Google Cloud Supercharges Vertex AI with A4X VMs and NVIDIA GB200 NVL72: GTC 2026 AI Breakthroughs

Google Cloud just unveiled game-changing Vertex AI upgrades at NVIDIA GTC 2026, including A4X VM support for NVIDIA GB200 NVL72 racks and resilient multi-week training. Discover how these advancements, plus Gemini 3.1 Flash-Lite and Nemotron 3 Super 120B, position Vertex AI as the leader in agentic AI.

Google Cloud Supercharges Vertex AI with A4X VMs and NVIDIA GB200 NVL72: GTC 2026 AI Breakthroughs

Published: March 21, 2026

Just days after the buzz of NVIDIA GTC 2026, Google Cloud is making waves in the AI world with transformative upgrades to Vertex AI. Announced at the event, these enhancements include A4X VM support for NVIDIA GB200 NVL72 rack-scale systems, hardware resiliency for uninterrupted multi-week training jobs, and an expanded Vertex AI Model Garden featuring NVIDIA Nemotron 3 Super 120B and Nemotron 3 Nano. Coupled with recent releases like Gemini 3.1 Flash-Lite public preview and Vector Search 2.0 GA, Vertex AI is cementing its position as the premier platform for enterprise agentic AI training.[1]

Infrastructure Upgrades: Powering Massive-Scale AI Training

At GTC 2026, Google Cloud highlighted its deepened partnership with NVIDIA, focusing on Vertex AI training clusters that can handle the demands of next-generation AI. The star of the show? Support for A4X VM domains on NVIDIA GB200 NVL72 rack-scale systems. This integration allows enterprises to leverage Vertex AI's managed infrastructure for massive-scale training without the headaches of custom setups.[1]

To tackle the challenges of long-running jobs, Google introduced hardware resiliency features with configurable, proactive fault detection scans. These capabilities identify and mitigate potential issues before they disrupt critical "hero" training runs, ensuring higher goodput and preventing costly restarts for multi-week workloads.[1]

Imgix's Head of Engineering, Alfonso Acosta,