Learn about this first
Artificial Analysis published the first results for NVIDIA Vera Rubin NVL72 on the agent-reasoning benchmark AA-AgentPerf. It says the platform achieved up to approximately 5.7 times the total throughput per megawatt of the previous-generation GB300 NVL72, although the multiple varies by model: approximately 5.7 times for DeepSeek V4 Pro, 5.4 times for DeepSeek V4.1 Flash, 3.0 times for GLM-5.2, and 2.8 times for Kimi K3. The data comes from preliminary evaluations and is not yet a final conclusion.
Google Cloud and NVIDIA subsequently announced that Vera Rubin infrastructure is now generally available. A5X instances based on the platform will open to all customers next month, with plans to expand into Europe. A5X integrates Vera Rubin NVL72, ConnectX-9 SuperNICs, Virgo networking, and fourth-generation liquid cooling, and is designed for large-scale agentic inference and reinforcement learning. Google Cloud plans to deploy up to 960,000 Rubin GPUs across multiple sites.
According to Rohan Paul’s account, WPP’s early tests showed that AI agent workloads previously requiring clusters of several servers could run on a single A5X instance.
AI-generated from reports · updated 26 days agoLatest developments
Google Cloud will make Vera Rubin-based A5X instances available to all customers next month
Event progress2
Google Cloud and NVIDIA announce Vera Rubin infrastructure GA
NVIDIA Blog: Google Cloud and NVIDIA Announce Vera Rubin Infrastructure Is Generally AvailableAA Releases Preliminary Evaluation Results for the Vera Rubin NVL72
Artificial Analysis full article: Artificial Analysis publishes the first AA-AgentPerf agent inference benchmark results for NVIDIA Vera Rubin NVL72