| Active sessions | Count of running notebook sessions, with delta vs yesterday. | Tracks interactive compute consumption and spot unexpected overnight sessions. |
| Active endpoints | Count of live inference endpoints, with warm replicas called out. | Warm replicas cost money even with no traffic; cold ones have latency on first hit. |
| Running training jobs | Count of in-progress training jobs + total GPU-hours consumed today. | Gives a real-time view of your cluster's training workload. |
| Spend today | Burn rate vs your daily cost ceiling. | Early warning before you breach a budget threshold. |
| Spend MTD | Month-to-date spend with an end-of-month forecast. | Capacity planning and finance reporting. |
| Idle GPU hours | Paid-for GPU time where utilization stayed under 5%. | The most actionable waste metric — drive this toward zero. |