Microsoft and NVIDIA: Advancing AI Infrastructure, Foundry, and Physical AI
- Microsoft and NVIDIA have expanded their strategic partnership to integrate accelerated computing and cloud-scale engineering into a unified platform for enterprise AI.
- Central to this effort is Microsoft Foundry, which functions as an operating system for building, deploying, and operating AI at enterprise scale.
- As of March 16, 2026, the Foundry Agent Service and Observability in the Foundry Control Plane are generally available.
Microsoft and NVIDIA have expanded their strategic partnership to integrate accelerated computing and cloud-scale engineering into a unified platform for enterprise AI. The collaboration focuses on scaling the deployment of production-ready AI agents, upgrading hyperscale infrastructure for reasoning-based workloads, and advancing Physical AI systems that bridge digital simulations with real-world operations.
Expanding Microsoft Foundry for Agentic AI
Central to this effort is Microsoft Foundry, which functions as an operating system for building, deploying, and operating AI at enterprise scale. By integrating models, tools, data, and observability into a single system, Foundry is designed to move organizations from AI experimentation to production-grade agents.
As of March 16, 2026, the Foundry Agent Service and Observability in the Foundry Control Plane are generally available. The Agent Service allows development teams to create agents capable of reasoning, planning, and acting across various workflows and data sets. To ensure enterprise trust and developer productivity, the Foundry Control Plane provides end-to-end visibility into agent behavior.
The platform now includes NVIDIA Nemotron models, which are available as secure NIM microservices. These open models allow developers to build specialized agentic systems directly within Foundry. To further optimize these assets, Microsoft has integrated Fireworks AI, enabling customers to fine-tune open-weight models like Nemotron into low-latency assets for distribution to the edge.
Additional capabilities introduced include a public preview of Voice Live API integration with the Foundry Agent Service, allowing for the creation of real-time, multimodal, voice-first agentic experiences. The ecosystem also features expanded integrations for Zenity and Palo Alto Networks’ Prisma AIRS to provide runtime security across the agent lifecycle.
Early adoption is already visible in industrial applications. Corvus Energy is utilizing Foundry to replace manual inspection workflows with agent-driven operational intelligence across its global fleet.
Next-Generation AI Infrastructure
To support inference-heavy and reasoning-based workloads, Microsoft is updating its Azure AI infrastructure. The company has announced that it is the first hyperscale cloud to power on NVIDIA Vera Rubin NVL72 systems in its labs, with a rollout to liquid-cooled Azure datacenters planned for the following months.
This hardware evolution follows the deployment of hundreds of thousands of liquid-cooled Grace Blackwell GPUs across Microsoft’s global footprint. Microsoft is deploying more than 100,000 Blackwell Ultra GPUs in NVIDIA GB300 NVL72 systems globally for inference purposes.
The infrastructure strategy extends to the Microsoft Fairwater AI superfactory, which connects data centers in Wisconsin and Atlanta, Georgia. This facility integrates hundreds of thousands of NVIDIA Blackwell GPUs for large-scale training and utilizes next-generation NVIDIA Spectrum-X Ethernet switches.
For organizations requiring higher levels of control, Microsoft has introduced initial support for the NVIDIA Vera Rubin platform on Azure Local. This allows the deployment of reasoning-based and agentic systems in sovereign or customer-controlled environments while maintaining consistent operations and governance through Azure Arc and Foundry Local.
Advancing Physical AI and Digital Twins
Microsoft and NVIDIA are also targeting Physical AI, which applies AI to robotics and physical assets. This initiative centers on the NVIDIA Physical AI Data Factory Blueprint, with Microsoft Foundry serving as the hosting and operating platform on Azure.
To facilitate this, Microsoft has released a public Azure Physical AI Toolchain GitHub repository. This toolchain integrates with the NVIDIA Physical AI Data Factory and core Azure services, allowing developers to create repeatable pipelines that connect physical assets, simulations, and cloud training environments.
The collaboration further deepens the integration between Microsoft Fabric and NVIDIA Omniverse libraries. By connecting live operational data with physically accurate digital twins, organizations in manufacturing and operations can move beyond static dashboards to AI-driven actions across facilities and machines.
To support these diverse AI needs, Microsoft Foundry now offers both NVIDIA Nemotron models for digital AI and NVIDIA Cosmos models for physical AI.
Industry Application and Scale
The combined infrastructure is designed to support the most demanding workloads for the Microsoft AI Superintelligence Team, Microsoft 365 Copilot, and OpenAI. Other industry players, such as Black Forest Labs, are utilizing NVIDIA GB200 NVL72 systems to train multimodal FLUX models for visual intelligence.
NVIDIA Nemotron integrations are being used to accelerate AI capabilities for Microsoft SQL Server 2025, while new capabilities are being introduced for onboarding AI agents directly within Microsoft 365.