Google Cloud and NVIDIA have announced general availability of NVIDIA Vera Rubin infrastructure. The A5X instances are built on Vera Rubin NVL72 systems with ConnectX-9 SuperNICs, the Google Cloud Virgo Network and fourth-generation liquid cooling. General availability is next month, with planned expansion into Europe. Google Cloud says it plans to scale its Vera Rubin infrastructure to as many as 960,000 Rubin GPUs across multiple sites.
The first customer on A5X is Ineffable Intelligence, the London lab founded by David Silver, which is using it for large-scale reinforcement learning. WPP demonstrated a creative agent reference architecture on A5X using NVIDIA Nemotron models and OpenShell, and reported that workloads which normally need a multi-node cluster ran on a single A5X instance in initial tests.
That single-instance claim is the one to watch, because it changes the unit of capacity planning. If a job that previously required a cluster now fits on one instance, the relevant comparison is no longer GPUs per dollar but jobs per instance - and NVIDIA published no generation-over-generation performance figures alongside the announcement.
Sources
NVIDIA · Unite.AI