Microsoft Foundry Agent Service Goes GA — Enterprise-Ready Agent Development Platform

General Availability Announced at GTC 2026

On March 16, 2026, Microsoft announced the general availability (GA) of Foundry Agent Service at GTC 2026. This unified platform for building, deploying, and monitoring AI agents has transitioned from preview to production-ready status, enabling enterprise organizations to confidently deploy AI agents at scale. The announcement also included GA status for Observability in Foundry Control Plane, providing centralized performance monitoring and governance management for agent ecosystems.

Unified ai.azure.com Portal for One-Stop Development

A key highlight of the GA release is the integration of the ai.azure.com portal. Previously, model catalogs, agent development tools, and governance settings were scattered across separate interfaces. Now, developers can access models, build agents, and configure governance policies from a single unified dashboard. This streamlined experience allows teams to select from Azure OpenAI Service, NVIDIA Nemotron open models, and other foundation models while designing, testing, and deploying agent workflows in one cohesive environment.

NVIDIA Nemotron Support and Next-Gen GPU Deployment

Foundry Agent Service now supports NVIDIA Nemotron open models, enabling commercial use of open-source LLMs within the agent platform. Microsoft also announced the world's first hyperscale cloud deployment of NVIDIA Vera Rubin NVL72, alongside plans to deploy hundreds of thousands of liquid-cooled Grace Blackwell GPUs in under one year. This massive infrastructure investment underscores Microsoft's commitment to scaling AI inference and training capabilities for enterprise workloads.

Global Expansion of Hosted Agents Including Japan East

The GA release expands hosted agent capabilities to six additional regions, including Japan East. This enables Japanese enterprises to deploy agent infrastructure while meeting data sovereignty requirements. For highly regulated industries such as finance, healthcare, and manufacturing, the ability to combine low latency with domestic data residency represents a significant competitive advantage in the Japanese market.

Open-Sourcing Physical AI Toolchain for IoT Integration

Microsoft introduced the Physical AI Toolchain, an open-source framework integrating Azure IoT Operations, Azure Fabric, and NVIDIA Omniverse for building agents that interact with the physical world. Available on GitHub, this toolchain supports use cases such as robotic control in manufacturing, warehouse automation, and smart building management. Developers can simulate agent behavior in virtual environments, analyze IoT data in real time, and deploy automated control workflows with no-code or low-code approaches.

Production-Ready Observability and Governance

The GA status of Observability in Foundry Control Plane provides centralized dashboards for monitoring agent logs, inference costs, error rates, and response times. In multi-agent environments where multiple agents collaborate on tasks, visibility into which agent handled which task and where bottlenecks occur is essential. Governance features allow fine-grained control over data sources agents can access, actions they can execute, and approval workflows, ensuring compliance with regulatory requirements.

Conclusion — Oflight Supports Your Agent Deployment

The general availability of Microsoft Foundry Agent Service marks the beginning of an era where enterprises can safely and reliably operate AI agents in production. With the deployment of NVIDIA's latest GPUs, Japan East region support, and the open-sourcing of Physical AI Toolchain, Japanese companies now have access to world-class agent infrastructure. Oflight Inc. provides comprehensive support for agent development and deployment using Azure Foundry Agent Service, including integration with existing systems and governance design. Please contact us for a consultation on enterprise AI agent infrastructure.