Google Cloud and Nvidia said A5X instances built on Vera Rubin NVL72 will reach general availability next month, with expansion to Europe planned. A5X combines Vera Rubin NVL72, ConnectX-9 SuperNICs, Google Cloud's Virgo network and fourth-generation liquid cooling, aimed at agentic inference and large-scale reinforcement learning. Google Cloud says it plans to scale Vera Rubin-based capacity to as many as 960,000 Rubin GPUs across multiple sites. Ineffable Intelligence, the London company founded by AlphaGo creator David Silver, is the first customer and is using the platform for reinforcement learning on systems it calls superlearners. Nvidia also cites WPP running a creative agent reference architecture with Nemotron models and OpenShell on A5X, where workloads that normally required a multi-node cluster ran on a single instance in initial tests, and notes rack-scale confidential computing on bare-metal A4X Max instances plus general availability of confidential computing on G4 VMs with RTX PRO 6000 GPUs.
Our takeFor reinforcement learning the binding constraint is usually the handoff between generating experience and training on it, not raw FLOPs, so rack-scale integration and fast fabric are the parts that matter here. The single-instance result is workload-specific and should be read as a reference-architecture finding rather than a general speedup. The more consequential addition for regulated buyers is confidential computing at rack scale, since it moves the trust boundary from the operator to the hardware.