh capital in less than two months.

The funding tells only half the story. Groq is no longer positioning itself primarily as the startup trying to beat Nvidia by selling a rival AI accelerator. It is increasingly positioning itself alongside Nvidia as a specialized cloud provider focused on one of AI’s biggest infrastructure problems: running models at scale.

A $3.5 billion valuation after being worth $6.9 billion

The new financing gives Groq a valuation roughly 49% below the $6.9 billion figure attached to its September 2025 funding round, when the company raised $750 million while still being viewed mainly as an independent AI-chip challenger.

That headline could make the latest deal look like a major down round.

Groq, however, told TechCrunch it does not view the financing that way. The company argues that the $3.5 billion figure establishes a valuation for the post-Nvidia-licensing version of Groq, rather than directly repricing the business investors valued last year.

That distinction matters because Groq changed significantly after December 2025.

Nvidia entered into a non-exclusive agreement to license Groq’s inference technology, while Groq founder Jonathan Ross, president Sunny Madra and other members of the team moved to Nvidia. Groq remained an independent company rather than being fully acquired.

The deal effectively separated much of Groq’s original chip story from the company that is now raising money.

Groq is becoming an infrastructure company

Groq’s new strategy is much more capital intensive.

Instead of concentrating primarily on designing and commercializing its own Language Processing Units, or LPUs, the company is building a globally distributed inference cloud capable of operating both its inference technology and Nvidia accelerated computing.

Groq currently says it operates 13 data centers across North America, Europe, the Middle East and Asia-Pacific. Its platform now serves more than six million developers, Fortune 500 enterprises and thousands of AI-native companies, which collectively generate trillions of tokens each week.

The company plans to increase infrastructure capacity from approximately 54 megawatts today to more than 200 megawatts in 2027. That would represent nearly a fourfold increase in available power capacity if the target is reached.

The new $350 million will help finance that expansion, particularly larger clusters of Nvidia accelerated computing designed for AI training and inference.

Nvidia has gone from rival to partner, and potential investor

Perhaps the clearest evidence of Groq’s transformation arrived five days before the funding announcement.

On August 12, Groq became an official Nvidia Cloud Partner, giving it certification to design, deploy and operate Nvidia accelerated-computing infrastructure according to Nvidia’s reference architecture and operating standards.

Groq said the certification also clears the way for it to install the latest Nvidia accelerated-computing technology inside its existing data centers.

That creates a striking reversal.

Groq originally built its LPU architecture as an alternative to GPUs for high-speed AI inference. Today, the company that once competed with Nvidia at the silicon layer is increasingly building a cloud business that will also purchase, deploy and operate Nvidia systems.

Nvidia is now expected to participate in Groq’s latest financing round as well.

The bigger bet is on inference

Groq’s new business is built around the belief that the next phase of AI spending will increasingly move from training models toward inference, the computing required every time a deployed AI model generates an answer, image, recommendation or action.

Groq said in June that it expects inference eventually to require 15 to 20 times more compute than model training, although that is the company's own estimate rather than an independently established industry forecast.

The company is therefore betting that developers and enterprises will increasingly need specialist infrastructure optimized for running models continuously rather than merely training them.

That puts Groq into the expanding “neocloud” market alongside companies offering specialized AI compute as an alternative or complement to traditional hyperscalers.

There is still a difficult economic question behind that opportunity. Building hundreds of megawatts of AI infrastructure requires enormous capital, and specialized cloud providers must generate enough utilization and cash flow to justify continuous spending on data centers and rapidly evolving computing hardware. TechCrunch noted similar concerns around the broader neocloud model, including high capital expenditure and hardware depreciation.

For Groq, however, investors are financing the pivot aggressively.

Within eight weeks, the company has secured $1 billion, expanded from more than five million developers in June to reporting more than six million in August, joined Nvidia’s cloud partner program and laid out plans to increase infrastructure capacity beyond 200 MW.

Groq’s first chapter was about building a chip fast enough to challenge the GPU establishment. Its second is shaping up very differently: building the cloud infrastructure that sits around those chips, including Nvidia’s, and betting that inference becomes one of the largest computing markets of the AI era.