Nvidia has entered a nonexclusive licensing agreement with Groq, an AI chip company, to access Groq’s technology for running AI models in real-world applications. As part of the deal, Groq founder and CEO Jonathan Ross will join Nvidia, along with other senior leaders and staff, to help integrate the licensed technology.

The agreement reflects growing demand for AI hardware designed not just to train models, but to run them efficiently once they are deployed in consumer and enterprise products.

Under the agreement, Jonathan Ross will join Nvidia to help integrate Groq’s licensed technology. Groq’s president and several staff members will also move to Nvidia. Groq’s finance chief, Simon Edwards, will take over as the company’s new CEO.

Importantly, this is not an acquisition. Groq will continue operating as an independent company, including its cloud business, GroqCloud, which allows software developers to access Groq’s AI processing without purchasing physical chips or servers.

In the AI industry, inference refers to what happens when a trained AI model is put to work — for example, answering questions, generating responses, or making predictions using new data. This is the phase users interact with every day.

Groq’s chips, which it calls language processing units, are designed specifically for this kind of work. According to Ross, their design uses embedded memory, allowing the chips to be produced and deployed faster while using less power than traditional graphics processing units (GPUs), which are more energy-intensive and typically used for training large models.

As AI adoption grows, inference has become a major bottleneck, pushing companies to look for hardware that can scale efficiently without sharply increasing energy costs.

The Nvidia–Groq agreement follows a broader industry pattern in which large technology companies license AI technology while bringing in startup leadership.

Meta invested $14 billion in Scale AI, leading Scale’s CEO to join Meta’s AI leadership. Alphabet’s Google hired top executives from Character.AI while licensing its technology. Microsoft struck a similar arrangement with Inflection AI.

These deals allow established companies to move faster by combining external innovation with internal expertise, without fully acquiring younger firms.

Groq was last valued at $6.9 billion following a $750 million funding round in September. Investors included BlackRock, Neuberger Berman, Cisco, and Samsung.

The company says its chips are designed, fabricated, and assembled in North America, working with manufacturing partners that include Samsung.

Ross previously worked at Google, where he helped develop the processors that later became known as tensor processing units (TPUs), and studied under AI pioneer Yann LeCun.

Nvidia remains the dominant supplier of advanced AI chips and has become the world’s most valuable company as demand for AI hardware has surged. The company has also increased the pace of its AI chip releases to maintain its lead.

At the same time, competition is intensifying. Cloud providers such as Google and Amazon are developing their own chips, while major Nvidia customers — including OpenAI and Meta — are designing custom hardware to reduce dependence on external suppliers.

Nvidia shares were flat in after-hours trading following the announcement, though the stock is up more than 35% year to date.

Q: What did Nvidia announce?
A: Nvidia announced a nonexclusive licensing agreement with Groq to use its AI technology for running AI models, while also bringing Groq’s founder and key staff into Nvidia.

Q: Is Nvidia acquiring Groq?
A: No. Groq will remain an independent company. Nvidia is licensing technology and hiring leadership to support integration.

Q: What is AI inference in simple terms?
A: AI inference is when a trained AI model is used to produce results — such as answering a question or making a prediction — in real-world situations.

Q: Why is inference becoming more important than training?
A: As AI products move into everyday use, the cost, speed, and energy required to run models at scale matter more than training them once.

Q: What happens to GroqCloud?
A: GroqCloud will continue operating independently, providing AI processing services to developers.

This deal highlights how AI’s focus is shifting from training ever-larger models to running those models reliably, affordably, and at scale. For businesses and consumers, inference efficiency determines whether AI features feel instant and accessible — or slow and expensive.

By licensing Groq’s technology rather than acquiring the company outright, Nvidia is prioritizing flexibility and speed. The move allows Nvidia to strengthen its inference capabilities while preserving optionality in a market where cloud providers, enterprise customers, and chipmakers are all racing to reduce dependence on any single hardware supplier while reducing energy costs. As AI adoption accelerates, how models run may matter just as much as how they are trained.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing, image, and idea-generation support from ChatGPT, an AI assistant. However, the final perspective and editorial choices are solely Alicia Shapiro’s. Special thanks to ChatGPT for assistance with research and editorial support in crafting this article.