The Pulse
Google Cloud Sets November Launch for Vera Rubin A5X Instances
Google Cloud and NVIDIA announced A5X cloud instances built on Vera Rubin NVL72 systems, with general availability planned for November 2026. The companies say the infrastructure will support AI agents and reinforcement learning, with Googl
AI.info Team · Written with AI · fact-checked ·
Google Cloud and NVIDIA announced on October 21 that Google’s A5X instances, powered by NVIDIA Vera Rubin NVL72 systems, are due to become generally available in November. The launch plan also includes expanding availability to new European regions for customers with data-residency and sovereignty requirements. The instances are not generally available yet; the companies describe the November date as a target.
The announcement, made at NVIDIA GTC Berlin, positions A5X as infrastructure for agentic workloads, including inference and large-scale reinforcement learning. Google is pairing NVIDIA’s rack-scale compute with its own networking and cloud infrastructure, aiming to support systems that generate experience, assess results and use those results to update models.
Google’s A5X combines Rubin GPUs and Virgo networking
A5X instances use NVIDIA Vera Rubin NVL72 systems alongside NVIDIA ConnectX-9 SuperNICs and Google Cloud’s Virgo Network. NVIDIA’s announcement also names Google’s fourth-generation advanced liquid-cooling technology as part of the system. Together, the components are intended to keep compute, network traffic and cooling coordinated across demanding workloads.
Google says Virgo will support clusters of up to 80,000 Rubin GPUs at a single site and as many as 960,000 GPUs across multiple sites. Those figures describe the scale the network is designed to support, not capacity that customers can access today. The distinction matters: Google has set a launch window for A5X, while its stated multi-site scale describes a broader infrastructure plan.
Reinforcement learning is an early use case
London-based Ineffable Intelligence was the first customer to access A5X, according to NVIDIA. The company, founded by AlphaGo creator David Silver, plans to use the platform for large-scale reinforcement learning and research into systems that learn continuously from experience.
“The true promise of agentic AI isn’t just speed, it’s freeing up human creativity from the friction of modern production pipelines,” said Stephan Pretorius, chief technology officer at WPP. Google’s announcement describes WPP teams using A5X to direct agents through creative production work, including frame-accurate quality checks and adaptation of brand assets. NVIDIA says WPP demonstrated a workflow spanning video generation, upscaling and restyling, with human approval before the agents execute tasks.
November availability will test the scale claims
The first practical test is whether customers can obtain the instances on the announced schedule and in the regions they need. Google says the November launch will be accompanied by expansion into new EMEA regions, but its announcement does not name those locations. That leaves customers with regional requirements waiting for details beyond the launch month.
Google and NVIDIA are pitching A5X as more than a pool of accelerators: it combines Rubin compute with high-speed networking and liquid cooling for training, inference and reinforcement learning. For now, the clearest customer evidence is Ineffable’s early access and WPP’s demonstrated creative workflow. November’s availability will show how quickly those examples can move from early access and demonstrations to workloads cloud customers can provision.