Nvidia to deploy Groq AI inference chips in data centers by end of 2025

Nvidia's $20 billion Groq deal moves forward as first benchmark results emerge and rack deployments are promised by year-end.

Nvidia has committed to bringing Groq-powered server racks online before the end of 2025, following a $20 billion deal that positions the company to compete in the fast-growing market for low-latency AI inference. The announcement comes alongside the release of early benchmark results for Groq's third-generation Language Processing Unit (LPU), tested against the Gemma 4 31B model. Groq's LPUs are purpose-built for inference — the process of running a trained AI model to generate outputs — rather than training. Unlike conventional GPUs, they are designed to deliver responses with minimal delay, a characteristic increasingly valued as AI services scale and user-facing applications demand faster replies. The benchmark figures, reported by The Register, reflect a best-case scenario for the Groq 3 LPU rather than typical workloads, a detail that complicates direct comparisons with competing hardware. Nvidia's involvement adds significant manufacturing and distribution weight behind Groq's chip technology, which previously operated at smaller scale.

Why it matters

Low-latency inference is becoming a key battleground in AI infrastructure as cloud providers and enterprises demand faster, cheaper AI responses at scale. A Nvidia-backed rollout of Groq hardware could meaningfully shift the competitive landscape away from GPU-dominated inference.

What's next

Watch for Groq rack availability announcements and independent benchmark results under real-world workloads later in 2025.

Key facts

• Nvidia struck a $20 billion deal involving Groq chip manufacturing and deployment
• Groq-powered server racks are expected to be available to customers before the end of 2025
• Groq's third-generation LPU (Language Processing Unit) is purpose-built for AI inference, not training
• Early benchmarks used Gemma 4 31B as the test model and represent best-case performance conditions
• The deal underscores the rising commercial value of low-latency inference in AI services

Bias & framing notes

CNBC focused on the business and deployment timeline angle, emphasizing the strategic importance of low-latency inference. The Register took a more technical and skeptical framing, flagging that the benchmark results reflect a best-case scenario — a caveat that CNBC did not surface. Neither source provided independent benchmark verification or detailed deal terms.

NewsClear — neutral news & congressional tracking · Bill of the Week