Microsoft & NVIDIA: Scaling Agentic & Physical AI

Out of NVIDIA GTC in San Jose, California, comes news of new collaboration with Microsoft announcing new capabilities across Microsoft Foundry, Azure AI infrastructure and physical AI to help its customers operate AI reliably at enterprise scale.

Working closely with NVIDIA, Microsoft is aiming to help organisations move from experimentation to production, building secure, safe AI agents that run across cloud, hybrid and sovereign environments.

Rather than treating AI as an experimental tool, the enhanced capabilities are designed to support organisations in running AI operations with the reliability and security demanded by production environments.

This could prove particularly valuable for businesses seeking to scale their AI initiatives beyond pilot projects.

The new tools in Foundry enable users to build, deploy and operate production-ready AI agents on NVIDIA accelerators – powerful, specialised computer chips that make AI work incredibly fast and efficiently – and open NVIDIA Nemotron models, which act as starting templates so users do not have to develop the AI's core intelligence from scratch.

Building production-ready AI agents

The Foundry Agent Service and Observability in Foundry Control Plane are now generally available, allowing users to build and operate AI agents.

Foundry Agent Service is a tool that helps teams quickly create AI agents that can reason, plan and act by connecting to and working with all the software, information and business processes a company already uses.

Once created, Foundry Control Plane provides the developer end-to-end visibility into agent behaviour with a complete view of exactly what the AI is doing.

This level of transparency could be crucial for organisations concerned about AI governance and the need to understand how autonomous systems make decisions.

Additionally, the Voice Live API integration with Foundry Agent Service, in public preview, enables developers to build voice-first, multimodal, real-time agentic experiences.

This functionality could open new possibilities for conversational AI applications across customer service, enterprise workflows and user interfaces.

Infrastructure optimised for reasoning workloads

The new Azure AI infrastructure is optimised for inference-heavy, reasoning-based workloads, including the first hyperscale cloud to power on next-generation NVIDIA Vera Rubin NVL72 systems.

Yina Arenas, Corporate Vice President of Microsoft Foundry, says: "Microsoft's AI infrastructure approach is engineered to seamlessly bring next-generation NVIDIA systems into Azure data centres that are designed for power, cooling networking and rapid generational upgrades.

"This allows our customers to move with speed and agility and stay at the leading edge from generation to generation."

According to Arenas, in less than a year, Microsoft has deployed hundreds of thousands of liquid-cooled Grace Blackwell GPUs across its global data centre footprint and now the company is the first hyperscale cloud to power on NVIDIA's newest Vera Rubin NVL72 in its labs.

"Over the next few months, Vera Rubin NVL72 will be rolled out into modern, liquid-cooled Azure data centres," Yina adds.

Industrial physical AI integration

Microsoft and NVIDIA are also working together to bring industrial physical AI to joint customers.

This work centres on NVIDIA Physical AI Data Factory Blueprint, with Foundry as the platform for hosting and operating physical AI systems on Azure.

Developers can use the integrated blueprint to build, train and operate physical AI and robotics workflows that connect physical assets, simulation and cloud training environments into repeatable pipelines.

Microsoft and NVIDIA are deepening the integration between Microsoft Fabric and NVIDIA Omniverse libraries, connecting live operational data with physically accurate digital twins and simulation.

This could allow organisations in manufacturing, logistics and other heavy operations to move beyond simple dashboards and alerts.

Instead of merely notifying a human that something is wrong, the system automatically triggers and coordinates a precise, optimal AI-driven action across all machines, facilities and processes, potentially resulting in faster decisions and minimised human error.

"Whether powering always-on agents, scaling next-generation AI infrastructure or deploying intelligent systems in factories, energy facilities and sovereign environments, Microsoft and NVIDIA are helping customers move faster from insight to action," says Arenas.

This is the next step in Microsoft's years of work with NVIDIA to integrate hardware, software and infrastructure to power AI breakthroughs.