Google Cloud A5X instances combine NVIDIA Vera Rubin and advanced networking

A5X instances integrate NVIDIA Vera Rubin NVL72, NVIDIA ConnectX-9 SuperNIC, Google Cloud Virgo Network and Google Cloud’s fourth-generation advanced water-cooling technology. Together, these technologies are designed to support demanding agent inference and reinforcement learning workloads at scale.

Google Cloud plans to scale Vera Rubin-based infrastructure to as many as 960,000 Rubin GPUs across multiple sites.

Ineffable Intelligence uses A5X for reinforcement learning

London-based technology company Ineffable Intelligence, founded by AlphaGo creator David Silver, was the first customer to gain access to A5X. The company is using the platform to power reinforcement learning at scale and develop “super learner” systems that continuously learn from experience.

Learning through interaction requires a continuous cycle of generating experiences, evaluating results and updating models. This process requires infrastructure that can support the entire workflow at scale.

A5X deploys NVIDIA Vera Rubin as an integrated system on Google Cloud’s AI hypercomputer, combining compute, networking and storage to support experience-driven learning at scale.

“We are building super learners who continually improve through their experiences,” said David Silver, Founder and CEO of Ineffable Intelligence. “This requires both significant compute and tight connectivity between the generation of the experience and its training, so each improvement in the agent can rapidly shape the next experience. The combination of Vera Rubin’s compute performance and high-bandwidth connectivity between Vera CPUs and Rubin GPUs, and between Rubin GPUs in racks, allows us to further the scale and ambition of our reinforcement learning research.”

WPP tests AI agents for multi-step creative workflows

London-based marketing and communications company WPP is using the NVIDIA Nemotron model and NVIDIA OpenShell Reference Architecture for safe AI agents.

The agent transforms conversational requests into multi-step creative workflows, including video generation, upscaling and restyling, with human approval required before execution.

OpenShell provides a managed sandbox for agent execution and helps control access to sensitive brand assets and creative tools. In WPP’s initial testing, workloads that typically require multi-node clusters were run on a single A5X instance.

Confidential computing for AI inference and fine-tuning

When organizations deploy agent systems, they need to protect the proprietary models and sensitive data that power those systems, even when they run on infrastructure they do not own.

Google Cloud will be one of the first cloud providers to offer NVIDIA Confidential Computing at rack scale on bare-metal instances, including A4X Max instances with NVIDIA GB300 NVL72.

Google Cloud is also making Confidential Computing commonly available on G4 virtual machines with NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs. The platform provides a flexible environment for confidential inference and fine-tuning.

Customers including Vertebrae, Polaris Cloud and Confidential Core AI use Confidential G4 for conversation privacy, verifiable agent execution and policy-driven enterprise intelligence.

Google Cloud and NVIDIA target energy-efficient AI infrastructure

Google Cloud and NVIDIA are in talks to support energy-efficient deployments at scale through NVIDIA DSX MaxLPS. Vera Rubin-based infrastructure features are designed to optimize performance per watt and help organizations scale capacity within their available data center power budgets.