01
Observe and model
Reconcile workload, electrical, cooling, reserve, and on-site generation inputs. Unknown or inconsistent evidence must remain visible.
Platform direction · specified
A control-plane direction, with evidence before authority.
GridNinja is developing an AI Data Center Virtual Capacity Control Plane for vendor-agnostic, inside-the-fence orchestration. The current commercial offer is a bounded historical-data assessment; this page describes the intended architecture.
The intended decision loop
Telemetry → model → decide → assure → prove. Each proposed action must carry its scope, constraints, margin to limit, evidence freshness, and authority boundary.
02
Decide and assure
Compare a requested profile with explicit conditions. Runtime assurance is intended to allow, repair, reject, or return no-proof; a website model does not operate equipment.
03
Prove and review
Keep the decision record, assumptions, audit log, and replay basis together. A reviewer must be able to reproduce the stated conclusion within its scope.
04
Integrate under site authority
Future integrations would work with existing operational systems and local policy. Supported interfaces, read access, and control permissions require separate validation.
Inspect a synthetic timed-dispatch example
The dispatch-envelope teaching example uses a separate timed maneuver and dataset from DEMO-01. Its model output is neither a commercial commitment nor an operationally accepted action.
Proof before autonomy
The intended path is historical assessment, separately agreed Shadow Mode, advisory review, and only then bounded autonomy under explicit envelopes. Expanded autonomy is conditional on accumulated evidence and approved authority, never implied by a successful demo.
Related operator resources
Continue the proof path
- How much virtual capacity is safe, for how long, and under which evidence?Learn how a dispatch envelope constrains data center actions by electrical, thermal, reserve, workload, water, and SLA limits before execution.
- Evidence informs a decision. Authority stays explicit.See how Shadow Mode, replay, allow / repair / reject decisions, audit logs, and proof packs establish evidence before bounded autonomy.
- See the decision, the conditions, and the unanswered question.Inspect a synthetic capacity decision brief. Compare requested and revised profiles, model screening, unknowns, and the unresolved business decision.
- A research direction for proof-backed AI capacity.Explore GridNinja's proof-backed AI capacity project framing for infrastructure partners, operators, and technical evaluators.
Proof before autonomy
Start with one capacity question.
Agree the decision, evidence boundary, responsibilities, and paid scope before work begins.