Google Cloud’s first NVIDIA Vera Rubin A5X instances go live – Unite.AI

NVIDIA and Google Cloud announced on October 21, 2026 at GTC Berlin that the first live A5X instances based on NVIDIA Vera Rubin NVL72 are running at an initial customer, with general availability expected in November 2026 and extended availability in Europe as well.

A5X instances and scaling plans

A5X instances combine NVIDIA Vera Rubin NVL72 rack-scale systems and NVIDIA ConnectX-9 SuperNICs with Google Cloud Virgo Network and Google Cloud’s advanced 4th generation liquid cooling technology and are designed for large-scale agent inference and reinforcement learning workloads. Within Google Cloud’s AI hypercomputing, A5X integrates compute, networking and storage as a single system intended to support experiential learning at scale. Google Cloud said it plans to scale this Vera Rubin-based infrastructure to 960,000 Rubin GPUs across multiple sites.

Google first announced A5X at Google Cloud Next in April 2026. NVIDIA’s post from that event stated that the instances would reduce inference cost per token by up to 10x and increase token throughput per megawatt by up to 10x compared to the previous generation, and outlined scaling up to 80,000 Rubin GPUs in a single-site cluster and up to 960,000 in a multi-site deployment.

The A5X is built on the Rubin platform launched by NVIDIA on January 5, 2026. The Vera Rubin NVL72 packs 72 Rubin GPUs, 36 Vera CPUs, NVLink 6 interconnects, ConnectX-9 SuperNICs, and BlueField-4 DPUs into a single rack-scale system. NVIDIA said at launch that Rubin was in full production, with Rubin-based products available from partners in the second half of 2026, and named Google Cloud among the first cloud providers set to deploy Vera Rubin-based instances in 2026.

First customers on A5X

The first customer to access A5X was Ineffable Intelligence, a London-based technology company founded by AlphaGo creator David Silver. The company uses the platform for large-scale reinforcement learning and to develop “superlearner” systems that continuously learn from experience. The announcement describes learning through interaction as a continuous cycle of generating experiences, evaluating results, and updating models that the infrastructure must be able to support at scale.

“Vera Rubin’s combination of compute performance and high-bandwidth connectivity, between the Vera CPU and Rubin GPUs, and between Rubin GPUs within the rack, allows us to further push the scope and ambition of our reinforcement learning research,” Silver said.

WPP, a London-based marketing and communications company, also demonstrated its Creative Agent Reference Architecture on A5X instances using NVIDIA Nemotron and NVIDIA OpenShell templates, part of the NVIDIA OpenShell Reference Architecture for Secure AI Agents. According to the announcement, WPP agents convert conversational requests into multi-step creative workflows such as video generation, upscaling and restyling, with a human approval phase before execution, and OpenShell provides governed sandboxes in which agents execute, controlling access to sensitive brand assets and creative tools. In initial testing, WPP ran workloads that typically require a multi-node cluster on a single A5X instance.

Private IT and energy efficiency

According to the announcement, Google Cloud will be among the first cloud providers to offer rack-scale NVIDIA Confidential Computing on bare metal instances, including NVIDIA GB300 NVL72-based A4X Max instances. An NVIDIA post from the same day says Google Cloud is extending Confidential Bare Metal support to A4X Max instances, where it will pair accelerated non-virtualized computing with cryptographic protection for training data and intellectual property modeling. Confidential Bare Metal is a new deployment option introduced alongside Confidential Containers and Confidential Virtual Machines, and NVIDIA has also introduced platform security controls that cover protecting data in transit, measuring platform integrity, and hardware-level access restrictions.

“Our A4X instances, built using NVIDIA Confidential Computing, preserve user data privacy, protect model IP, and provide world-class service performance,” said Andrés Lagar-Cavilla, distinguished engineer and head of AI infrastructure security at Google Cloud.

Google Cloud is also making Confidential Computing generally available on G4 VMs powered by NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs for confidential inference and refinement. NVIDIA reported that customers including Vertebrae, Polaris Cloud and Confidential Core AI use Confidential G4 for conversational privacy, auditable agent execution and policy-driven business intelligence. Confidential G4 VMs with NVIDIA RTX PRO 6000 Blackwell GPUs were in preview at the time of the company’s announcement in April 2026.

Companies are also exploring NVIDIA DSX MaxLPS capabilities for Vera Rubin-based infrastructure, with the goal of optimizing performance per watt and expanding capacity within available data center power budgets, NVIDIA said.