Google Cloud and NVIDIA Announce General Availability of NVIDIA Vera Rubin Infrastructure
NVIDIA Blog Uttara Kumar ● Covered by 2 sources
Google Cloud is bringing NVIDIA Vera Rubin instances to customers next month. It’s built for training, inference and giant reinforcement-learning jobs, with Europe next in line.
Based on reporting by NVIDIA Blog, Uttara Kumar — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Google Cloud and NVIDIA are moving Vera Rubin infrastructure into general availability next month, with Europe on the roadmap after that. The new A5X instances are meant to give cloud customers one integrated setup for training, inference and large-scale reinforcement learning, rather than a pile of separate parts stitched together after the fact.
The stack is a very NVIDIA-and-Google Cloud mix: Vera Rubin NVL72, ConnectX-9 SuperNICs, Google Cloud Virgo Network and fourth-generation liquid cooling. Google Cloud says it plans to scale this Vera Rubin-based infrastructure to as many as 960,000 Rubin GPUs across multiple sites. That is an eye-catching number, but the more interesting part is what it says about where the companies think demand is going: toward agentic inference and reinforcement learning at serious scale.
Ineffable Intelligence was the first customer to get access. The London-based company, founded by AlphaGo creator David Silver, is using A5X to push what it calls “superlearner” systems — models that keep improving through experience. Silver’s pitch is blunt: if the system is going to learn from interaction, the infrastructure has to let experience generation and training stay tightly connected, fast enough that each improvement can affect the next round.
WPP is also testing the platform. Its creative agent reference architecture on Vera Rubin-based A5X uses NVIDIA Nemotron models and NVIDIA OpenShell, with human approval before agents carry out tasks like video generation, upscaling and restyling. In WPP’s early tests, work that usually needed a multi-node cluster ran on a single A5X instance. That is the kind of detail enterprise buyers notice, because it turns “AI at scale” from a slogan into something you can actually budget for.
Google Cloud is also pushing confidential computing into the mix. It says it will be among the first cloud providers to offer NVIDIA Confidential Computing at rack scale on bare metal A4X Max instances powered by NVIDIA GB300 NVL72, and it is making Confidential Computing generally available on G4 VMs with RTX PRO 6000 Blackwell Server Edition GPUs. Customers including Vertebrae, Polaris Cloud and Confidential Core AI are already using Confidential G4 for privacy, verifiable agent execution and policy-governed enterprise intelligence. The other thread here is power: Google Cloud and NVIDIA are looking at NVIDIA DSX MaxLPS capabilities to improve performance per watt and stretch capacity within data-center power limits.
My take — AI-written commentary, not fact-checked reporting
This is the real cloud AI story now: not another model demo, but who can wire compute, networking, cooling and security into one machine that actually behaves. The industry spent years pretending scale was just buying more GPUs; turns out it also involves power budgets, permissions and the slightly dull business of keeping the thing cool. Boring engineering is finally getting the headline it deserves.
Read more about this at: NVIDIA Blog