Billed ~€1,100 for serverless GPU replicas that never started (Container Apps, Sweden Central)

On 24 and 28 Sept 2026, our Container Apps on the Consumption-GPU-NC24-A100 profile scheduled a new replica every minute for hours. The replicas failed with exceeded quota: consumption-gpu-a100 or never pulled their image. Only 1–3 replicas actually ran, but we were billed for ~170 A100 GPU-hours. The billing support form won't give us a refund route or a phone contact. Could a moderator help us open a billing case? We have system logs and cost exports ready.