- Closed loop
- The highest DCIM maturity stage, in which the facility model does not merely advise but drives response, adjusting cooling setpoints or rebalancing power automatically within defined, human-supervised envelopes. Approached carefully because automated actuation of a physical plant carries real risk.
- Compute-side orchestration (AI factory)
- The software layer that decides which workload runs on which processor, when, and for how long, schedulers such as Kubernetes with GPU awareness or Slurm. Its concern is the computation (FLOPs, job throughput, queue depth), forming a nervous system distinct from the facility's.
- Coupled constraint (power, cooling, space)
- The principle that a facility's three physical budgets are not independent. Nearly every watt delivered becomes heat that must be removed (power in equals heat out), and a rack's position governs how both power and heat reach it, so the three must be optimized jointly.
- DCIM (Data Center Infrastructure Management)
- The software layer that monitors, measures, and manages a facility's physical infrastructure, power, cooling, and space. It maintains a live, queryable model of the plant and enables provisioning, capacity forecasting, and failure detection. It does not schedule or run compute workloads.
- DCIM maturity gradient
- A four-stage scale of how an operator governs the physical plant: (1) spreadsheets and floor walks, (2) live sensor monitoring, (3) predictive modeling via a digital twin, (4) closed-loop automated response within bounded authority. Progress up it is costly and hard to fake, making position a signal of competence.
- Direct-to-chip liquid cooling
- Removing heat with fluid loops routed to the processor rather than with air, adopted as rising rack densities exceed what air can handle. It makes the facility a hydraulic system with finite flow, pressure, and supply-temperature limits, tightening the coupling of power, cooling, and space and raising the value of a physics-aware twin.
- Facility digital twin
- A physics-aware model of the physical plant that is continuously synchronized with live sensor telemetry and can answer counterfactual questions, predicting how power, cooling, and space respond to a proposed change or failure before it happens. Distinct from a static 3D model or a present-tense dashboard.
- Stranded capacity
- Usable capacity a facility cannot fill because available power, deliverable cooling, and physical position never coincide at any location, even though each budget individually shows slack. It is invisible to anyone tracking the three variables separately and represents money spent on a facility that cannot be filled.