Google Cloud A5X Vera Rubin Launch Targets 960,000 NVIDIA GPUs
Google Cloud’s A5X Vera Rubin instances launch next month, with plans for 960,000 GPUs, Europe expansion and rack-scale confidential computing for agentic AI.
Summary
Google Cloud and NVIDIA announced on October 21, 2026, that A5X instances will become generally available next month, with European expansion planned. Built into Google Cloud’s AI Hypercomputer, A5X combines NVIDIA Vera Rubin NVL72, ConnectX-9 SuperNICs, Google Cloud Virgo Network and fourth-generation advanced liquid cooling for training, inference, agentic inference and large-scale reinforcement learning. Google Cloud plans up to 960,000 Rubin GPUs across multiple sites.
Ineffable Intelligence, the London company founded by AlphaGo creator David Silver, received first access and is using A5X to develop continuously learning superlearner systems. WPP demonstrated creative agents using NVIDIA Nemotron and OpenShell, converting conversational requests into human-approved video generation, upscaling and restyling workflows inside governed sandboxes. Its initial tests ran workloads normally requiring a multi-node cluster on one A5X instance.
Google Cloud plans to become one of the first providers of rack-scale NVIDIA Confidential Computing through bare metal A4X Max instances powered by GB300 NVL72. Confidential Computing is now generally available on G4 VMs with RTX PRO 6000 Blackwell Server Edition GPUs. Vertebrae, Polaris Cloud and Confidential Core AI use Confidential G4 for conversational privacy, verifiable agent execution and policy-governed enterprise intelligence. Google Cloud and NVIDIA are also exploring DSX MaxLPS to improve performance per watt and expand capacity within data-center power limits.
Positives
- Google Cloud plans to scale Vera Rubin infrastructure to as many as 960,000 GPUs across multiple sites.
- WPP ran workloads normally requiring a multi-node cluster on a single A5X instance during initial tests.
- A5X integrates Vera Rubin NVL72, ConnectX-9, Virgo networking and fourth-generation liquid cooling within Google Cloud’s AI Hypercomputer.
- Confidential Computing is generally available on G4 VMs powered by RTX PRO 6000 Blackwell Server Edition GPUs.
- Ineffable Intelligence received first access to A5X for continuously learning reinforcement learning systems.
Risks & concerns
- European A5X availability remains planned, with no launch date specified.
- The 960,000 Rubin GPU footprint is a future scaling target rather than currently deployed capacity.
- Rack-scale Confidential Computing on A4X Max remains planned, while only G4 availability is confirmed.
- DSX MaxLPS support is still being explored as data-center power budgets constrain capacity.
- WPP’s agents still require human approval before executing multistep creative workflows.