AI Research #NVIDIA #Google Cloud #infrastructure #AI

NVIDIA and Google Cloud Unveil Next-Gen AI Infrastructure

The Vera Rubin Stack aims to tackle latency, cost, and security bottlenecks for agentic and physical AI at scale.

Saturday April 25, 2026

TL;DR — Quick Answer

At Cloud Next 2026, NVIDIA and Google Cloud unveiled the Vera Rubin Stack, a full-stack AI infrastructure for agentic and physical AI. The stack delivers generally available A5X-based instances, confidential VMs on Blackwell GPUs, a 40% reduction in PPO training overhead, and RLHF iteration time cut from 6 hours to 3.5 hours for 7B models.

Key Takeaways

→ The Vera Rubin Stack targets latency, cost, and security bottlenecks for agentic AI
→ A5X-based instances are now generally available
→ Confidential VMs run on Blackwell GPUs
→ PPO training overhead is reduced by 40%
→ RLHF iteration time dropped from 6 hours to 3.5 hours for 7B models
→ The stack could unlock widespread adoption of autonomous AI systems

NVIDIA and Google Cloud Unveil Next-Gen AI Infrastructure — AI news article illustration

NVIDIA and Google Cloud have unveiled a full-stack AI infrastructure for agentic and physical AI at Cloud Next 2026.

The Vera Rubin Stack

A5X-based instances — Now generally available
Confidential VMs — On Blackwell GPUs
40% reduction in PPO training overhead
RLHF iteration time reduced from 6 hours to 3.5 hours for 7B models

What It Solves

The new stack addresses:

Latency bottlenecks
Cost challenges
Security gaps in stateful agentic workflows

Why It Matters

The integration tackles the economic infeasibility of running agentic AI at scale, potentially unlocking widespread adoption of autonomous AI systems.

Written by Massin BSN

Frequently Asked Questions

What is the Vera Rubin Stack?

A full-stack AI infrastructure from NVIDIA and Google Cloud unveiled at Cloud Next 2026.

What does it solve?

Latency bottlenecks, cost challenges, and security gaps in stateful agentic workflows.

What is generally available?

A5X-based instances are now generally available, and confidential VMs run on Blackwell GPUs.

Who is it for?

Teams building agentic and physical AI that needs to run at scale economically.

AI Research