Global Standard is the default for a reason and the reason is not performance. It is a routing behaviour with quota consequences, and both halves surprise people who chose it because it was preselected.
The SKU name in code is GlobalStandard
. Microsoft describes it as using Azure’s global infrastructure to dynamically route traffic to available datacenters, and lists three concrete consequences: it provides the highest default quota, it eliminates the need to load balance across multiple resources for throughput purposes, and it is the type new models arrive on first.
The launch order is documented and it is a planning input. New deployment types become available Global first, then Data Zone, then single region — and single-region types arrive last, have no guaranteed availability date, and depend on capacity that frees up as older models retire. A design that requires a model pinned to one region is a design that may wait indefinitely for that model. Microsoft, Understanding deployment types in Foundry Models.
Global Standard also supports priority processing on a pay-as-you-go basis, which is a separate rate for faster responses on the same deployment.
The distinction Microsoft draws is between data at rest and data in flight, and only the second one varies by deployment type. Data stored at rest remains in the designated Azure geography for every type. Inferencing data is processed differently:
“Any Azure region” is the phrase to take to a compliance conversation before you deploy rather than after. Microsoft also notes it can add regions to a data zone without prior notice to improve capacity and availability, so a Data Zone commitment is a commitment to the zone, not to a fixed region list.
Quota accounting for Foundry changed recently and the change matters most to exactly this deployment type. Microsoft documents subscription-level quota management as having started after 7 May 2026, rolling out model by model, and describes the resulting pools directly:
The consequence is worth stating flatly: creating a second Global Standard deployment of the same model and version in another region does not give you more throughput. It gives you a second endpoint drawing on the same allowance. Teams reaching for a second region to escape 429s often find the 429s follow them, and this is why. Real headroom means a different model, a different deployment type, a higher quota tier, or a different subscription.
Quota itself is now tiered — Microsoft documents a Free Tier plus Tiers 1 through 6, with automatic upgrades based on consumption trends and on your commercial relationship, and an opt-out flag for customers who use quota as a spend control. Which tier you are in is readable through the control-plane API, and it is the number that actually sets your limits; any single figure quoted elsewhere is one row of one tier.
Two ARM APIs answer the two questions that actually matter here, and they are different questions. The Usages API takes a location and returns every quota line in that region with its currentValue
and limit
— a ledger of what you have consumed. The Model Capacities API takes a model name, version and format and returns available capacity per location and deployment type across the subscription — a map of where you could deploy. Use the first to alert on consumption, the second to pre-check before creating or scaling a deployment.
GET https://management.azure.com/subscriptions/{subscriptionId}
/providers/Microsoft.CognitiveServices/locations/{location}/usages
?api-version=2024-10-01
One caveat on both: Microsoft notes that the Foundry portal and the capacity APIs currently return quota and capacity information for models that are retired and no longer available for new deployments. A capacity figure alone is not evidence that a model can still be deployed.
A related trap on the way out. Deleting an Azure OpenAI resource through the portal is blocked while deployments still exist, which forces the quota to be released cleanly. Deleting it through the REST API or another programmatic route bypasses that check — and the associated quota allocation then stays unavailable for 48 hours until the resource is purged. A Terraform destroy-and-recreate against a quota-constrained model can therefore fail to recreate for reasons that have nothing to do with the new configuration.
Microsoft is consistent on this across three articles: Global Standard and Data Zone Standard give more consistent latency for customers with low to medium traffic, and customers with high consistent volume might experience greater latency variability. The threshold is set per model and is expressed as a monthly usage tier in tokens — a number that is per model and moves, so read it rather than remember it.
The documented consequence of exceeding it is not a hard error. Response latency may increase significantly, and Microsoft states it can vary and in some cases be more than twice as high as when operating within the tier. That degradation is invisible to any check that only looks at status codes, which is a good argument for alerting on your own p95 rather than on error rate alone.
Where latency variance is unacceptable, the documented answer is provisioned throughput — reserved capacity with lower and more consistent latency, at the cost of paying for it whether you use it or not.
“Dynamically routes to available datacenters” reads like resilience. Microsoft’s own note says otherwise, and it is the most important sentence on this page: with Global Standard and Data Zone Standard deployment types, if the primary region experiences an interruption in service, all traffic initially routed to this region is affected.
Global routing distributes load. It does not survive the loss of the region your requests entered through. Resilience against a regional outage is still an application-level concern — a second resource, a second endpoint, and code that fails over between them, which is what the failover page builds.
One organisational note: if you need to prevent Global deployments entirely for residency reasons, Azure Policy can deny them by SKU. The policy matches on Microsoft.CognitiveServices/accounts/deployments/sku.name
equal to GlobalStandard
, which is a far more reliable control than a written guideline about which dropdown to pick.
Deployment type SKU codes, the quota pooling behaviour and the usage tier thresholds are Microsoft’s documented state at the time of writing. Subscription-level quota pooling in particular is a recent and still-rolling change — verify how it applies to your model before sizing.