News
Google Cloud and NVIDIA say Vera Rubin-based A5X instances will reach general availability next month, with Europe expansion planned.
Google Cloud and NVIDIA said their Vera Rubin-based A5X instances will reach general availability next month, adding a new cloud platform for training, inference and large-scale reinforcement learning. The companies also said they plan to expand availability to Europe, while framing the offering as part of Google Cloud’s AI Hypercomputer stack.
Evidence and context
In the announcement published by NVIDIA on Oct. 21, 2026, A5X is described as combining NVIDIA Vera Rubin NVL72, NVIDIA ConnectX-9 SuperNICs, Google Cloud Virgo Network and Google Cloud’s fourth-generation advanced liquid-cooling technology. The companies said the system is intended for demanding agentic inference and reinforcement learning workloads at scale.
Google Cloud previously announced A5X on April 22, 2026 as bare-metal GPU-accelerated instances powered by NVIDIA Vera Rubin NVL72 within AI Hypercomputer. In the background material reviewed with the announcement, Google’s Virgo Network scaling targets were described as up to 80,000 GPUs at one site and as many as 960,000 Rubin GPUs across multiple sites. Those figures refer to cluster-level capability, not to a single instance.
The announcement said Ineffable Intelligence, a London-based company founded by David Silver, was the first customer to access A5X. According to the statement, the company is using the platform to advance reinforcement-learning systems that learn continuously from experience. David Silver said the setup needs both substantial compute and tight connectivity between generating experience and training on it, and that Vera Rubin’s compute performance and high-bandwidth connectivity support that workflow.
WPP also used A5X to demonstrate a creative-agent reference architecture with NVIDIA Nemotron models and NVIDIA OpenShell. In the company’s initial tests, workloads that typically required a multi-node cluster ran on a single A5X instance, according to the announcement. The material also said OpenShell provides governed sandboxes for agent execution to help control access to sensitive brand assets and creative tools.
Google Cloud said it will be among the first cloud providers to offer NVIDIA Confidential Computing at rack scale on bare metal instances such as A4X Max instances powered by NVIDIA GB300 NVL72. It also said Confidential Computing is becoming generally available on G4 VMs powered by NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs, for confidential inference and fine-tuning. The companies said customers including Vertebrae, Polaris Cloud and Confidential Core AI are using Confidential G4 for privacy, verifiable agent execution and policy-governed enterprise intelligence.
The reviewed materials do not provide a complete per-instance specification table for A5X, including vCPU count, host RAM, GPU count per instance, GPU memory allocation, local SSD capacity, per-instance network bandwidth or hourly pricing. General availability is scheduled for November 2026, but no exact launch day was supplied.
Be the first to comment