Archive · 25 Aug 2026
Nvidia Puts Groq 3 LPX Racks Into Full Production
Nvidia is making a new kind of computer rack that uses Groq chips.
These chips are designed to answer AI questions very quickly.
Faster answers can make AI agents feel more responsive, especially when helping with coding.
Nvidia bought Groq’s assets for $20 billion in December.
Each rack contains 256 Groq 3 chips.
Nvidia says one rack can produce up to 3,400 tokens, or small pieces of text, per second in a benchmark.
The racks will be used alongside Nvidia’s Vera processors and Rubin graphics processors.
Nebius is expected to begin receiving the systems later this year.
The Groq chips are meant to work alongside GPUs rather than replace them.
Nvidia says its Groq 3 LPX rack has entered full production.
The racks will be deployed with Vera CPUs and Rubin GPUs at Nebius, beginning later this year.
Nvidia acquired assets from Groq for $20 billion in December.
Each LPX rack packages 256 Groq 3 chips and can deliver up to 3,400 tokens per second, according to an Artificial Analysis benchmark.
Groq chips target low-latency AI inference, complementing rather than replacing GPUs used for broader AI workloads.
- Who
- Nvidia, Groq, and neocloud provider Nebius are the main companies involved.
- What
- Nvidia has put its Groq 3 LPX racks into full production for low-latency AI inference.
- When
- Nvidia announced the production milestone on Monday; deployments are expected to begin later this year.
- Where
- The racks are expected to be deployed at Nebius.
- Why
- The systems are intended to provide faster AI inference and make AI agents more responsive.
- Acquisition value
- Nvidia acquired assets from Groq for $20 billion in December.
- Rack configuration
- Each Groq 3 LPX rack contains 256 Groq 3 chips.
- Reported performance
- Nvidia says a rack can deliver up to 3,400 tokens per second, based on an Artificial Analysis benchmark.
- Deployment customer
- Nebius is expected to deploy the racks later this year.
- Chip manufacturing
- Samsung manufactures Groq chips, while Taiwan Semiconductor Manufacturing Company manufactures Nvidia GPUs.
- Primary use
- Groq chips focus mainly on the decode phase of AI model serving and low-latency inference.
- Related systems
- The racks will be deployed alongside Nvidia’s Vera central processors and Rubin graphics processors.