Google Cloud Launches NVIDIA Vera Rubin A5X Instances for Large‑Scale Agentic AI

Google Cloud has announced the general availability of its A5X instances, accelerated by NVIDIA Vera Rubin NVL72 systems and ConnectX‑9 SuperNICs, backed by the company’s fourth‑generation liquid‑cooling technology. The new instances are designed to handle demanding agentic inference and large‑scale reinforcement‑learning workloads in a single, tightly integrated rack.

The platform combines compute, high‑bandwidth networking, and storage within Google Cloud’s AI Hypercomputer, enabling a continuous loop of experience generation, evaluation, and model updates. Google plans to scale the Vera Rubin infrastructure to as many as 960,000 GPUs across multiple sites, with European availability slated for the coming month.

London‑based Ineffable Intelligence, founded by AlphaGo creator David Silver, is the first customer to adopt A5X. The firm is using the system to advance “superlearner” agents that improve continuously from their own experience, leveraging the tight CPU‑GPU and GPU‑GPU connectivity of Vera Rubin to push reinforcement‑learning research further.

Why it matters for GPU / AI infrastructure

WPP has also demonstrated a creative‑agent reference architecture on a single A5X instance, turning conversational prompts into multi‑step video generation pipelines while protecting brand assets through NVIDIA OpenShell sandboxes. In addition, Google Cloud will be among the first providers to offer NVIDIA Confidential Computing at rack scale on bare‑metal A4X Max instances and on G4 VMs powered by RTX PRO 6000 Blackwell Server Edition GPUs, giving enterprises hardware‑level data protection for proprietary models.