Google Cloud and NVIDIA launch A5X instances with Vera Rubin NVL72 for agentic AI and reinforcement learning

Google Cloud and NVIDIA have announced general availability of NVIDIA Vera Rubin infrastructure on Google Cloud through the new A5X instances. According to the announcement posted on the NVIDIA blog on October 21, 2026, the instances will become generally available the following month. An expansion to Europe is also planned. The offering is an integrated platform for large-scale training, inference and reinforcement learning.

The A5X instances combine NVIDIA Vera Rubin NVL72 systems, NVIDIA ConnectX-9 SuperNICs, Google Cloud's Virgo Network and the fourth generation of Google Cloud's advanced liquid cooling technology. They are designed for demanding agentic inference and reinforcement learning workloads. Google Cloud plans to scale its Vera Rubin-based infrastructure to as many as 960,000 Rubin GPUs spread across multiple sites.

The technical reasoning behind the announcement is that learning through interaction requires a continuous loop: generating experience, evaluating results and updating models. That calls for infrastructure that can sustain the entire process at scale. A5X brings Vera Rubin into Google Cloud's AI Hypercomputer as an integrated system that combines compute, networking and storage for this kind of experience-based learning.

The first customer to access A5X was Ineffable Intelligence, a London-based technology company founded by David Silver, the creator of AlphaGo. The company is using the platform to advance large-scale reinforcement learning and to develop what it calls “superlearner” systems, which learn continuously from experience. Silver, the company's founder and CEO, says this demands a great deal of compute and a tight coupling between generating experience and training on it, so that each improvement in the agent quickly shapes its next experience. He highlights the high-bandwidth connectivity between the Vera CPU and the Rubin GPUs, and among the Rubin GPUs within the rack.

WPP, the London-based marketing and communications company, has also demonstrated its creative agent reference architecture on Vera Rubin-based A5X instances. It uses NVIDIA Nemotron models and NVIDIA OpenShell, part of the NVIDIA OpenShell Reference Architecture for secure AI agents. The agents turn conversational requests into multi-step creative workflows, such as video generation, upscaling and redesign, with human approval before execution. OpenShell provides governed sandboxes for agent execution, helping control access to sensitive brand assets and creative tools. In WPP's initial tests, workloads that would normally require a multi-node cluster ran on a single A5X instance.

The announcement also addresses confidentiality. Google Cloud will be one of the first providers to offer rack-scale NVIDIA Confidential Computing on bare metal instances, such as the A4X Max with NVIDIA GB300 NVL72. In addition, Confidential Computing is now generally available on G4 VMs with NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs, a flexible option for confidential inference and fine-tuning. Vertebrae, Polaris Cloud and Confidential Core AI use confidential G4 for conversational privacy, verifiable agent execution and policy-governed enterprise intelligence.

On energy efficiency, Google Cloud and NVIDIA are exploring NVIDIA DSX MaxLPS capabilities for the Vera Rubin-based infrastructure. The goal is to optimize performance per watt and expand capacity within data centers' available power budgets.

The text is an NVIDIA corporate post and includes no performance figures, pricing or exact launch dates beyond “next month.” Capacity figures, such as the 960,000 GPUs, are plans, not existing deployments.