Google Cloud Launches A5X Bare Metal Instances, Supporting Multi-Site Clusters with Up to 960,000 Rubin GPUs

Google Cloud launched the A5X bare metal instance at the Google Cloud Next conference, powered by NVIDIA's Vera Rubin rack-level system. Bare metal instances are physical servers on the cloud with independent resources and no virtualization overhead.

The A5X instance uses NVIDIA's ConnectX-9 SuperNIC and Google's latest Virgo Network large-scale expandable AI data center architecture. A single-site cluster can scale up to 80,000 Rubin GPUs, while multi-site clusters support up to 960,000 GPUs, helping customers run large-scale AI workloads.

Mark Lohmeyer, VP of Google Cloud AI and Compute Infrastructure, stated that AI development in the next decade depends on integrated AI-optimized infrastructure. By combining Google Cloud's scalable infrastructure with NVIDIA's advanced platforms, customers can flexibly train and deploy models, optimizing performance and cost.