Annanim News

Google Cloud and NVIDIA: A5X instances with Vera Rubin NVL72 reach general availability

At GTC Berlin, Google Cloud and NVIDIA announced that A5X instances, built on Vera Rubin NVL72 systems, will be generally available next month, with planned expansion to Europe. Google plans to scale to as many as 960,000 Rubin GPUs across multiple sites.

Google Cloud and NVIDIA: A5X instances with Vera Rubin NVL72 reach general availability
Image: NVIDIA Blog (opens the source site in a new tab)

A5X instances combine NVIDIA's Vera Rubin NVL72 system, ConnectX-9 SuperNICs, Google Cloud's Virgo network and fourth-generation liquid cooling. They target agentic inference and large-scale reinforcement learning workloads, and join Google's AI Hypercomputer as one integrated system of compute, networking and storage.

The first customer with access is London-based Ineffable Intelligence, founded by AlphaGo creator David Silver, which uses the platform for large-scale reinforcement learning of systems that learn continuously from experience. Advertising group WPP demonstrated a creative-agent architecture on A5X using Nemotron models and OpenShell; according to NVIDIA, workloads that previously needed a multi-node cluster ran on a single instance.

Separately, Google Cloud will be among the first cloud providers to offer NVIDIA Confidential Computing at rack scale on bare-metal instances such as A4X Max, based on GB300 NVL72, to protect proprietary models and sensitive data even when they run on infrastructure the organization does not own.

Sources

Topics