Google Cloud and Nvidia announce Rubin-based AI infrastructure

Google Cloud and Nvidia have announced the upcoming launch of A5X instances, which are built on the Nvidia Vera Rubin NVL72 infrastructure. Scheduled for general availability in November 2026, these bare-metal systems will utilize liquid cooling and the Google Virgo network.

Technical specifications for the new architecture include ConnectX-9 SuperNICs and the NVLink-6 interface, which achieves transfer rates of 3.6 TB/s between GPUs. The infrastructure is designed for massive scalability, with targets of up to 80,000 Rubin GPUs per location and a total of 960,000 units across multiple sites. Ineffable Intelligence has been named the first customer for these instances.

Despite these technical advancements, Nvidia faces market pressure due to concerns regarding the high costs of financing AI data center expansions, rising oil prices, and high valuations. While the company continues to announce infrastructure projects with partners, investors remain cautious about the economic framework and credit costs associated with such large-scale deployments.