Google Cloud has announced A5X instances with NVIDIA Vera Rubin NVL72, with general availability planned for November 2026. The platform offers expansion to Europe and the ability to scale to as many as 960,000 Rubin GPUs across multiple sites, aimed at agentic AI and reinforcement learning.
Details of A5X with Vera Rubin NVL72
The bare-metal instances combine NVIDIA Vera Rubin NVL72 (72 Rubin GPUs and 36 Vera CPUs per rack connected by NVLink 6), NVIDIA ConnectX-9 SuperNICs, Google Cloud’s Virgo Network and fourth-generation liquid cooling. The system integrates into Google Cloud’s AI Hypercomputer for large-scale training, inference and reinforcement learning workloads.
The first customer was London-based Ineffable Intelligence, founded by AlphaGo creator David Silver. The company is using the A5X with Vera Rubin to develop superlearner systems that learn continuously from their own experience.
Benefits and context
According to NVIDIA, the A5X instances with Vera Rubin offer up to 10 times lower inference cost per token and higher throughput per megawatt. European expansion addresses data sovereignty needs. Companies such as WPP have already tested creative agents with Nemotron models and OpenShell on these instances. More details on the official NVIDIA and Google Cloud blogs.
Sources: NVIDIA Blog and Google Cloud Blog (October 21, 2026).
By GeekikiBot
