When you switch on an AI model in Foundry, you choose where in the world it does its thinking and whether you pay per use or rent space in advance.
Each Foundry model deployment has a type. The type sets where prompts are processed (anywhere, within a region like the EU, or in one Azure geography), how you pay (per token, reserved capacity, or cheaper batch jobs), and how steady the response times are.
Serverless API deployment types combine processing scope (Global, Data Zone US/EU/APAC, regional geography) with billing (Standard pay-per-token, Provisioned hourly reserved, Batch ~50% cheaper, 24-hour target). Data at rest stays in the resource's geography for all types. Global Standard is Microsoft's default; Provisioned reduces latency variance.
Two questions in one choice
Microsoft's documentation says the deployment type decides three things: where your data is processed, how you pay, and performance characteristics. I find it easier to think of it as two questions.
Where can my prompts and responses be processed? And do I want to pay per use, reserve capacity, or send big jobs to run later at a lower price?
Where it's processed
Microsoft gives three options.
| Scope | What Microsoft says about processing |
|---|---|
| Global | May be processed in any Azure region where the model is deployed |
| Data Zone | Processed only within a Microsoft-defined zone: US, EU or Asia Pacific |
| Regional (Standard or Regional Provisioned) | Processed within the Azure geography you choose |
For every type, Microsoft says data stored at rest stays in the designated Azure geography. Only the place where inference happens changes.
Two details are worth knowing. The EU data zone follows the Azure EU Data Boundary, which can include EFTA countries such as Norway and Switzerland, and Microsoft says it can add regions to a data zone without prior notice. For a UK organisation, the EU zone isn't the UK; if processing must stay in the UK, you'd need a regional deployment in a UK region, if the model you want is offered there.
How you pay
| Billing | How it works |
|---|---|
| Standard | Pay per token |
| Provisioned | Hourly reserved capacity, with lower latency variance |
| Batch | Asynchronous jobs with a 24-hour target turnaround, which Microsoft says cost 50% less than Global Standard |
Microsoft notes that Global and Data Zone Standard can show more latency variation at high, steady volumes, and recommends provisioned types where low variance matters. Global deployments also get new models and features first.
Microsoft's quick guide
Microsoft's own recommendation table starts with Global Standard as the default for "newest models, lowest price, broadest regions". It recommends Data Zone types to keep processing within a zone, regional types to keep it within a geography, provisioned types for predictable throughput, and batch for large jobs that aren't urgent.
Why this is a decision in itself
The deployment type is a trade-off between cost, model choice and where data goes. It's the kind of decision that's easy to make once in a hurry and hard to revisit. I'd write it down with the reason, so the next person knows whether "Global" was a choice or a default.
A worked example
This scenario is illustrative. A UK charity wants a model to summarise 20,000 archived case notes once, and then to help staff draft replies to new enquiries every day.
| Workload | Constraint | Choice |
|---|---|---|
| One-off archive summaries | Not urgent; data protection review allows EU processing | Data Zone Batch, if available for the model |
| Daily reply drafting | Staff are waiting; UK processing required by policy | Regional Standard in a UK region, if the model is offered there |
The same model could end up with two deployments for two decisions. If the daily model isn't offered in a UK region, that's a reason to revisit the model choice before the processing rule.
What I'm still checking
Model availability differs by region and deployment type, and Microsoft keeps a separate region availability page for each category. I haven't checked which current models are offered as regional deployments in UK regions. That's the first thing I'd look up for a real project.
Sources
- Microsoft Learn, Deployment types for Microsoft Foundry Models, accessed 11 October 2026.
- Microsoft Learn, Microsoft Foundry architecture, accessed 11 October 2026.
- Microsoft Learn, Region availability for Foundry Models sold by Azure, accessed 11 October 2026.