2026-11-04 · Fonte：Nvidia Blog
NVIDIA extends Vera Rubin NVL72 with fast token generation for agentic systems, saying the next era of inference hinges on every layer of the AI factory working together.
NVIDIA said it is extending its Vera Rubin NVL72 rack-scale system with fast token generation purpose-built for agentic systems, an announcement that also confirms Groq 3 LPX has reached full production. The company framed the move around a central thesis: the next era of AI inference will not be decided by any single breakthrough chip, network or system, but by how every layer of the AI factory works together.
The announcement, made today, positions the Vera Rubin rack-scale platform as the vehicle for inference workloads driven by autonomous agents. NVIDIA's framing is that agentic applications demand fast token generation sustained across every layer of the AI factory — silicon, networking and system architecture working in concert. Fast token generation is the performance lever the company emphasizes for these deployments, and extending NVL72 for agentic workloads is its concrete response to that thesis.
The scope of the posting also reaches into Spectrum-X and NVLink Fusion, NVIDIA's networking and interconnect technologies, signaling that the inference push is being pitched as a full-stack story rather than a chip launch. That emphasis matters for buyers: NVIDIA is arguing that token throughput for agents is a property of the integrated factory, not of any one component.
For the data center market, the significance is positioning. With Groq 3 LPX in full production and Vera Rubin NVL72 extended for agentic inference, NVIDIA is staking its claim that competitive advantage in the agent era accrues to operators who deploy coordinated stacks — compute, network and system designed as one — rather than those optimizing individual components in isolation. The throughline is clear: NVIDIA wants inference judged at the level of the integrated AI factory, where Vera Rubin NVL72, Groq 3 LPX, Spectrum-X and NVLink Fusion operate as one system.
Tags: NVIDIA · Vera Rubin NVL72 · Groq 3 LPX · AI Inference · Agentic AI · NVLink Fusion
萬安算交所 AIXX · AIXX · Exchange verificado · Ir ao mercado