Google Cloud to open Vera Rubin-based A5X instances next month
A single instance replacing a multi-server cluster signals a real drop in infrastructure cost and complexity for RL and agentic workloads, not just a spec bump. Teams running agent training or large-scale inference should compare A5X pricing against current GPU cluster setups when the instances become generally available. The 960K-GPU scaling claim also sets a benchmark competitors like AWS and Azure will be measured against.
Google Cloud will open its Vera Rubin-based A5X instances to all customers next month. In early tests by London advertising group WPP, AI agent workloads that previously required a cluster of several servers ran on a single A5X instance. Each rack uses ConnectX-9 network cards and Google's Virgo network to link racks together, allowing scaling to 960K Rubin GPUs across multiple sites. The design targets reinforcement learning, where faster CPU-to-GPU and GPU-to-GPU links shorten each training round.
Google Cloud will open its Vera Rubin-based A5X instances to all customers next month.
In early tests by London advertising group WPP, AI agent workloads that previously needed a cluster of several servers ran on a single A5X instance.
Outside each rack, ConnectX-9 network cards and Google’s Virgo network tie racks together, letting Google Cloud scale the setup to 960K Rubin GPUs across multiple sites.
The design targets reinforcement learning, where models repeatedly generate experience, score results and retrain, so faster CPU-to-GPU and GPU-to-GPU links shorten each round.
