Nvidia Begins Production of Groq 3 LPX Racks, Plans Deployment This Year

Nvidia (NASDAQ: NVDA) said Monday that its Groq 3 LPX rack is in full production and will be deployed at neocloud operator Nebius (NASDAQ: NBIS) later this year alongside the company’s Vera CPUs and Rubin GPUs.

The rack combines 256 Groq 3 chips and is designed for low-latency AI inference. Nvidia said the system can generate 3,400 tokens per second, citing a benchmark from Artificial Analysis. Groq’s architecture incorporates 500 megabytes of SRAM directly on each chip to reduce memory bottlenecks. The Groq processors are manufactured by Samsung, while Taiwan Semiconductor Manufacturing Co. (NASDAQ: TSM) produces Nvidia’s GPUs.

Nvidia is targeting the technology at latency-sensitive workloads, including AI agents and coding applications where faster response times are required. Senior director Dion Harris said the performance can allow companies serving AI tokens to offer premium service tiers for customers seeking lower latency. The Groq processors are intended to complement rather than replace Nvidia’s GPUs. GPUs remain capable of handling both AI training and inference, while chips such as Groq are focused primarily on the decode portion of inference workloads. “This isn’t about replacing GPUs,” Harris said. “It’s about using the right price, right processor for the right part of the workload.”

Other chipmakers are also developing systems aimed at faster inference. Advanced Micro Devices (NASDAQ: AMD) said earlier this year that it would integrate its rack-scale systems with Cerebras (NASDAQ: CBRS) chips. OpenAI’s Ultrafast mode, which is powered by Cerebras, promises throughput of 750 tokens per second. Nvidia obtained the Groq technology through a $20 billion purchase of assets from the chip startup in December. The transaction was Nvidia’s largest acquisition on record, and the start of Groq 3 LPX production marks the commercialization of technology acquired through that deal.

The company is also increasing shipments of its Vera Rubin systems after production began earlier this year. Nvidia introduced Vera Rubin and Groq 3 LPX in March, when Chief Executive Jensen Huang projected $1 trillion in cumulative sales from the current Blackwell generation and Vera Rubin systems through 2027.

Huang said at the March event that he planned to dedicate 25% of data center capacity intended for coding applications to Groq chips, with the remainder allocated to Vera Rubin.

About Nvidia

Nvidia (NASDAQ: NVDA) was founded in 1993 and is headquartered in Santa Clara, California. The company develops graphics processors, accelerated computing systems, networking technology and software used in artificial intelligence, data centers, gaming and professional computing.

At the time of publishing, Stocks.News does not hold positions in companies mentioned in the article.