Skip to content

Platform direction · In development

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.

Observe and model

Reconcile workload, electrical, cooling, reserve, and on-site generation inputs. Unknown or inconsistent evidence must remain visible.

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.

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.

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.

Inside-the-fence decisions have an outside context.

The development direction is to examine facility decisions within explicit technical limits while making relevant grid and local constraints visible. Any future grid-facing service would require separately validated capability, agreements and operating authority.

Grid constraintsFacility perimeterLocal community

A fenced data center, separate grid infrastructure, and a nearby local community. The scene does not define an assessment boundary or operating authority. Synthetic concept. Grid services, resource impacts and community benefits require separate evidence and review.

The current offer remains a paid, bounded assessment of authorized historical inputs. No live connection or equipment control.

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.

Explore the dispatch-envelope example

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.

Review evidence and authority boundaries

Proof before autonomy

Start with one capacity question.

Agree the decision, evidence boundary, responsibilities, and paid scope before work begins.