🇺🇸 USA · Azure OpenAI Service
Status: 🟩 COMPLETE 🟦 LIVING Last updated: 2026-06-26 Plain-English tagline: Microsoft’s enterprise channel for OpenAI models. Same GPT / o-series / Sora / gpt-image-1 you’d get from OpenAI directly, but wrapped in Microsoft enterprise terms with no-train + AUS data residency.
Front-matter facts
| Field | Value |
|---|---|
| Vendor | Microsoft (Redmond, USA) — via partnership with OpenAI |
| Country / origin | 🇺🇸 USA |
| Recommended for Australian users? | ✅ Yes — Australia East region available for AUS data residency |
| Privacy summary | No training on customer data (contractually); AUS data residency; “Limited Access” exemption available for high-compliance customers (no abuse-monitoring retention) |
| Free tier | Limited via Azure Free Account |
| Paid tiers | Pay-per-token + PTU (Provisioned Throughput Units) for reserved capacity; AUD via Microsoft Australia |
| First released | 2021 (initial preview); generally available 2022 |
| Last reviewed | 2026-06-26 |
| Official site | https://azure.microsoft.com/products/ai-services/openai-service |
What it is
Azure OpenAI Service is Microsoft’s enterprise way to access OpenAI models — GPT-5 family, o-series reasoning models, gpt-image-1, Sora, Whisper, embeddings models, all the OpenAI catalog — but delivered through Microsoft Azure infrastructure with Microsoft enterprise commitments.
Microsoft has a deep, complex, evolving partnership with OpenAI:
- Microsoft has invested ~US$13B+ in OpenAI
- Microsoft is OpenAI’s largest cloud partner (most of OpenAI’s compute runs on Azure)
- Azure OpenAI Service is the Microsoft-branded enterprise distribution of OpenAI models
- In return, Azure customers get exclusive enterprise access patterns
What you get via Azure OpenAI vs direct OpenAI:
- AUS data residency (Australia East region)
- No-training-by-default + tenant isolation (matches direct OpenAI API)
- Microsoft Azure enterprise terms — same legal / billing / support model as the rest of Azure
- PTU (Provisioned Throughput Units) — reserved capacity for guaranteed latency at scale
- Limited Access mode — option to disable abuse-monitoring retention for high-compliance customers (medical / legal / government)
- IRAP compatible for AUS government
- Microsoft Purview integration for data governance
What you DON’T get via Azure OpenAI vs direct OpenAI:
- Cutting-edge model availability lags slightly (new GPT models may launch on OpenAI direct first, then Azure)
- Some experimental features (early Operator / Voice / etc.) are OpenAI direct first
- Slightly different pricing structure
What you’d use it for
- Enterprise production AI on GPT models with Microsoft enterprise terms
- AUS data residency for OpenAI models specifically
- Regulated industries (banking, healthcare, government) needing OpenAI access with compliance wrapper
- Microsoft-stack organisations already in Azure
- High-compliance use cases with Limited Access (no abuse-monitoring retention)
- High-volume production with PTU (Provisioned Throughput Units)
When NOT to use Azure OpenAI:
- For prototyping / personal use (direct OpenAI API or ChatGPT simpler)
- For latest-and-greatest model access (direct OpenAI often gets new models first)
- For non-OpenAI models (use Azure AI Foundry’s broader catalog, or AWS Bedrock for Claude / Llama)
How to use from Australia
- Azure subscription
- Request access to Azure OpenAI Service (small approval form; was previously gated, increasingly open)
- Create Azure OpenAI resource in Australia East region
- Deploy specific model variants (gpt-5, gpt-image-1, etc.) into your resource
- Call via REST API or SDK (OpenAI-compatible)
- Billed via Azure billing in AUD
Example call
from openai import AzureOpenAI
client = AzureOpenAI(
azure_endpoint="https://your-resource.openai.azure.com/",
api_key="...",
api_version="2025-03-01"
)
response = client.chat.completions.create(
model="gpt-5", # deployment name
messages=[{"role": "user", "content": "Hello"}]
)What it costs
Pay-per-token (most use)
- Similar to OpenAI direct pricing with small Microsoft Azure margin
- Examples (verify current Azure OpenAI pricing):
- GPT-5: ~US$X per million input/output tokens
- GPT-5 mini: much cheaper
- gpt-image-1: per-image pricing
- Sora: per-second pricing
- Whisper: per-minute audio
PTU (Provisioned Throughput Units)
- Reserved compute capacity for guaranteed throughput / latency
- Monthly commitment
- For high-volume production where consistent latency matters
Free Azure credits
- AUD $300 with Azure Free Account can be applied to Azure OpenAI usage
How it compares to alternatives
| Aspect | Azure OpenAI Service | OpenAI direct API | AWS Bedrock | Vertex AI |
|---|---|---|---|---|
| Model breadth | OpenAI only | OpenAI only | Multi-vendor | Multi-vendor |
| AUS data residency | Yes (Australia East) | Limited | Yes (Sydney) | Yes (Sydney + Melbourne) |
| Enterprise contracts | Microsoft enterprise | OpenAI enterprise | AWS enterprise | Google Cloud enterprise |
| Latest GPT models | Slight lag | First access | N/A | N/A |
| Pricing | Small Azure margin | Direct (slightly cheaper) | Per-vendor | Per-vendor |
| Limited Access (no abuse retention) | Yes | No | No | Limited |
| PTU / provisioned throughput | Yes | Limited | Yes | Yes |
| Best for | Enterprise OpenAI on Azure | Direct usage / prototyping | AWS shops / multi-model | GCP shops / Gemini |
For AUS enterprise wanting OpenAI models with compliance + data residency, Azure OpenAI Service is the standard pick.
Privacy / data handling
- No training on customer data — contractually committed
- Inputs retained for 30 days for abuse-monitoring by default — can be disabled via Limited Access application
- 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
Recent changes
- 2026: Sora + gpt-image-1 + GPT-5 family broadly available
- 2025: PTU expansion; Limited Access expanded
- 2024: Australia East availability for most models
- 2022-23: Initial GA + expansion
Gotchas
- Access approval was previously a barrier; mostly removed for established Azure customers
- Model availability per region varies — verify your specific models in Australia East
- Quota management — request uplift for production use
- Latest OpenAI features sometimes lag Azure availability (Operator, Voice Engine, etc. may be direct-only initially)
- Two billing channels — Azure usage (cheaper for AUS data residency) + OpenAI direct API (cheaper for prototyping). Don’t double-bill.
- Limited Access application — for HIPAA / sensitive workloads needing no-retention; involves an approval process
- Pricing differences vs direct OpenAI are real but small
See also
- Microsoft Azure overview 🟩 🟦
- Azure AI Foundry 🟥 — broader Microsoft AI platform
- ChatGPT 🟩 🟦
- OpenAI API 🟥 — direct alternative
- GPT models lineup 🟥
- Microsoft Copilot 🟩 🟦
- Microsoft 365 Copilot 🟩 🟦
- Microsoft MAI 🟩 🟦
- AWS Bedrock 🟩 🟦 — multi-vendor alternative
- Vertex AI 🟥 — Google equivalent
- privacy-and-data-training.md 🟩 🟦