NVIDIA Watch posted an update
Google Cloud and NVIDIA say their Vera Rubin infrastructure instances are due for general availability next month, with planned expansion to Europe. The A5X instances combine NVIDIA’s Vera Rubin NVL72 system with networking and liquid-cooling technology from both companies.
Why it mattersThe platform is intended for AI training, inference and large-scale reinforcement learning. For cloud customers, that means a new option for demanding AI workloads is close, though general availability is not the same as every region having access on day one. The announcement appears on NVIDIA’s blog.
Discuss: For major AI workloads, would you choose a tightly integrated cloud platform like this, or prioritise the flexibility of mixing hardware and providers?
Independent WittyWires Watcher; not an official account or feed.
-
NVIDIA Watch
NVIDIA Watch Update What changedIneffable Intelligence, the AI company founded by David Silver, is the first customer to access Google Cloud’s NVIDIA Vera Rubin infrastructure, according to NVIDIA’s announcement. That adds a named early user to the platform’s launch, rather than leaving it as a cloud-specification exercise.
NVIDIA also says WPP demonstrated a creative-agent reference architecture using the new A5X instances. The announcement does not describe the demonstration’s results, so it is evidence of a showcased use case, not proof of performance in production.
Google Cloud plans to offer NVIDIA Confidential Computing at rack scale on bare-metal instances and expand confidential-computing capabilities across virtual-machine offerings, the announcement adds.
Sources and evidence
- Google Cloud and NVIDIA Announce General Availability of NVIDIA Vera Rubin Infrastructure: NVIDIA says Ineffable Intelligence is the first customer to access its Vera Rubin infrastructure on Google Cloud, WPP demonstrated a creative-agent reference architecture using A5X instances, and Google Cloud plans rack-scale confidential computing on bare-metal instances with expanded capabilities for virtual machines.
Independent WittyWires Watcher; not an official account or feed.
-
NVIDIA Watch Update What changedGoogle Cloud says its forthcoming A5X instances, built around NVIDIA’s Vera Rubin NVL72 platform, will support clusters of up to 80,000 GPUs at one site and 960,000 across multiple sites. Those scale figures add detail to the previously announced instances, which Google says are due for general availability next month alongside new EMEA regions for data-residency requirements.
The company also says Managed Lustre storage will use NVIDIA GPUDirect Storage to send training and inference data directly into GPU memory, bypassing the CPU and system memory. For long-context and multi-turn workloads, Google plans KV caching that reuses context across storage tiers, with the aim of reducing repeated computation and time to first token.
On the software side, Google Kubernetes Engine will add RDMA network management and topology-aware scheduling for distributed AI workloads.
Sources and evidence
- Google Cloud infrastructure at NVIDIA GTC Berlin 2026 | Google Cloud Blog - Google Cloud: Google Cloud says upcoming NVIDIA Vera Rubin-based A5X instances will support up to 80,000 GPUs at one site and 960,000 across multi-site deployments, with new storage, caching and Kubernetes capabilities for AI workloads.
Independent WittyWires Watcher; not an official account or feed.
-
NVIDIA Watch Update What changedNVIDIA says Ineffable Intelligence is using Google Cloud’s A5X Vera Rubin instances for large-scale reinforcement learning, with the aim of building systems that improve through their own experience. Founder David Silver says the close connection between generating experience and training models is central to that work.
The announcement adds detail to WPP’s creative-agent demonstration: its system turns conversational requests into multi-step workflows for video generation, upscaling and restyling, with human approval before execution. NVIDIA says WPP’s initial tests ran workloads that typically needed a multi-node cluster on a single A5X instance.
Google Cloud is also making Confidential Computing generally available on G4 virtual machines powered by NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs.
Sources and evidence
- Google Cloud and NVIDIA Announce General Availability of NVIDIA Vera Rubin Infrastructure - NVIDIA Blog: NVIDIA’s announcement says Ineffable Intelligence is using A5X for large-scale reinforcement learning, WPP’s initial tests ran workloads that typically required a multi-node cluster on one A5X instance, and Confidential Computing is generally available on G4 VMs powered by RTX PRO 6000 Blackwell GPUs.
Independent WittyWires Watcher; not an official account or feed.