Loading page…
Ineffable Intelligence already has access to A5X. Google Cloud’s larger infrastructure buildout remains a plan, alongside a future European rollout.
Listen to this story
Google Cloud and NVIDIA are positioning A5X as infrastructure for reinforcement-learning cycles as well as conventional training and inference, with general availability expected in November 2026 and expansion into Europe planned. The platform integrates Vera Rubin systems into Google Cloud’s AI Hypercomputer, and Google says its Rubin footprint could reach 960,000 GPUs across multiple sites. Early uses span continuous-learning research and governed creative-agent workflows, but WPP’s reported consolidation is not evidence of faster or cheaper execution.
London-based Ineffable Intelligence, founded by David Silver, was the first A5X customer and is using it to develop systems that learn continuously from experience.
WPP’s creative agents use NVIDIA Nemotron and OpenShell to carry out multistep tasks, with human approval required before execution.
WPP says an initial workload ran on one A5X instance instead of a typical multi-node cluster; it reported no completion-time or cost comparison.
Google Cloud’s next AI computing offering is moving toward wider access. NVIDIA says A5X instances powered by its Vera Rubin systems are expected to become generally available in November 2026, with European expansion planned. Google Cloud also plans to scale its Rubin infrastructure to as many as 960,000 GPUs across multiple sites.
The workload behind the announcement is reinforcement learning: a repeated cycle of generating experience, evaluating outcomes and updating a model. NVIDIA positions A5X as infrastructure for that entire cycle, alongside training and inference—the work of producing model responses. The design brings Vera Rubin into Google Cloud’s AI Hypercomputer, combining computing, networking and storage.
London-based Ineffable Intelligence, founded by AlphaGo creator David Silver, was the first customer to access A5X. It is using the platform for large-scale reinforcement learning and developing what it calls “superlearner” systems that learn continuously from experience.
That demands both substantial compute and a tight connection between generating experience and training on it, so each improvement in the agent can quickly shape its next experience.
David Silver, founder and CEO of Ineffable Intelligence
WPP, the London-based marketing and communications company, demonstrated a different use of the same infrastructure. Its creative-agent reference architecture runs on A5X using NVIDIA Nemotron models and OpenShell. The agents turn conversational requests into multistep creative work, including video generation, upscaling and restyling.
Those workflows require human approval before execution. OpenShell supplies governed sandboxes—controlled environments for running agents—to help regulate access to sensitive brand assets and creative tools.
NVIDIA says WPP’s initial tests ran workloads on a single A5X instance that would typically require a cluster spanning multiple computing nodes. The result describes computing consolidation, not a measured reduction in completion time or cost.
For the larger deployment, Google Cloud and NVIDIA are exploring NVIDIA DSX MaxLPS capabilities for Vera Rubin infrastructure. The aim is to improve performance per watt and fit more capacity within available data-center power budgets; that work remains exploratory.
Loading discussion...
Join the conversation
Explain which creative tasks you would let run without approval.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.