Microsoft & NVIDIA: Scaling AI Infrastructure with Foundry & New Accelerators | GTC 2024

Microsoft is significantly expanding its artificial intelligence capabilities through a deepened collaboration with NVIDIA, unveiled at this week’s NVIDIA GTC conference in San Jose, California. The company is positioning Azure as a comprehensive platform for building, deploying, and operating AI agents and “Physical AI” systems, moving beyond experimentation to full-scale production.

At the core of Microsoft’s strategy is Microsoft Foundry, now described as the “operating system for AI” at the enterprise level. Foundry integrates models, tools, data, and observability features within Azure, specifically designed for production-ready AI agents. The next generation of Foundry Agent Service and its associated Observability features within Foundry Control Plane are now generally available, enabling organizations to build and operate AI agents at scale. The Agent Service allows for the rapid development of agents capable of reasoning, planning, and executing tasks across various tools, data sources, and workflows. Foundry Control Plane provides developers with finish-to-end visibility into agent behavior, aiming to improve both productivity, and trust.

Corvus Energy is among the early adopters, utilizing Foundry to automate inspection workflows across its global fleet, replacing manual processes with agent-driven operational intelligence, according to Microsoft.

Microsoft is also integrating NVIDIA Nemotron models into Foundry, expanding the selection of models available on Azure to include the latest reasoning, frontier, and open models. This builds on a recent partnership bringing Fireworks AI to Foundry, allowing customers to fine-tune open-weight models like NVIDIA Nemotron for low-latency performance at the edge.

To support the increasing demands of AI workloads, Microsoft is upgrading its data center infrastructure with NVIDIA’s latest technologies. The company is the first hyperscale cloud provider to power on NVIDIA’s Vera Rubin NVL72 systems in its labs, with plans to roll them out into liquid-cooled Azure datacenters in the coming months. Hundreds of thousands of liquid-cooled Grace Blackwell GPUs have already been deployed across Microsoft’s global datacenter footprint.

The infrastructure enhancements extend to sovereign and regulated environments, with initial support for the NVIDIA Vera Rubin platform now available on Azure Local, providing accelerated AI capabilities within customer-controlled environments. This allows organizations to prepare for next-generation AI workloads while maintaining Azure-consistent operations, governance, and security through Azure Arc and Foundry Local.

Beyond digital applications, Microsoft and NVIDIA are collaborating on Physical AI, integrating NVIDIA’s Physical AI Data Factory Blueprint with Microsoft Foundry as the platform for hosting and operating Physical AI systems on Azure. A public Azure Physical AI Toolchain GitHub repository has been launched, integrated with the Nvidia Physical AI Data Factory and core Azure services, to support developers building and operating physical AI and robotics workflows.

The companies are also deepening integration between Microsoft Fabric and NVIDIA Omniverse libraries, connecting live operational data with physically accurate digital twins and simulations. This aims to move beyond traditional dashboards and alerts, enabling coordinated, AI-driven action across machines, facilities, and workflows, particularly in manufacturing and operations.

Yina Arenas, corporate vice president for Microsoft Foundry, emphasized the company’s commitment to delivering a reliable, production-scale AI platform. “Microsoft is delivering reliable, production-scale AI by bringing together its global AI infrastructure, platforms and real-world systems with the latest innovation from NVIDIA,” she stated in a blog post.