Google Cloud and NVIDIA Announce Upcoming General Availability of Vera Rubin A5X

Google Cloud and NVIDIA announced that general availability of A5X instances is coming next month. The rollout targets cloud customers requiring integrated platforms for training, inference and large-scale reinforcement learning. This development impacts organizations deploying agentic AI systems that demand high-bandwidth connectivity and substantial compute resources for experience-driven learning.

What hardware powers the new A5X instances?

The A5X instances integrate NVIDIA Vera Rubin NVL72 systems with NVIDIA ConnectX-9 SuperNICs. They also utilize Google Cloud’s Virgo Network and fourth-generation advanced liquid-cooling technology. This combination is designed to support demanding agentic inference and reinforcement learning workloads at scale by unifying compute, networking and storage within Google Cloud’s AI Hypercomputer.

Which organizations are using the platform?

Ineffable Intelligence, a London-based company founded by David Silver, was the first customer to access A5X. The firm uses the platform to develop "superlearner" systems that improve continuously through experience. WPP also demonstrated a creative agent reference architecture on the instances, using NVIDIA Nemotron models and OpenShell to manage multistep creative workflows with human approval.

How does this relate to previous Google AI tools?

As TechSec Feed reported on 2026-10-18, Google launched a public AI detector that scans for watermarks from Gemini, OpenAI, NVIDIA and Kakao models. This new infrastructure complements that verification capability by providing the underlying compute for the agentic systems those detectors may eventually need to analyze. The unified Gemini agent announced in October 2026 also relies on similar API-driven interfaces.

What security features accompany the rollout?

Google Cloud is making NVIDIA Confidential Computing generally available on G4 VMs powered by RTX PRO 6000 Blackwell Server Edition GPUs. This offers a platform for confidential inference and fine-tuning. Customers like Vertebrae and Polaris Cloud use these capabilities for conversational privacy and verifiable agent execution, addressing concerns about protecting proprietary models on shared infrastructure.

Questions to ask your suppliers

Security teams should ask vendors how agent execution is isolated from other tenants. Check whether the supplier provides verifiable proofs of model integrity during inference. Operations leaders need to know if data encryption keys remain under customer control when workloads move between nodes. Ensure that audit logs capture every step of an agent's decision-making process. These questions help verify that the supplier’s architecture meets your organization’s governance requirements for autonomous systems.

What to do and how to stay safe: NVIDIA

• Review current contracts to ensure they cover liability for autonomous agent actions.
• Verify that suppliers provide transparent logging for all agent-initiated network requests.
• Confirm that confidential computing features are enabled for sensitive model inference workloads.
• Establish clear human-in-the-loop approval workflows for high-risk agent decisions.

Step-by-step guide: Browser Security Settings for Small Business Teams

General security guidance from the TechSec Feed newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

When will A5X instances be generally available?

General availability is coming next month, according to the NVIDIA blog.

Will the infrastructure expand outside the US?

Google Cloud plans to expand availability to Europe, as stated in the announcement.

How many GPUs are planned for the Vera Rubin infrastructure?

Google Cloud plans to scale the infrastructure to as many as 960,000 Rubin GPUs.