Google Cloud Unveils NVIDIA-Powered AI Supercomputer for Advanced AI Workloads

Google Cloud has announced a major expansion of its artificial intelligence infrastructure through a partnership with NVIDIA, introducing its new A5X computing instances powered by NVIDIA’s Vera Rubin NVL72 platform. Announced on October 21, 2026, the system is designed to handle demanding AI workloads, including model training, real-time inference and reinforcement learning. The infrastructure combines NVIDIA’s latest GPU technology with Google’s networking, storage and advanced liquid-cooling systems to improve performance when running large-scale AI applications. The announcement reflects the growing demand for computing power as technology companies develop increasingly capable AI models and autonomous agents that can perform complex tasks with limited human supervision. Google Cloud says the integrated platform is intended to reduce infrastructure bottlenecks and help businesses deploy advanced AI systems more efficiently.

At the centre of the announcement is NVIDIA’s Vera Rubin NVL72 architecture, which integrates 72 Rubin GPUs and 36 Vera CPUs into a rack-scale computing system. The A5X instances also incorporate NVIDIA ConnectX-9 networking technology and Google’s Virgo Network, designed to move large volumes of data between processors and computing systems. This high-speed connectivity is particularly important for advanced AI models, which require substantial computing resources to process information and update their parameters. Google Cloud plans to scale the infrastructure to support as many as 80,000 Rubin GPUs at a single site and up to 960,000 GPUs across multiple sites. These figures represent the company’s planned deployment capacity rather than the number of GPUs currently operating in a single supercomputer.

One of the primary applications for the new infrastructure is reinforcement learning, a technique in which AI systems improve their performance through repeated interaction, feedback and evaluation. The first customer to access the A5X platform is Ineffable Intelligence, a London-based AI company founded by David Silver, who previously led influential reinforcement-learning research at Google DeepMind. The company intends to use the computing platform to develop what it calls “superlearner” systems capable of improving through experience. Such systems require substantial computing resources because they must generate training experiences, evaluate the results and update their models repeatedly. Google and NVIDIA believe the new infrastructure can support this continuous learning process at a scale that would be difficult to achieve with less integrated computing systems.

The infrastructure is also expected to support businesses developing AI agents for practical applications. Marketing and communications company WPP has demonstrated a system that uses AI agents to manage multistep creative workflows, including video generation, image upscaling and visual restyling. The system uses NVIDIA Nemotron models alongside OpenShell, a framework designed to provide controlled execution environments for AI agents. These safeguards help restrict access to sensitive brand assets and tools, while human approval is required before certain actions are carried out. WPP reported that some workloads that would normally require a cluster of multiple computing nodes could run on a single A5X instance during its initial tests. The demonstration illustrates how advanced AI infrastructure could support automated production processes while retaining human oversight.

Security and data protection are also central to the partnership. As AI systems gain access to confidential documents, proprietary models and business applications, companies need ways to protect information while it is being processed. Google Cloud and NVIDIA are expanding confidential-computing capabilities designed to protect sensitive data and model intellectual property during AI training and inference. Google Cloud also plans to expand the availability of its Vera Rubin-based infrastructure to European regions, helping organisations address data-residency and sovereignty requirements. The A5X instances are expected to become generally available in November 2026, with additional regional deployments planned. The announcement highlights the increasing importance of combining computing performance with security, energy efficiency and regulatory compliance as AI workloads expand.

The launch strengthens Google Cloud’s position in the increasingly competitive market for AI computing infrastructure. Major cloud providers are investing heavily in specialised chips, high-speed networking and large data-centre systems to attract businesses developing advanced models and autonomous agents. By combining NVIDIA’s hardware with Google’s cloud infrastructure, the companies aim to provide a platform that can support everything from large-scale model training to continuous AI-driven operations. However, the infrastructure’s long-term impact will depend on its real-world performance, operating costs, energy requirements and availability to customers. If the deployment proceeds as planned, A5X could become an important computing platform for organisations seeking to build more capable AI systems and automate increasingly complex tasks.