With Groq 3 LPX in Full Production, NVIDIA Extends Vera Rubin Inference for Agents

NVIDIA announced on August 24 that it is extending its Vera Rubin NVL72 rack-scale system with fast token generation for agentic systems, and that Groq 3 LPX has entered full production. The company says the next era of AI inference will be defined not by a single breakthrough chip, network, or system, but by how every layer of the AI factory works together.

The announcement positions the Vera Rubin platform as an inference engine for AI agents, which require fast, continuous token generation to sustain multi-step reasoning and tool use. Extending NVL72 with fast token generation is aimed directly at that workload pattern, according to the company.

The second part of the announcement is that Groq 3 LPX has reached full production. Moving from roadmap to shipping scale matters for customers planning data-center deployments around the Vera Rubin generation, because it means the capability is available now rather than promised for later.

NVIDIA framed the update around its broader AI factory strategy, in which infrastructure is designed and built as a complete system rather than a collection of individual accelerators. The same announcement also highlighted the networking and interconnect layers of the AI factory, including NVLink Fusion and Spectrum-X.

Why it matters: agentic AI is shifting the bottleneck in data centers from training throughput to inference latency. Systems that can sustain fast token generation at rack scale are becoming a strategic differentiator for both NVIDIA and the cloud providers building on its platforms.

What to watch next: when Vera Rubin NVL72 systems with the new capabilities become broadly available, and how they perform on real agent workloads compared with the previous generation. For enterprises building agent infrastructure, the production status of Groq 3 LPX is the more immediately actionable signal.