Step 01 — Real-time sensing
One controller reads
every power and compute signal in the facility.
The MaatiAI edge controller connects locally to facility meters, UPS systems, batteries, generators, rack-level PDUs, and GPU workload schedulers — all at once, with no data leaving the perimeter.
- PowerFacility and rack-level draw, UPS loading, battery state
- ComputeGPU utilization, workload queue, priority flags
- BackupGenerator availability, cooling headroom, grid signals
Step 02 — Constraint forecasting
Know the peak
before it happens.
The system maintains a live operating model of site power and predicts when incoming GPU demand will exceed safe facility limits — before the constraint hits, not after.
- ScenarioSite limit 1 MW — current demand 870 kW — new workload adds 180 kW
- DetectionUncontrolled total would reach 1.05 MW in the next 4 minutes
- WindowMaatiAI flags the constraint with time to act, not time to react
Step 03 — Power + compute in one loop
Least disruptive action.
Every time.
MaatiAI finds the coordinated response — across batteries, GPU power caps, and workload scheduling — that keeps the site inside its power envelope without rejecting the workload.
- BatteryDispatch 40 kW from on-site storage to bridge the peak
- GPU capTemporarily reduce selected non-priority GPUs by 30 kW
- SchedulerDefer one 20 kW background job — priority workload runs in full
