Microsoft launched Foundry Managed Compute: thousands of Hugging Face models with one click

On July 7, 2026, Microsoft launched Foundry Managed Compute — a managed GPU platform for deploying open-source models from Hugging Face in the Azure cloud. Thousands of models from a curated catalog are now available with one click, complete with enterprise security, monitoring, and unified billing.

What is Foundry Managed Compute

Foundry Managed Compute is a managed GPU platform that handles GPU topology selection and infrastructure management. Developers work in terms of models: number of parameters, context length, latency-throughput trade-offs — Microsoft manages the servers.

The Microsoft Foundry platform itself is positioned as an agent AI stack with the widest model selection among cloud providers. It offers models from OpenAI, Anthropic, Meta, Mistral, DeepSeek, and Hugging Face — through a single endpoint and SDKs in Python, C#, JavaScript, and Java. The platform includes multi-agent orchestration, end-to-end tracing, content security filters, AI Red Teaming Agent, and Azure Policy integration.

Which Hugging Face Models Are Now Available?

The Hugging Face collection in Foundry covers all key modalities:

• Catalog updates — weekly, based on community trends and customer requests
• Formats: SafeTensors only, no executable third-party code
• Runtimes: vLLM, SGLang, TensorRT-LLM, NIM, Text Embeddings Inference, llama.cpp, hf-serve
• Accelerators in preview: NVIDIA A100, H100 (80 GB), AMD MI300X
• Modalities: LLM, multimodal models (VLM), ASR, embeddings, segmentation, image generation

Each model goes through a five-stage selection pipeline: trend identification → license and repository security verification → runtime assembly with CVE scanning → weight upload to Azure Storage → API validation and catalog publication.

Pre-built weights and runtime images are stored in Azure Storage in advance, so production deployment does not require outbound access to the Hugging Face Hub. Security patches and runtime updates are applied automatically without re-deploying the model.

How to Deploy a Model

Deployment takes five steps: select a model from the catalog, choose a deployment template, specify the number of instances, launch via the portal, CLI, SDK, or REST — and call through Foundry's unified endpoint. For example, for the Qwen3-32B model, four templates are available: combinations of vLLM with A100 or H100 accelerators and 40K or 128K token contexts.

Via the Python SDK, launching boils down to a single `begin_create_or_update` call, after which the model is invoked through the standard OpenAI-compatible SDK — exactly like OpenAI or Anthropic frontier models. Foundry agents integrate Hugging Face models through the same paths as closed-source models.

"We've combined the breadth of the open-source ecosystem with the operational layer that powers

Microsoft" — from the official partner announcement.

What This Means

The integration lowers the barrier for enterprise adoption of open-source models: instead of self-managing GPU clusters, runtimes, and security systems, organizations receive a managed service with unified Azure RBAC access policies, Azure Monitor monitoring, and per-deployment billing. Hugging Face brings together 15 million developers, 400,000 organizations, and over 3 million open-source models — now this entire ecosystem becomes a corporate asset on Azure infrastructure.

Frequently Asked Questions

Do I need access to the Hugging Face Hub for deployment?

No. Microsoft pre-loads model weights and runtime images into Azure Storage, so deployment proceeds without outbound requests to the Hugging Face Hub. This is essential for enterprise environments with restricted internet access.

How often is the model catalog updated?

The catalog is expanded weekly based on signals from the Hugging Face community and direct customer requests. All new models go through a five-stage security pipeline before publication.

*Meta is recognized as an extremist organization and banned in Russia.