@Tomislav Đurić : when the requests first reach ARM, the requests are checked on the ARM level and throttled if the requests exceed the limit. Once the request passes through the ARM, it goes to the particular resource provider ( for example, CRP - compute resource provider, NRP- network resource provider), and then limits are checked against those resource provider and the corresponding resource.
For more details, kindly refer : https://learn.microsoft.com/en-us/azure/azure-resource-manager/management/request-limits-and-throttling