Key Highlights

  • Nvidia's new Groq 3 LPX chip is now in full production, built to make AI systems respond faster, especially for tasks like coding.
  • In testing, the chip generated 3,400 words per second, the fastest speed ever recorded for the AI model it was tested on.
  • Cloud company Nebius will be the first to offer this chip to its customers, later this year.
  • The chip comes from technology Nvidia gained by buying AI startup Groq for about $20 billion last December, the biggest purchase Nvidia has ever made.

A Faster Chip for AI That Needs to Think and Act Quickly

Nvidia (NASDAQ:NVDA) said its new chip, called Groq 3 LPX, is now in full production and ready for use. The chip is built to make AI systems respond much faster, which matters most for AI "agents," programs that don't just answer questions but actually complete tasks like writing and testing code. The faster a chip can generate responses, the quicker these AI agents can finish their work without making people wait.

In independent testing, Groq 3 LPX generated 3,400 words per second while running a well-known AI model, the fastest speed ever recorded for that model. Nvidia said this lets AI agents complete coding tasks in minutes instead of hours, and the chip responds about four times faster than the closest competing option. Nvidia CEO Jensen Huang said this is another major step forward in making AI faster and more efficient, at a time when the world's Demand for AI computing keeps growing.

Where the Technology Came From

This chip is the first major product to come from Nvidia's roughly $20 billion purchase of AI startup Groq in December, the largest Acquisition Nvidia has ever made. The chip includes a large amount of fast onboard memory, which helps it avoid the slowdowns that usually happen when a chip has to constantly fetch data from elsewhere. Nvidia packs 256 of these chips into each server rack it sells.

A senior Nvidia executive explained that this kind of speed matters because cloud companies can charge more money for faster AI responses, especially to customers who need instant results. He was clear that this chip isn't meant to replace Nvidia's main AI chips, which do the heavy lifting of Training and running AI models. Instead, this new chip handles one specific, speed-critical part of the process, working alongside Nvidia's other chips rather than competing with them.

Nebius Is First in Line, But Competition Is Heating Up

Nebius, a Cloud Computing company, will be the first to offer this new chip to its own customers, planning to add it to its systems later this year. Right behind Nebius, the company Groq itself, now largely owned by Nvidia, is also expected to be an early user.

This isn't happening in a vacuum. Other chip companies are racing to build faster AI response times too. A rival chip maker recently teamed up with a company called Cerebras to compete in this exact space, and even OpenAI has announced a fast-response mode built using Cerebras technology.

Conclusion

This new chip shows Nvidia successfully turning a major acquisition into an actual product just months after buying the company. With demand growing fast for AI that can act quickly, not just answer questions, and competitors also racing to solve the same speed problem, how well this chip performs once it's fully rolled out with Nebius will be an important signal to watch. Nvidia is also expected to report its latest Earnings soon, which may offer more details on how this new chip is performing in the real world.