EngiMetric
Medium-voltage switchgear, transformer and cable trays in an electrical plant room

EngiQuest / Mission-critical electrical and thermal systems

AI Data Center: Full Engineering Workbench

Carry an AI workload from rack power through electrical demand, redundancy, cooling, PUE, cost and commissioning to a defended preliminary design package.

  • Advanced
  • 14 stages
  • 234 min
  • Mission-critical electrical and thermal systems
Development preview All EngiQuest journeys

A data-center design is an energy-balance problem before it is a building. The workbench keeps electrical capacity, cooling duty and cost tied to the same rack schedule.

Development preview. The calculations on this journey run on existing EngiMetric engines, but the journey text and scope have not completed independent engineering review. The route is excluded from search while it remains a preview.
Professional review required.Engines and workbench are implemented and tested. Defaults, sources and engineering assumptions still require independent review before publication or use on a real project. Current verification state: Needs engineering review.

The engineering outcome

Mission-critical electrical and thermal systems

A fourteen-stage workbench for a 4 MW teaching campus. You own the rack mix, load fractions, topology, cooling approach, tariffs and cost rates; the model recomputes the 8,760-hour energy balance, equipment capacity, failure scenarios, costs and report from one saved state.

Who it is for
Electrical, mechanical and mission-critical engineers, project owners, developers and technical decision-makers who need a traceable first-pass data-center design and an honest list of what remains unverified.
Stages
14
Estimated time
234 min
Level
Advanced

What you will be able to do

Learning objectives
  • Write a design basis that distinguishes owner requirements, assumptions, missing information and verified data.
  • Build a rack-level IT load model without confusing nameplate capacity with operating demand.
  • Carry concurrent IT load through UPS, transformer, generator and cooling capacity with explicit loading targets and topology rules.
  • Separate capacity redundancy, path redundancy, failure tolerance and maintainability instead of inferring a Tier rating.
  • Calculate cooling duty, air/liquid flow and plant power from a stated thermal boundary.
  • Build a genuine 8,760-hour energy model and a PUE statement with an explicit boundary.
  • Compare utilisation, cooling, redundancy and tariff scenarios using complete model re-runs.
  • Export a preliminary report that preserves calculated, assumed, missing and unverified values.

Journey map

Ordered stages
  1. Write the project brief and design basis Fix the workload type, capacity target, availability objective, site conditions, utility status, water status and cost boundary before a single number is computed. Introduction / 15 min
  2. Model the AI workload and rack capacity Turn a heterogeneous rack schedule into installed nameplate, operating average, operating peak, concurrent peak, annual IT energy and growth headroom. Calculation / 20 min
  3. Test the space and rack-layout envelope Translate rack count and density into a conceptual white-space and support-area allowance, then challenge the site and floor height. Calculation / 12 min
  4. Size the electrical infrastructure and losses Convert concurrent IT peak and critical auxiliary load into UPS input, facility demand, transformer capacity, battery energy and generator capacity using explicit efficiencies and loading targets. Calculation / 22 min
  5. Evaluate redundancy, failure and maintainability Separate capacity redundancy, path redundancy and maintainability, then test failure scenarios without claiming a Tier certification. Comparison / 20 min
  6. Build the cooling load and thermal boundary Convert IT heat and in-boundary electrical losses into air and liquid duties, flow requirements and a cooling reserve. Calculation / 18 min
  7. Select heat rejection and cooling equipment Size cooling units and heat rejection from the calculated duty, and expose the water and ambient assumptions behind the selection. Calculation / 16 min
  8. Build the energy model and PUE statement Calculate annual facility energy from a genuine 8,760-hour model and state PUE with a defined boundary. Calculation / 20 min
  9. Compare utilisation, cooling and redundancy scenarios Re-run the complete model for each option instead of scaling the base case, then compare PUE, peak demand, cost and feasibility. Comparison / 15 min
  10. Build the capital and lifecycle cost case Build a transparent CAPEX and OPEX estimate from user rates, then report cost per IT kW, NPV of costs and cost per delivered IT kWh. Calculation / 18 min
  11. Assemble schedules, diagrams and the assumptions register Turn the same model into a coherent preliminary package: capacity summaries, equipment schedules, schematics and an explicit exclusions register. Implementation / 14 min
  12. Test site, grid, water and implementation readiness Convert the model into an implementation checklist of utility, site, water, permitting and long-lead dependencies. Implementation / 12 min
  13. Plan commissioning, testing and operations Define the tests and operational evidence needed to prove that the installed facility matches the design, without pretending the tests have been performed. Commissioning / 12 min
  14. Defend the preliminary design and export the report Integrate the stages into a defended preliminary design, state bottlenecks and missing information, and export the model-backed report. Capstone / 20 min

Before you start

Prerequisites
  • Comfort with power and energy units: kW, kVA, kWh, power factor and load factor.
  • Basic electrical distribution and UPS/generator concepts.
  • Willingness to read failed checks and missing-input warnings instead of treating a summary as approval.

Defend the preliminary AI data-center design

Capstone

Produce an integrated preliminary design review for the campus. Use the saved project state, the converged model and the failed-check register. The defence must show the chain from AI workload to rack power, electrical demand, cooling, energy, cost and implementation readiness, and it must be explicit about what remains unverified.

Deliverable
A written design review of 1,200-2,000 words plus the exported Markdown or JSON report and a one-page bottleneck and missing-study register.
Evidence model
Rubric against produced artifacts, not completion ticks

Requirements

  • State the project type, IT target, rack mix and availability objective.
  • Present nameplate, average, peak and concurrent IT demand without conflating them.
  • Explain the electrical topology, losses, UPS/battery autonomy and generator strategy.
  • Separate capacity redundancy, path redundancy and maintainability; do not claim a Tier rating.
  • State the cooling boundary, air/liquid split and heat-rejection assumptions.
  • Report annual energy and PUE with the measurement boundary and modelled/measured distinction.
  • Present CAPEX, OPEX and cost per delivered IT kWh with currency and price basis.
  • Identify the limiting system, margin, single points of failure and missing studies.
  • State which decisions require owner, utility or professional engineering review.

Rubric

Each criterion is scored from evidence, and weights sum to 100.
Weight Criterion Evidence
10% Design basis and requirement traceability Owner targets, assumptions, missing data and verified data are clearly separated.
12% IT and rack capacity argument Nameplate, average, peak and concurrent values are correctly distinguished and justified.
15% Electrical demand and capacity UPS, transformer, battery and generator sizing follows from the load with explicit losses and targets.
12% Redundancy and failure evaluation Capacity, path and maintainability are separated; availability is not presented as certification.
12% Cooling load and heat rejection Duty, air/liquid split, plant power and water boundary are consistent with the model.
12% Energy and PUE statement Annual energy and PUE are reported with boundary, method and modelled/measured status.
10% Cost and sensitivity CAPEX, OPEX and cost per kWh use a stated basis and identify dominant assumptions.
12% Bottlenecks, missing information and review The limiting system, failed checks, missing studies and required reviews are named honestly.
5% Clarity and professional presentation The argument is structured, unit-consistent and free of unsupported claims.
What this is not. This is an educational preliminary design review, not a stamped design, Tier certification, utility approval, permit, procurement package or investment case.

Replace illustrative climate, grid, water and cost data with evidenced site information, then repeat the model and the review.

Journey limits and sources

Provenance

Limitations

  • Status is preview. The journey is noindex and has not been independently engineering-reviewed.
  • Climate, grid, water, tariff and cost defaults are illustrative planning values, not site evidence.
  • The model is a screening tool. It does not perform short-circuit, protection-coordination, arc-flash, structural, fire, acoustic, CFD, water-treatment or utility interconnection studies.
  • PUE is modelled from a synthetic 8,760-hour profile, not measured from meters.
  • Availability is a binomial capacity-block estimate, not a Tier certification or uptime guarantee.
  • Costs are owner-side estimates with no revenue, financing, tax or market-price basis.
  • No equipment vendor performance data is bundled; user-supplied COP and availability values must be replaced with evidence.

Review status

  • Needs engineering review
  • Professional review is required before publication or use as a design basis.