AIDC·AI.IO
ContactOpen Demo
v0.1 · ENGINEERING PREVIEWRUBIN-ERA READY

AI Data Center
Infrastructure Design Engine

Plan Rubin-era AI data centers with a unified engine for electrical sizing, liquid cooling, layout, validation, and export-ready engineering evidence.

Explore PlatformView Engineering WorkflowPrivate demo →
Rack density basis
≥ 120 kW
MV distribution
22.9 kV
Cooling topology
L2L · D2C
SYSTEM TOPOLOGY · OVERVIEWEPOD · RMU · CDU
UTILITY22.9 kVRMU RINGMV SWGRTRANSFORMERMV → LVUPSBattery / FlywheelRACK · ≥120kWRubin GPUD2C cold plateCDU · L2LIn-RowPRIMARY LOOP · PLANTChillers · Dry coolers

Schematic only · power and cooling decoupled at the CDU; secondary loop sized to per-rack thermal profile.

01 · WHY THIS EXISTS

Rubin-era density breaks the old design pattern.

The legacy AI data center workflow — siloed Excel sheets for electrical, separate decks for cooling, a layout drawn after both — was tuned for a 30–60 kW rack and air as the primary heat-transport medium. Rubin-class GPUs and ≥ 120 kW racks invalidate the assumptions underneath that workflow, not just the numbers on top.

Rack thermal envelope

Legacy basis

30–60 kW air-cooled

Rubin-era basis

≥ 120 kW direct-to-chip

Consequence · Rear-door and CRAH-only halls cannot absorb the heat flux. Plant interface moves upstream.

MV distribution

Legacy basis

Single-radial 22.9 kV feeders

Rubin-era basis

Ring-grouped RMU with redundant step-down

Consequence · Concurrent maintenance becomes a sizing constraint, not an afterthought.

Cooling topology

Legacy basis

CRAH / RDHx in white space

Rubin-era basis

L2L CDUs decoupling primary and secondary loops

Consequence · Secondary ΔT and CDU placement (In-Row vs EoR) drives floor layout, not the other way around.

Validation evidence

Legacy basis

Spreadsheets, screenshots, marked-up PDFs

Rubin-era basis

Engine output feeding pandapower / OpenFOAM / EnergyPlus

Consequence · Reviewers ask for the model, not the deck. The deck must be a view of the model.

DESIGN BASIS We treat power, cooling, layout, validation, and evidence as one engine output — not four hand-offs that drift apart between revisions.

02 · PLATFORM CAPABILITIES

One engine. Power, cooling, layout, validation, and evidence.

AIDC-AI.IO is a generic automatic design engine — not a site-specific template. Each capability shares the same data model, so changes propagate through every view, panel, and export.

ELECTRICALPROFILE-AWARE

Electrical sizing & 22.9 kV topology

MV distribution from utility intake to LV bus. RMU ring groups, transformer capacity, generator step-up, and UPS sizing planned against the actual hall load.

  • RMU rating & ring topology
  • MV→LV step-down strategy
  • UPS / generator sizing
  • Busway distribution
COOLINGPROFILE-AWARE

Liquid cooling & CDU planning

Direct-to-chip and rear-door options modeled per rack. CDUs placed in-row or end-of-row, decoupling primary and secondary loops at the right capacity step.

  • D2C cold-plate planning
  • Row vs End-of-Row CDU
  • Secondary loop sizing
  • Primary plant interface
LAYOUTPROFILE-AWARE

EPOD / RMU / UPS / generator layout

Hall, tech space, plant, and EX-plant zoning. EPOD blocks, RMU rooms, UPS rooms, generator yards, and substation footprints aligned to one source of truth.

  • EPOD & hall zoning
  • RMU & substation footprint
  • UPS / battery rooms
  • Generator yard & fuel
VALIDATIONPROFILE-AWARE

Validation & RFI tracking

Rule families across electrical, cooling, layout, safety, and vendor data run continuously. Uncertainty is tracked as RFIs until vendor or owner data closes the gap.

  • Electrical & cooling rules
  • Layout & safety constraints
  • Vendor / RFI ledger
  • Uncertain data flagging
EXPORTPROFILE-AWARE

Export-ready engineering evidence

Site totals, hall panels, Sankey, SLD, and Excel exports consume the same engine output, so what reviewers see matches what the engine computed.

  • Site & hall panels
  • Sankey & SLD views
  • Excel / CSV exports
  • External validation handoff

03 · UNIFIED SOURCE OF TRUTH

Every view consumes the same engine output.

Site totals, hall panels, Sankey, SLD, Excel exports, validators, and configurators all read the same engine model. No spreadsheet drift, no manual reconciliation between a deck and a one-line.

  • Change a load profile → totals, panels, Sankey, SLD, and exports update together.
  • Validation rules read the same model, so reviewer findings map back to source.
  • Uncertain data is held as RFIs until vendor confirmation closes the gap.
ENGINE → CONSUMERSsingle model

ENGINE

AIDC design engine

Generic, profile-aware, deterministic.

  • Site totalsVIEW
  • Hall panelsVIEW
  • Sankey energy flowDIAGRAM
  • Single-line diagramDIAGRAM
  • Excel / CSV ledgerEXPORT
  • ValidatorsRULE
  • ConfiguratorsUI

04 · RUBIN-ERA INFRASTRUCTURE

Built for the thermal and electrical reality of Rubin-class GPUs.

Rack thermal profiles, MV distribution, and CDU topology are not after-thoughts. The engine assumes high rack density and direct-to-chip liquid cooling as the default basis, with optional capability profiles for future vendor- or Rubin-specific variants.

ParameterEngine basisNote
Rack density basis≥ 120 kWPlan against per-rack thermal envelope, not nameplate averages.
MV distribution22.9 kVMV-to-LV step-down strategies sized to peak AI workload.
Cooling topologyL2L · D2CDirect-to-chip primary; rear-door / immersion options as profile variants.
CDU placementIn-Row / End-of-RowDecouples primary and secondary loops at the right capacity step.
PUE basisProfile-awareStated as a basis assumption, not a marketing number.
Fabric scale1.6T-classInfiniBand / Ethernet topology and NVLink switch planning.

05 · VALIDATION WORKFLOW

Rules run continuously. Uncertainty is tracked, not hidden.

Rule families across electrical, layout, cooling, safety, and data fire whenever the model changes. Items that depend on vendor or owner input become RFIs until confirmed, so the design's evidentiary basis is always traceable.

Electrical

  • RMU rating
  • 22.9 kV topology
  • Transformer capacity
  • Generator step-up

Layout

  • Rack width
  • EPOD spacing
  • Service clearance
  • Egress

Cooling

  • Cooling capacity
  • Secondary loop ΔT
  • CDU capacity step
  • Plant interface

Safety

  • Battery constraints
  • Fire compartment
  • Bund / spill
  • Smoke control

Data

  • Vendor-required
  • RFI-required uncertain data
  • Owner-provided basis
  • Profile assumptions
RFI LEDGER Items requiring vendor / owner data are routed to the RFI ledger and re-validated automatically when answered.

06 · EXTERNAL EVIDENCE / EXPORT LEDGER

External validation as an evidence workflow, not a claim of integration.

The engine produces structured outputs intended to feed external validators. Where an integration is not actually implemented, we surface it as an evidence workflow — a deliberate handoff with documented inputs and assumptions — rather than a finished tie-in.

External toolDomainStatus
pandapowerPower flow / network analysisEVIDENCE WORKFLOW
SLD workflowsSingle-line diagram reviewEVIDENCE WORKFLOW
EnergyPlus / OpenStudioBuilding energy modelingEVIDENCE WORKFLOW
OpenFOAMCFD / airflow modelingEVIDENCE WORKFLOW
EcodialLV electrical sizingEVIDENCE WORKFLOW
SIMARISElectrical planningEVIDENCE WORKFLOW
DOCCoordination / discriminationEVIDENCE WORKFLOW
xSpiderSelectivity analysisEVIDENCE WORKFLOW

07 · ENTER THE PLATFORM

Design Rubin-era infrastructure with engineering evidence, not slides.

Architects, engineers, developers, and owners — plan power, cooling, and layout with one source of truth, then export the evidence reviewers actually accept.

Enter PlatformReview Engineering Workflow

Private demo · access requires passphrase. Contact hello@aidc-ai.io