Google Cloud’s A5X instances, built on NVIDIA’s Vera Rubin NVL72 platform, reach general availability next month, NVIDIA announced on October 21 at GTC Berlin, with a planned expansion to Europe. Google Cloud plans to scale its Vera Rubin-based infrastructure to as many as 960,000 Rubin GPUs across multiple sites, and the first customer is already live: Ineffable Intelligence, the London company founded by AlphaGo creator David Silver.
A5X brings Vera Rubin into Google Cloud’s AI Hypercomputer as an integrated system, combining the NVL72 racks with NVIDIA ConnectX-9 SuperNICs, Google’s Virgo Network and Google’s fourth-generation advanced liquid cooling, aimed at agentic inference and large-scale reinforcement learning. NVIDIA described the instances as an integrated platform for training, inference and reinforcement learning rather than a bare GPU rental, with the networking and cooling engineered as part of the product.
Built For Reinforcement Learning At Scale
Ineffable Intelligence is using A5X to advance large-scale reinforcement learning and build what Silver calls “superlearners” — systems that learn continuously from experience. That workload needs a tight cycle of generating experience, evaluating outcomes and updating models, which is why the connection between the Vera CPU and the Rubin GPUs, and among the GPUs inside the rack, matters as much as raw compute.
“We’re building superlearners that continually improve through their own experience,” Silver said. “Vera Rubin’s combination of compute performance and high-bandwidth connectivity allow us to push the scale and ambition of our reinforcement learning research further.” WPP, the London marketing group, has also run its creative agent reference architecture on A5X instances using Nemotron models and OpenShell — workloads that typically needed a multi-node cluster ran on a single A5X instance in its initial tests, with agents turning conversational requests into video generation and restyling workflows under human approval, inside the governed sandboxes of the OpenShell reference architecture.
Confidential Computing Comes With The Racks
Google Cloud will be among the first cloud providers to offer NVIDIA Confidential Computing at rack scale on bare-metal A4X Max instances powered by GB300 NVL72, and it is making Confidential Computing generally available on G4 virtual machines powered by RTX PRO 6000 Blackwell Server Edition GPUs. Customers including Vertebrae, Polaris Cloud and Confidential Core AI are using Confidential G4 for conversational privacy, verifiable agent execution and policy-governed enterprise intelligence.
The two companies are also exploring NVIDIA DSX MaxLPS capabilities for Vera Rubin infrastructure to optimise performance per watt within data centre power budgets — the same DSX programme behind the containerised DSX POD factories announced at the same conference, and fed by the supply chain capacity expansion NVIDIA signed with six partners.
Why It Matters
The A5X launch matters because Vera Rubin is NVIDIA’s next platform generation, and its first large cloud home sets the terms for everyone training on it: rack-scale systems sold as an integrated AI Hypercomputer, with confidential computing attached for the regulated customers who cannot run frontier workloads on ordinary instances. Google’s plan to scale toward 960,000 Rubin GPUs is one of the largest single-platform commitments yet disclosed, and landing Ineffable Intelligence as the reference customer ties the platform to reinforcement learning — the training method behind the current generation of reasoning and agent models — rather than to conventional pretraining alone. For rivals, the November general-availability date and the planned European expansion set the clock.
Background
Vera Rubin is NVIDIA’s successor platform to Blackwell, pairing Vera CPUs with Rubin GPUs in NVL72 rack systems; Google Cloud has been one of the earliest and largest buyers of each NVIDIA generation for its AI Hypercomputer line, including the A4X series built on GB300. GTC Berlin, NVIDIA’s European developer conference, has been the launch stage for the company’s autumn infrastructure announcements, and the A5X news arrived alongside a full slate of them on October 21.
Source: NVIDIA official blog; Unite.AI
0 Comments