Microsoft Adds Hugging Face Model Support to Foundry

Microsoft announced Foundry Managed Compute and Hugging Face model integration at Microsoft Build 2026. This allows developers to easily deploy open-weight models from the Hugging Face ecosystem on top of Azure infrastructure with a single click.

Key Features of the Foundry Platform:

• Broad Model Selection: Access to a wide range of models including Microsoft, OpenAI, Anthropic, Meta, Mistral, DeepSeek, and Hugging Face through a single endpoint and SDK (Python, C#, JS, Java).
• Foundry Agent Service: Provides multi-agent orchestration, memory, knowledge base (Foundry IQ), and tool connection capabilities.
• Managed Compute: Supports the latest runtimes such as vLLM, SGLang, and TensorRT-LLM, offering a PaaS (Platform-as-a-Service) style GPU management service that automatically manages GPU topology.

Enterprise Features and Security:

• All models share enterprise-grade security, governance, observability, and a unified billing system.
• Supports content safety filters, task compliance guardrails, AI red-teaming agents, private networking, and Azure Policy integration.
• In addition to Pay-per-token pricing and Provisioned Throughput for high-performance workloads, a Managed Compute option that charges for GPUs on an hourly basis is also offered, enhancing cost efficiency.