🇺🇸 USA · Azure AI Foundry
Status: 🟩 COMPLETE 🟦 LIVING Last updated: 2026-06-26 Plain-English tagline: Microsoft’s enterprise AI development platform on Azure. Broader than Azure OpenAI Service — Foundry includes MAI, Llama, Mistral, Anthropic Claude, plus tooling for building / deploying AI apps end-to-end.
Front-matter facts
| Field | Value |
|---|---|
| Vendor | Microsoft (Redmond, USA) |
| Country / origin | 🇺🇸 USA |
| Recommended for Australian users? | ✅ Yes — available in Australia East with AUS data residency for relevant models |
| Privacy summary | No training on customer data; tenant-isolated; AUS data residency; IRAP compatible |
| Free tier | Limited free via Azure Free Account |
| Paid tiers | Per-token pricing per model + Azure infrastructure costs; AUD billing |
| First released | 2024 (consolidated from Azure AI Studio + Azure ML) |
| Last reviewed | 2026-06-26 |
| Official site | https://ai.azure.com |
What it is
Azure AI Foundry is Microsoft’s unified AI development platform — the successor / consolidation of earlier “Azure AI Studio” and “Azure Machine Learning Studio.” Foundry provides:
Model catalog (broader than Azure OpenAI Service)
- OpenAI models (GPT family, gpt-image-1, Sora) — Azure OpenAI Service is essentially Foundry’s OpenAI section
- Microsoft MAI — MAI-1, MAI-Voice, MAI-Vision
- Anthropic Claude (via Foundry’s Anthropic partnership)
- Meta Llama (4 / 5 family)
- Mistral models
- Cohere models
- DeepSeek open weights (Western-hosted) — note: Chinese-origin open weights, see DeepSeek entry
- Hundreds of other models from Hugging Face + partners
Platform features
- Prompt Flow — visual prompt engineering workflows
- Evaluation tools — model comparison, scoring
- AI agents — orchestrate multi-step workflows
- AI Search integration — Microsoft’s enterprise search for RAG
- Content safety — content moderation
- Responsible AI dashboard — fairness / bias / explainability metrics
- Fine-tuning — adapt models on your data
- Deployment — managed endpoints with auto-scaling
Azure AI Foundry is Microsoft’s answer to AWS Bedrock + SageMaker combined + Google Vertex AI combined — one platform for the full AI development lifecycle.
What you’d use it for
- Multi-model strategy within Azure — pick OpenAI / Claude / Llama / Mistral / MAI from one platform
- End-to-end AI development — prototype → evaluate → fine-tune → deploy
- Enterprise AI compliance — Microsoft enterprise terms + Responsible AI tooling
- Microsoft-stack organisations wanting non-OpenAI models too
- Production AI on Azure with broader model choice than Azure OpenAI Service alone
- AUS data residency for broader model catalog (Azure OpenAI alone covers OpenAI; Foundry extends this)
How to use from Australia
- Azure subscription
- Go to
ai.azure.com - Create AI Foundry project / hub
- Pick Australia East region for AUS data residency
- Browse Model Catalog
- Deploy selected models to endpoints
- Use Prompt Flow for development
- Billed via Azure billing in AUD
What it costs
Per-token (per model)
- Same pricing as Azure OpenAI Service for OpenAI models
- Anthropic Claude via Foundry: Anthropic’s commercial terms within Azure billing
- Llama / Mistral / Cohere: per-model pricing
- MAI models: Microsoft pricing
Infrastructure
- Endpoint hosting (managed compute) — typically per-hour
- AI Search index storage
- Fine-tuning compute (training) — significant for large models
Hidden costs
- Endpoint hosting accumulates if you leave unused endpoints running
- AI Search index has its own pricing tier
How it compares to alternatives
| Aspect | Azure AI Foundry | AWS Bedrock | Vertex AI | Azure OpenAI Service alone |
|---|---|---|---|---|
| Scope | Broadest within Azure | Broad (AWS) | Broad (Google) | OpenAI-only |
| Microsoft-stack integration | Best | Limited | Limited | Best for OpenAI |
| End-to-end ML lifecycle | Yes | SageMaker for training | Yes | Limited |
| Multi-model catalog | OpenAI + MAI + Claude + Llama + Mistral + Cohere | Multi-vendor | Multi-vendor | OpenAI-only |
| AUS data residency | Yes (Australia East) | Yes (Sydney) | Yes (Sydney + Melbourne) | Yes (Australia East) |
| Best for | Azure shops + multi-model + end-to-end | AWS shops + multi-model | GCP shops + multi-model | Pure OpenAI Azure use |
For Azure-shop organisations wanting more than just OpenAI, Foundry is the natural choice.
Privacy / data handling
- No training on customer data — contractually committed
- AUS data residency via Australia East
- IRAP compatible
- HIPAA, GDPR, SOC 2, ISO 27001 compliant
- Tenant-isolated by Azure subscription
- Microsoft Purview integration for governance
- Responsible AI dashboard for explainability / bias detection
Recent changes
- 2026: MAI-1.5 / 2 added; expanded model catalog
- 2025: Foundry consolidated from Azure AI Studio + Azure ML Studio
- 2024: Initial Foundry launch
Gotchas
- Foundry vs Azure OpenAI Service vs Azure ML confusion — Foundry is the umbrella; Azure OpenAI Service is a subset; legacy Azure ML still exists
- Model availability per region varies — verify your models in Australia East
- Endpoint management matters — don’t leave unused endpoints running
- Pricing complexity with multiple model providers + infrastructure costs
- For pure OpenAI use, Azure OpenAI Service alone is simpler
- For multi-vendor without Microsoft preference, AWS Bedrock often simpler
- Quota management required for production
See also
- Microsoft Azure overview 🟩 🟦
- Azure OpenAI Service 🟩 🟦 — OpenAI subset of Foundry
- Microsoft MAI 🟩 🟦
- Microsoft Copilot 🟩 🟦
- Microsoft 365 Copilot 🟩 🟦
- AWS Bedrock 🟩 🟦 — AWS equivalent
- Vertex AI 🟩 🟦 — Google equivalent
- Claude API overview 🟩 🟦
- Decision frameworks — AWS vs Azure vs GCP for AI 🟥