AI Model Availability and Processing Constraints

Question

Hi, I need to set up a deployment using an LLM model where all data is processed and stored in Australia - i.e. it cannot leave Australian soil.

Is there a GPT model/a list of models we can use that allow this? Not including the dedicated/provisioned allocation, which costs much more than a pay-per-use setup?

If so, which deployment type should I select for this? I see "Standard" has GPT 4o, but I am not sure if the data stays in Australia.

Answer 1

Hello Guy,

Welcome to Microsoft Q&A and Thank you for reaching out.

I understand that you're trying to set up a deployment using an LLM model while ensuring that all your data stays within Australia.

Here’s the guidance for your scenario:

Can I keep all LLM data processing inside Australia?

Yes, but only if you use Azure OpenAI with the correct deployment type and region.

The correct setup is:

• Deploy in Australia East
• Use Standard (pay-per-use) deployment
• Choose a model available in that region (e.g., GPT-4o)

1. Model Availability in Australia

Before deployment, you should verify which models are available in Australia East by checking the official Azure OpenAI model availability documentation.

Currently in Standard regional deployment in Australia East:

• GPT-4o is available

Newer global models (e.g., GPT-5 family or certain reasoning models) may not yet be available in that region

So for a consumption-based (pay-per-use) setup fully hosted in Australia, GPT-4o via Standard deployment is the practical choice today.

2. Data Residency vs Processing Location

This is the most important distinction:

Standard Regional Deployment (Australia East)

If you:

Create your Azure OpenAI resource in Australia East

Deploy the model as Standard (regional)

Then:

Prompts and responses are processed in Australia

Model inference (GPU compute) occurs in Australia

Customer content is stored in Australia (if logging enabled)

The service operates within Azure’s Australian datacenters

This satisfies typical “data must not leave Australia” compliance requirements.

3. Should You Use “Data Zone” Deployment?

Azure OpenAI supports Data Zone deployment, which ensures:

Data is processed and stored within the designated geography

It does not leave that geography for model processing

If your compliance team requires strict geographic enforcement, Data Zone deployment in Australia is the safest architectural choice.

However, Data Zone availability depends on region and model support.

If available for your chosen model in Australia, it provides stronger residency guarantees.

If Data Zone is not available for your required model, then Standard Regional in Australia East is still compliant for most Australian residency requirements because inference and storage remain in-region.

4. Avoiding Provisioned / Dedicated Costs

You mentioned avoiding provisioned allocation (PTU) due to cost.

Standard (on-demand) deployment is consumption-based

No minimum throughput commitment

No dedicated GPU reservation

You pay per token usage

Provisioned throughput is only required if you need:

Guaranteed capacity

Predictable latency at scale

Enterprise SLA for heavy workloads

For most workloads, Standard deployment is sufficient and cost-effective.

Recommended Setup,

1. Create Azure OpenAI resource in Australia East
2. Select Standard (regional) deployment
3. Deploy GPT-4o
4. Confirm endpoint region is Australia East
5. Enable Data Zone deployment if available and required by compliance

Please refer this

1. Azure OpenAI Standard Deployment Model Availability
2. Understand Azure OpenAI Deployment Types and Data Processing

I hope this helps, do let me know if you have any further queries.

Thank you!