Google Cloud's Vera Rubin A5X Arrives Next Month

Cloud customers waiting for NVIDIA’s next rack-scale system now have a month, if not a date. Google Cloud’s A5X instances, built on NVIDIA’s Vera Rubin NVL72, will reach general availability next month, the two companies said at NVIDIA’s GTC Berlin conference. Availability in Europe will follow.

A5X itself is not new. Google announced it in April, promising to be among the first to offer Vera Rubin “when it becomes available later this year”. What has changed is the timing: Google Cloud now puts general availability a month away, and has named a first customer. Neither company has given a launch day, prices or per-instance specifications.

London companies were first to try it

According to NVIDIA’s blog, Ineffable Intelligence, a London-based company founded by AlphaGo creator David Silver, “was the first customer to access A5X”. NVIDIA says Ineffable is using it for large-scale reinforcement learning, to shape “superlearner” systems that learn continuously from experience.

“We’re building superlearners that continually improve through their own experience,” Silver said in the post. He said the work needs both heavy compute and a tight link between generating experience and training on it. “Vera Rubin’s combination of compute performance and high-bandwidth connectivity, between the Vera CPU and Rubin GPUs and between the Rubin GPUs within the rack, allow us to push the scale and ambition of our reinforcement learning research further.”

WPP, the London-based marketing group, has also shown its creative-agent setup running on A5X, using NVIDIA’s Nemotron models and its OpenShell software for running agents in controlled sandboxes. NVIDIA says that in WPP’s initial tests, workloads that typically needed a multi-node cluster ran on a single A5X instance. That is a company test result, not an independent benchmark. In Google’s post, WPP chief technology officer Stephan Pretorius said Google Cloud gives it “the reliability, scale and sovereign security required to meet our clients’ most ambitious creative demands”.

The 960,000-GPU figure is a ceiling, not a count

Each A5X instance combines a Vera Rubin NVL72 rack with NVIDIA ConnectX-9 SuperNICs, Google’s Virgo Network and Google’s fourth-generation liquid cooling, NVIDIA says. It adds that Google Cloud “plans to scale its Vera Rubin-based infrastructure to as many as 960,000 Rubin GPUs across multiple sites”.

Google’s own announcement puts that number differently. It describes it as what the Virgo network fabric will allow: clusters of up to 80,000 Rubin GPUs at one site and up to 960,000 across several. Google gave the same two figures in April. Neither company has said how many Rubin GPUs Google has installed so far.

Google is also adding support for its Managed Lustre file system with NVIDIA GPUDirect Storage on A5X, which it says gives data a direct path into GPU memory. It is promising KV caching that extends context memory beyond the GPUs’ own high-bandwidth memory for long agentic tasks. A5X will also run on Google Kubernetes Engine with Kueue, a scheduler Google says boosts distributed workloads through topology-aware scheduling.

Europe gets regions, but none named yet

On Europe, Google says the launch comes with an expansion “to support EMEA data residency and sovereignty requirements with availability in new local regions”. It did not say which regions or when. NVIDIA describes the same plan as an “expansion of availability to Europe”. Neither says whether the UK is included.

The companies made two smaller security announcements alongside A5X. Google Cloud will be among the first cloud providers to offer NVIDIA Confidential Computing at rack scale on bare-metal instances such as A4X Max, which runs on GB300 NVL72, NVIDIA says. Confidential Computing is now generally available on G4 virtual machines with RTX PRO 6000 Blackwell GPUs. The two companies are also exploring NVIDIA’s DSX MaxLPS power software for Vera Rubin systems.

The next marker is Google’s general-availability notice in November. When it comes, it should show the first prices, and whether any European region launches on day one.