We noticed this behavior on ci.jenkins.io where a number of agents failed to provision due to timeout errors in the Azure API. The resource group was filled with VMs but there were not corresponding agents listed in Jenkins.
Looking at the logs I noticed that the instance was complaining that it could not provision more machines due to the quota, despite very few agents actually being listed under /computer.
After deleting the resource group in question, Jenkins started provisioning new agents while still complaining about the quota in the logs.
I have no idea what happened here, but it seemed like the Azure timeout exceptions accumulated and caused some error condition with managing the quota.