Constraint model
Capacity is bound by the tightest cross-domain constraint
Usable AI data center capacity is bounded by coupled constraints, not a single electrical ceiling. Workload timing changes heat, cooling demand, storage posture, generation reserves, water exposure, network dependencies, and SLA risk. A cross-domain capacity process evaluates those interactions in the same operating window and accepts only the capacity whose binding constraint and margin can be shown.
Definition
What it is
A cross-domain constraint model represents the dependencies that can make an otherwise feasible power action unsafe, unavailable, or commercially unusable.
Category boundary
What it is not
It is not a collection of independent dashboards or a sum of best-case headroom values. Capacity cannot be counted twice across domains that depend on the same reserve or recovery path.
Operational significance
Why operators care
Operators see which constraint actually binds a request and can evaluate repairs—such as timing, placement, or asset commitment—without hiding the cost in another facility domain.
Mechanism
How the decision path works
- Step 01
Normalize constraints into a shared time window and topology context.
- Step 02
Propagate candidate actions across power, thermal, water, storage, workload, and SLA models.
- Step 03
Identify the earliest binding constraint and margin to limit.
- Step 04
Repair the action only when the changed action remains provable across every domain.
Textual diagram equivalent
Decision flow
Requested AI load ↔ workload placement ↔ heat and cooling ↔ power and storage ↔ bridge power and reserves ↔ SLA and recovery → one accepted or refused capacity result.
Synthetic coupled-constraint example
Visibly scoped example
Synthetic illustrative scenario—not a customer or production result. Electrical headroom exists, but a proposed workload placement is rejected because the thermal recovery assumption is unsupported after a cooling-component outage.
Failure and no-proof cases
- Models use different timestamps, topology versions, or operating windows.
- One reserve is credited to multiple independent capacity claims.
- A domain reports a scalar score without the constraints needed for runtime checking.
Limitations
- The model is only as complete as the declared dependencies and trusted telemetry.
- Site engineering limits and protective settings remain authoritative.
- Uncertain or unavailable domains should produce no-proof rather than inferred capacity.
Evidence and sources
Inspect the basis, not just the answer
Local evidence objects show the GridNinja proof contract. Primary sources provide external standards and risk-management context; they do not validate GridNinja performance.
Related operator resources
Continue the proof path
- The platform for runtime-assured virtual capacitySee how GridNinja coordinates workloads, cooling, storage, and bridge power inside a runtime-assured dispatch envelope.
- For AI Clouds Racing Against the GridAccelerate AI cloud time-to-power by proving virtual capacity across power, cooling, workloads, reserves, and bridge power.
- Make bridge power provably usefulCoordinate bridge power, storage, generation, cooling, and AI workloads inside a visible, runtime-assured dispatch envelope.
Proof before autonomy
Test the evidence boundary against your capacity question
Start with a Capacity Audit or read-only Shadow Mode discussion. GridNinja does not require control authority to identify where a capacity claim remains unproven.