Microsoft Azure → Amazon Web Services
AWS equivalent of Microsoft Foundry Models (incl. Azure OpenAI)
The closest AWS equivalents of Microsoft Foundry Models (incl.
About Microsoft Foundry Models (incl. Azure OpenAI)
| Provider | Microsoft Azure |
|---|---|
| Status | Generally available |
| What it is | Catalog of 10,000+ models. "Sold by Azure" (Azure OpenAI GPT family, Microsoft MAI/Phi, xAI Grok, DeepSeek, Meta, Mistral, Cohere, Black Forest Labs, Moonshot) plus partner/community models (Anthropic Claude, Hugging Face, and more). Deploy serverless (per-token) or on managed compute. |
| Documentation | learn.microsoft.com |
| Pricing | Official pricing page |
AWS mappings (4)
| Capability | AWS equivalent | Match |
|---|---|---|
| Foundation model API platform | Amazon Bedrock | Direct match |
| Cloud provider's own foundation models | Amazon Nova | Partial match |
| Open-weight model hub on dedicated compute | Amazon SageMaker JumpStart, Amazon Bedrock Marketplace | Direct match |
| Prebuilt image analysis | Amazon Rekognition | Partial match |
Generative AI platforms & models(3)
Foundation model API platform Direct match
The Azure equivalent of Amazon Bedrock is Microsoft Foundry Models (which includes Azure OpenAI). Both are managed, pay-per-token, multi-vendor model APIs.
- Model overlap is now large. Both offer OpenAI GPT models, Anthropic Claude, xAI Grok, Meta Llama, Mistral, DeepSeek, Cohere, and Moonshot Kimi. Check per-model region and lifecycle on each side.
- Only on Bedrock (first-party), Amazon Nova. Bedrock also lists Qwen, Stability AI, TwelveLabs, Writer, Luma, AI21, Z.ai GLM, MiniMax, NVIDIA Nemotron, and Google Gemma.
- Only on Foundry (first-party), Microsoft MAI and Phi models plus Model Router. Foundry also lists Black Forest Labs FLUX, Sora (video, preview), and 10,000+ catalog models, including Hugging Face models on managed compute.
Cloud provider's own foundation models Partial match
Amazon Nova's closest Azure counterpart is Microsoft's own models in Foundry (MAI and Phi, plus Model Router). But Microsoft's flagship LLMs on Azure are mainly OpenAI GPT models, so there's no one-to-one family match.
- Nova covers understanding (Micro/Lite/Pro, Nova 2 Lite), creative (Canvas image, Reel video), speech-to-speech (Nova 2 Sonic), and multimodal embeddings, all from Amazon.
- On Azure, first-party models include the MAI models (for example MAI-Image), Phi small language models, Model Router, and healthcare AI models. General-purpose frontier LLM work on Azure usually means Azure OpenAI GPT models, which are "sold by Azure" but built by OpenAI.
- For speech-to-speech, Nova Sonic maps more naturally to Azure Voice Live or realtime audio models than to a Microsoft-built LLM.
Open-weight model hub on dedicated compute Direct match
SageMaker JumpStart and Bedrock Marketplace map to the Microsoft Foundry model catalog's managed-compute deployments (and the Azure Machine Learning model catalog).
- Foundry advertises 10,000+ catalog models (including Hugging Face) and states that partner/community models are the large majority. Hugging Face managed-compute deployments currently use hub-based projects in the Foundry (classic) portal.
- JumpStart supports private and curated hubs for governing which models teams may use. Bedrock Marketplace models are called through Bedrock APIs, so Guardrails, Knowledge Bases, and Agents integrate with them.
- Billing is per instance/VM-hour on both sides when you deploy to dedicated compute.
Vision & documents(1)
Prebuilt image analysis Partial match
Amazon Rekognition's image features map partly to Azure Vision Image Analysis, but that API is deprecated and retires on September 25, 2028. Microsoft now recommends Foundry multimodal models and Content Understanding instead.
- Microsoft's own AWS comparison now maps Rekognition to Foundry models plus Azure ML AutoML for vision. It says no single current Azure service covers every Rekognition capability.
- Microsoft's Image Analysis migration guide sends OCR to Document Intelligence, faces to the Face API, embeddings to Cohere Embed (Foundry), and everything else to GPT models or Content Understanding.
- Rekognition offers celebrity recognition and Face Liveness, which Azure's guidance doesn't cover the same way (Azure Face has liveness under Limited Access).