Google Cloud to offer Nvidia Vera Rubin instances next month
A5X scales toward 960,000 GPUs, with Europe expansion planned
The facts
- General availability of A5X instances arrives next month, with planned expansion to Europe
- A5X combines Nvidia Vera Rubin NVL72, ConnectX-9 SuperNICs, Google Cloud's Virgo Network and fourth-generation liquid cooling
- Google Cloud plans to scale its Vera Rubin infrastructure to as many as 960,000 Rubin GPUs across multiple sites
- Ineffable Intelligence, founded by AlphaGo creator David Silver, is the first customer, using A5X for large-scale reinforcement learning
- WPP's creative agent reference architecture ran on a single A5X instance workloads that typically required a multi-node cluster
- Google Cloud will offer Nvidia Confidential Computing at rack scale on A4X Max instances powered by GB300 NVL72, and is making it generally available on G4 VMs with RTX PRO 6000 Blackwell Server Edition GPUs
- Google Cloud and Nvidia are exploring Nvidia DSX MaxLPS capabilities for Vera Rubin infrastructure to optimise performance per watt
Google Cloud will bring Nvidia's Vera Rubin NVL72 platform to general availability next month. The new A5X instances arrive first, with a planned expansion to Europe.
The instances pair Vera Rubin with Nvidia ConnectX-9 SuperNICs. They also draw on Google Cloud's Virgo Network and its fourth-generation liquid-cooling system. Google aims the combination at agentic inference and reinforcement learning at scale. The company says it plans to scale its Vera Rubin infrastructure to as many as 960,000 Rubin GPUs across multiple sites. It is folding the platform into its existing AI Hypercomputer, an integrated compute, networking and storage system.
The first named customer is Ineffable Intelligence, a London startup founded by AlphaGo creator David Silver. It is using A5X to build what it calls "superlearner" systems that improve continuously from their own experience. Silver said the work demands both heavy compute and tight coupling between generating experience and training on it. He credited Vera Rubin's bandwidth, between its Vera CPU and Rubin GPUs and among GPUs within the rack, with letting the company push its reinforcement learning research further. WPP has also tested its creative agent architecture on A5X, using Nvidia Nemotron models and the OpenShell safety framework. It reports that workloads normally requiring a multi-node cluster ran on a single instance.
Alongside A5X, Google Cloud is rolling out Nvidia Confidential Computing at rack scale. The feature runs on bare-metal A4X Max instances powered by GB300 NVL72 chips, making Google one of the first cloud providers to offer it. The company is also making Confidential Computing generally available on G4 VMs built on RTX PRO 6000 Blackwell Server Edition GPUs. Customers including Vertebrae, Polaris Cloud and Confidential Core AI are using it for private conversational agents and policy-governed enterprise workloads. Separately, Google Cloud and Nvidia are exploring Nvidia's DSX MaxLPS technology. The goal is better performance per watt as Vera Rubin deployments grow within fixed data-centre power budgets.
Why it matters
A 960,000-GPU target makes Google Cloud one of the largest declared Vera Rubin deployments. The single-instance result WPP reports suggests real savings on cluster overhead for agentic workloads. Anyone evaluating reinforcement-learning infrastructure now has a concrete bandwidth-driven reference case in Ineffable Intelligence, not just a vendor benchmark.
Primary source
NVIDIA — blogs.nvidia.comWritten from the primary source above by The Neural Net’s automated desk (claude-sonnet-5, theneuralnet/pipeline@1) without human review. How this works. Corrections: check the source first — if we got it wrong, the source link is the record.