Google Cloud가 NVIDIA Vera Rubin A5X 인스턴스를 처음으로 정식 출시했습니다.

NVIDIA and Google Cloud announced on October 21, 2026, at GTC Berlin that the first live A5X instances powered by NVIDIA Vera Rubin NVL72 are running with an initial customer. General availability is planned for November 2026, with expanded availability in Europe also planned.

**A5X Instances and Scale Plans**

The A5X instances pair NVIDIA Vera Rubin NVL72 rack-scale systems and NVIDIA ConnectX-9 SuperNICs with the Google Cloud Virgo Network and Google Cloud's fourth-generation advanced liquid-cooling technology. They are designed for large-scale agentic inference and reinforcement learning workloads. Within Google Cloud's AI Hypercomputer, A5X integrates compute, networking, and storage as a single system intended to support experience-driven learning at scale. Google Cloud said it plans to scale this Vera Rubin-based infrastructure to as many as 960,000 Rubin GPUs across multiple sites.

Google first announced A5X at Google Cloud Next in April 2026. NVIDIA's post from that event said the instances would cut inference cost per token by up to 10x and raise token throughput per megawatt by up to 10x relative to the prior generation. It outlined scaling of up to 80,000 Rubin GPUs in a single-site cluster and up to 960,000 in a multisite deployment.

A5X is built on the Rubin platform that NVIDIA launched 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 one rack-scale system. At launch, NVIDIA said Rubin was in full production, with Rubin-based products available from partners in the second half of 2026. It 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 is Ineffable Intelligence, a London-based technology company founded by AlphaGo creator David Silver. The company is using the platform for large-scale reinforcement learning and to develop "superlearner" systems that learn continuously from experience. The announcement describes learning through interaction as a continuous cycle of experience generation, outcome evaluation, and model updating, which 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 the Rubin GPUs within the rack, allow us to push the scale and ambition of our reinforcement learning research further," Silver said.

WPP, a London-based marketing and communications company, has also demonstrated its creative agent reference architecture on A5X instances using NVIDIA Nemotron models and NVIDIA OpenShell, part of the NVIDIA OpenShell Reference Architecture for safe AI agents. According to the announcement, WPP's agents convert conversational requests into multistep creative workflows, such as video generation, upscaling, and restyling, with a human approval step before execution. OpenShell supplies governed sandboxes in which the agents execute, controlling access to sensitive brand assets and creative tools. In initial tests, WPP ran workloads that typically call for a multi-node cluster on a single A5X instance.

**Confidential Computing and Power Efficiency**

Google Cloud is set to be among the first cloud providers offering NVIDIA Confidential Computing at rack scale on bare-metal instances, including A4X Max instances powered by NVIDIA GB300 NVL72, according to the announcement. A same-day NVIDIA post said Google Cloud is extending Confidential Bare Metal support to A4X Max instances, pairing non-virtualized accelerated computing with cryptographic protection for training data and model intellectual property. Confidential Bare Metal is a new deployment option introduced alongside Confidential Containers and Confidential Virtual Machines. NVIDIA also introduced platform security controls covering data protection in transit, platform-integrity measurement, and hardware-level access restrictions.

"Our A4X instances, built using NVIDIA Confidential Computing, preserve the privacy of user data, protect model IP, provide top-line serving performance," said Andrés Lagar-Cavilla, distinguished engineer and AI infrastructure security lead 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 fine-tuning. NVIDIA reported that customers including Vertebrae, Polaris Cloud, and Confidential Core AI are using Confidential G4 for conversational privacy, verifiable agent execution, and policy-governed enterprise intelligence. Confidential G4 VMs with NVIDIA RTX PRO 6000 Blackwell GPUs were in preview at the time of the companies' April 2026 announcement.

The companies are also exploring NVIDIA DSX MaxLPS capabilities for the Vera Rubin-based infrastructure, aimed at optimizing performance per watt and expanding capacity within available data-center power budgets, NVIDIA said.