Microsoft is using NVIDIA GTC 2026 to make a big statement: Azure is not just a place to train giant models, it’s positioning itself as the end‑to‑end platform for building, running and operating AI agents and “Physical AI” systems in the real world.
Microsoft doubles down on Foundry as the “OS for AI”
At the core of Microsoft’s GTC message is Foundry, which the company now openly calls the operating system for building, deploying and operating AI at enterprise scale. Built on Azure, Foundry pulls models, tools, data, and observability into a single environment designed specifically for production‑grade agents rather than just experimentation.
In a new blog post by Yina Arenas, corporate vice president for Microsoft Foundry, Microsoft outlines how it is expanding Foundry across Foundry Agent Service and NVIDIA Nemotron models to give enterprises a more opinionated, production‑ready stack for agentic AI. Foundry is also becoming the front door for a growing ecosystem of partners, observability tools and security integrations aimed at making agents something IT can actually govern, not just prototype.
Foundry Agent Service GA, Nemotron and voice agents
One of the biggest product milestones: the next‑generation Foundry Agent Service and its Observability features in Foundry Control Plane are now generally available. Foundry Agent Service lets teams build agents that can reason, plan and act across tools, data and workflows, while the Control Plane gives developers end‑to‑end visibility into how those agents behave in production.
Microsoft says customers like Corvus Energy are already using Foundry to replace manual inspection workflows with agent‑driven operational intelligence across their fleets. To make it easier to go from demo to deployment, the company is rolling out a Voice Live API integration with Foundry Agent Service in public preview, enabling voice‑first, multimodal, real‑time agent experiences that feel more like talking to a colleague than clicking through a bot.
The refreshed Foundry portal adds deeper integrations with Palo Alto Networks’ Prisma AIRS and Zenity, which Microsoft positions as a way to give builders richer diagnostics while security and compliance teams retain runtime control over how agents access systems and data. On the model side, NVIDIA Nemotron models are now available through Microsoft Foundry, joining Fireworks AI and a wide selection of frontier, reasoning and open‑weight models that can be fine‑tuned into low‑latency assets and pushed all the way to the edge.
Azure AI infrastructure: Vera Rubin NVL72 and Grace Blackwell at hyperscale
Microsoft is also using GTC to underline how aggressively it is scaling the hardware underneath all this agent activity. The company says it has already deployed “hundreds of thousands” of liquid‑cooled NVIDIA Grace Blackwell GPUs across its global datacenter footprint in under a year, engineered into facilities optimized for power, cooling, networking and rapid generational upgrades.
Microsoft claims it is the first hyperscale cloud to power on NVIDIA’s new Vera Rubin NVL72 systems in its labs, with broader rollout to modern, liquid‑cooled Azure datacenters over the coming months. The goal is to give customers infrastructure tuned for inference‑heavy, reasoning‑based workloads that can be deployed consistently across global and regulated environments, rather than forcing them to choose between performance and compliance.
That story extends to sovereign and disconnected environments as well. Foundry Local now supports modern infrastructure and large AI models, and Microsoft is adding initial support for the NVIDIA Vera Rubin platform on Azure Local so customers can bring accelerated AI into tightly controlled, customer‑operated settings. Through Azure Arc and Foundry Local, Microsoft is promising “Azure‑consistent” operations, governance and security, whether AI runs in a hyperscale region or a sovereign data enclave.
Pushing AI into the physical world with Physical AI Toolchain
Beyond data centers and dashboards, Microsoft and NVIDIA are sharpening their focus on what they call Physical AI – intelligent systems that sense, simulate and act in the real world. At GTC, this work centers on NVIDIA’s Physical AI Data Factory Blueprint, with Microsoft Foundry as the platform for hosting and operating Physical AI systems at Azure’s cloud scale.
To make that practical, Microsoft is introducing a public Azure Physical AI Toolchain GitHub repository that plugs directly into NVIDIA’s Physical AI Data Factory and core Azure services. The idea is to give developers a repeatable way to build, train and operate robotics and Physical AI workflows that link physical assets, high‑fidelity simulation and cloud training environments into enterprise‑grade pipelines.
Microsoft is also deepening the integration between Microsoft Fabric and NVIDIA Omniverse libraries, bringing together live operational data, physically accurate digital twins and simulation in one loop. That lets organizations see what is happening across their physical systems, understand it in real time and then use AI to decide and coordinate what to do next – moving beyond static dashboards and alerts to AI‑driven action across machines, facilities and end‑to‑end workflows.
Why this matters for customers
Microsoft’s GTC announcements are about turning AI from a set of labs projects into a continuously running, production‑scale system that spans cloud, edge and the physical world. By tightening its partnership with NVIDIA across Foundry, Azure AI infrastructure and Physical AI tooling, Microsoft is pitching customers a single, consistent stack for always‑on agents, reasoning‑heavy workloads and real‑world automation in factories, energy facilities and sovereign environments.
For enterprises already piloting Copilot or custom agents, the new Foundry capabilities and infrastructure roadmap offer a clearer path from prototype agents to governed, observable and scalable systems. And for Microsoft, this GTC moment is another chance to signal that it intends to compete at every layer of the AI stack—from GPUs and datacenters up to the agents making decisions on factory floors.
Recent Posts You Might Like
- Microsoft To Bring Gaming Copilot to Xbox Series X|S: Ultimate AI Game Help Is Coming (Sooner Than Expected) to Your Couch
- Rajesh Jha to Retire as EVP of Microsoft Experiences + Devices on July 1st: What Exciting Changes Are Coming for Windows, Surface, and Copilot Now?
- Microsoft unifies Copilot under new leadership team to drive “agentic” AI future in 2026
- Xbox Game Pass March 2026 Wave 2 adds Disco Elysium, Resident Evil 7, Final Fantasy IV, and more through early April
Discover more from Microsoft News Now
Subscribe to get the latest posts sent to your email.
