Google Cloud and NVIDIA have announced the upcoming general availability of A5X instances, which are built on the NVIDIA Vera Rubin infrastructure. These new cloud resources are designed to handle demanding artificial intelligence workloads, specifically targeting agentic inference and large-scale reinforcement learning. The instances will be accessible to customers starting next month, with further expansion planned for European data centers.
Google Cloud A5X Infrastructure Details
The A5X instances combine several high-performance components to support complex AI tasks. The hardware includes NVIDIA Vera Rubin NVL72 systems, NVIDIA ConnectX-9 SuperNICs, and Google Cloud’s Virgo Network. To manage the heat generated by these powerful systems, Google Cloud is utilizing its fourth-generation advanced liquid-cooling technology. This integrated approach allows the platform to support the entire cycle of experience-driven learning, from generating data to evaluating outcomes and updating models.
Google Cloud has outlined ambitious plans to scale this infrastructure significantly. The company intends to deploy up to 960,000 Rubin GPUs across multiple sites. This massive expansion aims to provide the compute power and high-bandwidth connectivity required for continuous model improvement.
Early Adopters and Use Cases
Ineffable Intelligence, a London-based technology firm founded by AlphaGo creator David Silver, was the first customer to access the A5X platform. The company is using the infrastructure to develop “superlearner” systems that improve continuously through experience. Silver noted that building these systems requires substantial compute power and a tight connection between generating experience and training on it. The Vera Rubin architecture allows Ineffable to push the scale of its reinforcement learning research further.
WPP, a London-based marketing and communications company, has also demonstrated its creative agent reference architecture on the new instances. WPP uses NVIDIA Nemotron models and NVIDIA OpenShell to turn conversational requests into multistep creative workflows. These workflows include video generation, upscaling, and restyling, with human approval required before execution. In initial tests, workloads that typically required a multi-node cluster ran successfully on a single A5X instance.
Confidential Computing and Energy Efficiency
As organizations deploy agentic systems, protecting proprietary models and sensitive data is a priority. Google Cloud will offer NVIDIA Confidential Computing at rack scale on bare metal instances, including A4X Max instances powered by NVIDIA GB300 NVL72. Confidential Computing is also becoming generally available on G4 virtual machines powered by NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs. This provides a flexible platform for confidential inference and fine-tuning. Customers such as Vertebrae, Polaris Cloud, and Confidential Core AI are already using Confidential G4 for conversational privacy and policy-governed enterprise intelligence.
To support energy-efficient deployment at scale, Google Cloud and NVIDIA are exploring NVIDIA DSX MaxLPS capabilities for Vera Rubin-based infrastructure. This technology helps optimize performance per watt, allowing companies to expand capacity within existing data-center power budgets.
Source: NVIDIA
