Google Cloud announced that its new A5X instances, powered by NVIDIA Vera Rubin GPUs, will be generally available next month. The rollout will start in the U.S. and is slated to expand to Europe later this year. Each A5X server combines a Vera Rubin NVL72 GPU, NVIDIA ConnectX‑9 networking cards, Google’s Virgo network, and a fourth‑generation liquid‑cooling system. Google says it will eventually run up to 960,000 Rubin GPUs across several data‑center sites.
First customers put A5X to work
London‑based Ineffable Intelligence, founded by AlphaGo creator David Silver, is the first user of the A5X. The company is using the platform for large‑scale reinforcement learning, building “superlearners” that improve through continual interaction. Ineffable says the tight link between the Vera CPU and the Rubin GPUs helps speed up the loop of generating experience, evaluating outcomes, and updating models. Marketing firm WPP also tested the A5X with NVIDIA’s Nemotron models and the OpenShell reference architecture. Their test showed that workloads that usually needed a multi‑node cluster could run on a single A5X instance, speeding up creative‑agent pipelines such as video generation and upscaling.
Security and efficiency features
Google Cloud will be one of the first providers to offer NVIDIA Confidential Computing on bare‑metal A4X Max instances and on G4 VMs that use RTX PRO 6000 Blackwell GPUs. The technology encrypts data while it is being processed, which helps protect proprietary models and sensitive data. Both companies are also looking at NVIDIA’s DSX MaxLPS tools to improve performance per watt, aiming to keep power use in check as the infrastructure scales.
Why it matters
For AI developers, the new A5X instances promise faster training loops and lower hardware complexity, which can cut project costs and time-to‑market. Enterprises that handle confidential data gain an option to run models in encrypted memory, reducing the risk of data leaks.
