Summary
Google Cloud and NVIDIA say A5X instances, built around NVIDIA Vera Rubin NVL72 systems, are expected to reach general availability the month after their October 21, 2026 announcement. The platform is designed for agentic AI and large-scale reinforcement learning workloads.
Google Cloud and NVIDIA say their Vera Rubin-based A5X cloud instances are expected to reach general availability in November 2026, following the companies’ October 21 announcement. They plan to expand availability to Europe. The platform combines NVIDIA Vera Rubin NVL72 systems with NVIDIA ConnectX-9 SuperNICs, Google Cloud’s Virgo Network and its fourth-generation advanced liquid-cooling technology for agentic AI and large-scale reinforcement learning workloads.
Google Cloud says it plans to scale its Vera Rubin-based infrastructure to as many as 960,000 Rubin GPUs across multiple sites. That is a stated infrastructure plan; the announcement describes A5X availability as forthcoming, rather than already generally available.
An integrated platform for learning through interaction
Reinforcement learning improves a model or agent through cycles of experience, evaluation and updates. For systems that learn by interacting, those stages need to work in close succession: new experience is generated, outcomes are assessed, and the system is updated before the next round. A5X brings Vera Rubin into Google Cloud’s AI Hypercomputer, combining compute, networking and storage in one cloud platform for that workflow.
Ineffable Intelligence, a London-based company founded by AlphaGo creator David Silver, was the first customer to access A5X, according to NVIDIA. The company is using the platform for large-scale reinforcement learning and its work on “superlearner” systems that learn continuously from experience. Silver said the combination of compute and high-bandwidth connections between CPUs and GPUs, and among GPUs within a rack, supports the rapid exchange between generating experience and training on it.
Early demonstrations on A5X
WPP has demonstrated a creative-agent architecture on A5X using NVIDIA Nemotron models and NVIDIA OpenShell. The agents can turn conversational requests into multistep creative workflows, including video generation, upscaling and restyling, with human approval before execution. OpenShell provides governed sandboxes for agent execution, intended to help control access to sensitive brand assets and creative tools.
WPP said its initial tests ran workloads that typically required a multi-node cluster on a single A5X instance. This is an early, company-reported result for its particular creative workflow. The announcement presents the broader platform as infrastructure for demanding AI workloads, with November general availability still ahead.