Data Centers: executive and technical guide to design, planning, and construction

Designing a data center means managing consequences. The final product is not the building, but the continuous, secure delivery of computing capacity, connectivity, and storage. This guide covers the full chain of decisions — from the business case and risk analysis to thermal and electrical engineering, telecommunications, protection, construction, commissioning, and operation.

The sequence is always the same: the business case defines criticality; criticality defines requirements; requirements define the design; the design is proven through testing; and operation preserves the performance achieved. Skipping a step transfers uncertainty to the construction phase, where changes cost more and jeopardize schedule.

About this material: the architectures presented are reference examples and do not replace an executive design, technical report, professional liability, or consultation with the current editions of applicable standards. Redundancy is not automatically synonymous with availability: the outcome depends on physical independence, selectivity, automation, procedures, integrated testing, and operational competence.

Executive summary

Each part of the guide answers a specific decision. Use the table as a reading map: managers should start with the first four rows; engineering should move through the technical disciplines; operations should focus on commissioning, documentation, and indicators; procurement should use the matrices and checklists to structure the RFP, technical equalization, and acceptance.

Part	Content	Main decision
Fundamentals	Objective, typologies, and life cycle	Build, contract colocation, edge, or hybrid?
Requirements and risks	BIA, criticality, threat, and tolerance	Which interruption is acceptable and at what cost?
Availability	Classes, topologies, maintenance, and failures	Which architecture meets the service?
Planning	Capacity, area, density, and expansion	How much, when, and where to build?
Thermal	Airflow, DX, chilled water, and liquid cooling	How to remove heat with stability and efficiency?
Electrical	Grid, GMG, UPS, batteries, ATS, PDU, and protection	How to deliver safe power up to the IT load?
Telecom	Routes, MMR, cabling, and topologies	How to eliminate hidden dependencies?
Protection and control	Fire, access, CCTV, BMS, and DCIM	How to detect, contain, and respond?
Construction and delivery	Materials, stages, QA/QC, and commissioning	How to prove the design works?
Operation and future	SLA, maintenance, PUE, AI, and high density	How to sustain reliability over time?

What is a data center

A data center is the coordinated set of spaces, utilities, controls, and processes that sustains technology and telecommunications equipment. Power, cooling, and networks are service chains; each link needs capacity, protection, monitoring, and a maintenance path.

Layer	Examples	Typical failure
Business	Services, customers, legislation, revenue	Poorly defined criticality
IT and telecom	Compute, storage, network, platforms	Demand exceeding capacity
Facilities	Power, cooling, fire, security	Single point of failure
Operations	People, processes, parts, and suppliers	Human error or delayed response

Principle: critical infrastructure must be designed around the service that needs to survive, not just the equipment that will be installed.

The full life cycle runs from strategy and requirements to the preliminary study, conceptual design, basic and executive design, contracting, construction, testing, operation, optimization, and finally expansion or decommissioning.

Types and deployment models

Model	Advantage	Attention point
Enterprise	Control and integration with the business	CAPEX, scale, and specialized team
Colocation	Speed, connectivity, and shared infrastructure	Contract, demarcation of responsibilities, and cross-connects
Hyperscale	Scale, standardization, and automation	High demand for power, water, and land
Edge	Low latency and proximity to the user	Distributed sites, maintenance, and physical security
Modular/prefabricated	Schedule and repeatability	Civil interfaces, logistics, and expansion
Hybrid	Flexibility across sites and clouds	End-to-end operational architecture and observability

The build-versus-contract comparison must incorporate total cost, schedule, regulatory risk, available power, connectivity, expansion capacity, useful life, cost of capital, staffing, obsolescence, and contractual flexibility. A seemingly cheap in-house room may require disproportionate electrical and thermal retrofitting; colocation can reduce lead time, but requires an SLA, audit, and exit plan.

• Map loads that cannot leave the site and those that can be distributed.
• Model at least three growth scenarios and one stress scenario.
• Include the cost of downtime, not just CAPEX and monthly fees.
• Define who is responsible for each segment: utility, data center, rack, equipment, and application.

Requirements in balance

A viable project balances three groups: business defines what needs to continue; engineering translates that into capacity, topology, and performance; finance sets the investment and operating envelope. This balance must be documented in a Basis of Design (BoD), with assumptions, exclusions, and acceptance criteria.

Group	Essential questions
Business	Which services? Who is affected? What window? What contractual or legal obligation?
Technical	Initial and future load? Density? Autonomy? Routes? Maintainability?
Financial	CAPEX, OPEX, contingency, cost of capital, energy, maintenance, and renewal?
Operational	Staff, 24x7 coverage, spare parts, suppliers, procedures, and training?

The Owner's Project Requirements (OPR) document consolidates this decision and must contain:

• Project objectives and scope.
• Load profile and growth horizon.
• Availability, security, and efficiency criteria.
• Environmental conditions and operating limits.
• Metering, alarms, integration, and data retention.
• Testing, documentation, training, and warranty.

BIA and business risk

The Business Impact Analysis identifies critical processes, dependencies, maximum tolerable downtime, and progressive impacts. The RTO guides the recovery time; the RPO guides the maximum acceptable data loss. They do not replace the facilities analysis, but they prevent infrastructure from being sized based on perception.

Impact	Example evidence	Treatment
Financial	Lost revenue, penalties, rework	Quantify per hour and per event
Operational	Backlog buildup, productivity loss	Model degradation and recovery
Legal/regulatory	Deadlines, data retention, security	Validate with legal and compliance
Reputation	Churn, press, trust	Use scenarios and indicators
Life/safety	Medical services, emergency, control	Minimum tolerance and formal response

Probability and consequence must be supported by history, asset condition, exposure, and existing controls. Residual risk is what remains after mitigation — accepting it is a decision by the risk owner, not an omission in the design.

Example: if a power failure can interrupt R$ 300 thousand per hour and the likely recovery time is 4 hours, the direct reference impact is R$ 1.2 million — before penalties, image, and operational recovery.

Internal and external risks

Domain	Threats	Verification
Location	Flood, landslide, external fire, aircraft crash	Maps, history, exposure radius, and inspection
Utilities	Weak power grid, fuel, water, telecom	Feasibility letters and physical routes
Building	Structural load, water above critical room, vibration	Survey and testing
Fire	Combustible load, compartmentalization, detection	Legal design and cause/spread analysis
People	Improper access, error, lack of training	Segregation, procedure, and training
Supply chain	Lead time, exclusive part, single supplier	Critical list and strategic stock
Cyber-physical	Exposed BMS/DCIM, weak credentials	Segmentation, hardening, backup, and logging

• Risk register with owner, deadline, and evidence.
• FMEA for failure modes, effects, and controls.
• HAZOP or What-if for process deviations.
• Fault Tree for combinations leading to the top event.
• Bow-tie to visualize prevention, event, and mitigation.

Availability without oversimplification

Availability is the proportion of time the service is able to perform its function. For a repairable component, a common approximation is A = MTBF / (MTBF + MTTR). In a data center, however, common dependencies, human failures, maintenance, controls, and operational response make systemic behavior far more complex than the formula suggests.

Concept	Practical meaning
N	Capacity needed to support the design load
N+1	One additional module beyond what is needed
2N	Two complete and independent systems
2(N+1)	Two complete systems, each with an additional reserve
Concurrent maintenance	Remove a component or path in a planned manner without stopping the load
Fault tolerance	A failure does not cause immediate interruption of the critical load

• Two devices fed by the same panel do not form independent paths.
• Logical redundancy without physical separation can fail from a single event.
• A single-cord load requires an STS or a specific architecture; dual-cord loads must have consistent A/B distribution.
• Rated battery autonomy varies with load, temperature, age, and discharge regime.

Classes and reference standards

The ISO/IEC 22237 series classifies infrastructure by availability, security, and efficiency across the life cycle. The Uptime Institute uses Tier criteria oriented to topology and operational performance. Levels and classes from different reference standards should not be treated as automatically equivalent: the contractual requirement must indicate the standard, edition, scope, and method of proof.

Intent	Typical architecture	Validation question
Basic capacity	Single path and sized components	Does a planned maintenance require shutdown?
Redundant components	N+1 in parts of the chain	Is there still a single path?
Maintainability	Multiple paths and isolation	Can each component be maintained with an active load?
Fault tolerance	Segregation and automatic response	Does an isolated failure interrupt the load?

Caution: do not advertise certification, class, or Tier based solely on a diagram. Certification depends on the program, scope, documentation, construction, and/or operation as evaluated by the relevant authority.

Dual-path electrical architecture

The A/B principle creates two power paths up to the critical load. To be effective, it must consider source, generation, transformation, switching, UPS, distribution, cables, PDU, and equipment connection. Shared components must be identified and justified — not discovered during a failure.

• N capacity during maintenance and under the defined failure condition.
• Selectivity and protection coordination across all operating modes.
• Testable interlocks and transfer logic.
• Physical separation against fire, water, and intervention.
• Metering points from the utility to the rack.
• Safe bypass plan and return to normal condition.

From IT inventory to design power

Sizing begins with an inventory of equipment, measured or declared power, coincidence factor, usage profile, growth, and reserves. Nameplate power is not average consumption — but the design also cannot ignore peaks, transients, simultaneous boot, battery recharge, and degradation.

Stage	Reference calculation
Initial IT load	Sum per equipment × utilization and coincidence factor
Growth	Annualized scenarios or by installation waves
Rack capacity	Electrical, thermal, physical, and floor limits — the lowest applies
IT thermal load	Approximately equal to the electrical power dissipated in steady state
Infrastructure	IT + cooling + losses + lighting + auxiliaries
Reserve	Explicit margin, without duplicating conservative factors

Example: 100 racks × 8 kW = 800 kW of IT load. With a design PUE of 1.45, the average total reference power would be 1.16 MW. The electrical system must still account for failure modes, maintenance, startups, and rated capacity.

Density and rack strategy

Density can be expressed in kW/rack, kW/m² of room, or per computing unit. The average hides hotspots: an 8 kW/rack room can contain 30 kW islands. The design must separate zones, define per-rack limits, and reserve evolution paths for liquid cooling.

Indicative range	Typical application	Thermal strategy
Up to 5 kW/rack	Conventional IT or low occupancy	Organized aisles and bypass control
5 to 15 kW/rack	Virtualization and storage	Containment and units with modulation
15 to 30 kW/rack	Dense compute and moderate accelerators	In-row/rear-door or highly controlled air
Above 30 kW/rack	HPC and high-density AI	Evaluate direct-to-chip, CDU, and technology water

The ranges are indicative. The decision depends on the required flow rate, equipment ΔT, available pressure, layout, thermal tolerance, and manufacturer specification.

• Width and depth compatible with equipment and cabling.
• Aisles for circulation, maintenance, and component removal.
• Cable management without blocking exhaust.
• Blanking panels and sealing of openings.
• Grounding, bonding, and labeling.

Site selection

Criterion	Due diligence questions
Power	Firm capacity? Voltage? External redundancy? Connection lead time? Quality?
Telecom	How many carriers? Truly diverse routes? Opposite entrances?
Environmental	Climate, water, noise, emissions, permits, flooding, and neighborhood
Civil	Soil, load, vibration, height, access, and equipment logistics
Operations	Labor, fuel, parts, security, and response time
Expansion	Area, utilities, phases, interferences, and continuity during construction

In an existing building (brownfield), perform an as-built survey, scanning when applicable, load testing, panel and cable inspection, structural analysis, and verification of water above or beside critical areas. The invisible constraint often defines the design: shaft, exhaust route, ceiling height, floor load, fire protection, or rigging access.

Thermal fundamentals

Nearly all electrical energy consumed by IT converts into heat. The thermal system's task is to capture that heat at the source, transport it, and reject it to the environment, keeping the equipment's inlet within the defined range. Average room temperature does not reveal recirculation, bypass, or hotspots.

Phenomenon	Effect	Control
Recirculation	Hot air returns to the rack inlet	Containment, sealing, and proper pressure
Bypass	Cold air does not pass through the IT	Flow adjustment, blanking panels, and layout
Mixing	Reduces ΔT and efficiency	Physical segregation of flows
Low flow	Inlet temperature rises	Capacity, fans, and clear paths
Excessive flow	Wasted energy and undue pressure	Demand-based modulation

Environmental parameters

Environmental classes must be defined according to the equipment installed. As a widely used reference, ASHRAE TC 9.9 indicates a recommended range of 18 °C to 27 °C for classes A1 through A4; allowable limits vary by class and should not be confused with a permanent operating target. Humidity must be managed by dew point, condensation risk, electrostatics, and corrosion.

Parameter	Why measure it	Best practice
Inlet temperature	The air actually received by the IT	Sensors per rack and height, not just on the wall
Humidity and dew point	Condensation, ESD, and corrosion	Trend and consistent alarms
Differential pressure	Confirms flow direction	Measure between relevant zones
Particulates and corrosion	Electronic reliability	Filtration, sealing, and coupons when necessary
Water leakage	Early detection	Sensors per zone and periodic testing

Tape equipment, batteries, and power components may have different limits. The manufacturer's specification and warranty conditions prevail.

Cooling technologies

Solution	When it makes sense	Points of attention
Direct expansion (DX)	Small, modular, or distributed sites	Refrigerant, distance, redundancy, and partial efficiency
Chilled water	Medium and large scale, centralization	Pumps, chiller, tower/dry cooler, water, and controls
In-row	Hotspots and proximity to the load	Hydraulic or refrigerant distribution in the room
Rear-door	Retrofit and high density	Weight, hoses, condensation, and maintenance
Direct-to-chip	High-density CPU and GPU	CDU, water quality, pressure, detection, and interfaces
Immersion	Specific loads and very high density	Fluid, maintenance, compatibility, and operation

Compare total cost, local climate, water availability, density, modularity, part-load efficiency, concurrent maintenance, area, noise, refrigerants, leak risk, and team capability. The best COP of an isolated unit does not guarantee the best system performance.

Thermal sizing and controls

Capacity must be verified for outdoor design conditions, sensible load, losses, redundancy, and degraded modes. Adding up rated capacities without confirming curves, water and air temperature, altitude, and flow rates creates false confidence.

Verification	Acceptance question
Thermal balance	Have all heat sources and scenarios been included?
Psychrometrics	Is there control against condensation and excessive humidification?
Hydraulics	Are pressure drop, valve authority, and balancing proven?
Controls	Have the sequence of operation and setpoints been reviewed?
Failure and maintenance	Does the temperature remain acceptable under the defined event?
Dynamic testing	Have load steps and transfers been simulated?

Metric: installed capacity is not enough. Track available, used, and committed capacity per zone, always considering the limiting component.

Critical electrical chain

The typical chain includes the utility, medium voltage, transformation, panels, generation, transfer, UPS, distribution, and rack. Each stage must be analyzed under normal condition, maintenance, failure, emergency, and return to normal.

Component	Function	Critical risk
Entry and MV	Receive, switch, and protect	Arc flash, coordination, and external outage
Transformer	Adjust voltage and isolate	Capacity, temperature, harmonics, and fire
GMG	Supply prolonged outage	Startup, fuel, emissions, and heat rejection
ATS/STS	Transfer sources and paths	Logic, synchronism, and switching failure
UPS	Condition and sustain transition	Bypass, battery, overload, and common-mode failure
PDU/RPP/busway	Distribute and meter	Selectivity, expansion, and connection error

Generator sets and fuel

The generator must meet steady-state load and transient steps, including motors, UPS, recharge, cooling, and auxiliaries. The analysis must consider startup sequence, voltage and frequency dip, short-circuit capacity, altitude, temperature, fuel, and emissions.

Topic	Criterion
Autonomy	Defined by risk, refueling contract, and crisis scenario
Tanks	Usable capacity, containment, detection, transfer, and fuel quality
Startup	Redundant batteries and chargers, pre-heating, and testing
Paralleling	Control, synchronism, load sharing, and protection
Maintenance	Access, parts, load bank, and planned downtime
Exhaust and noise	Back pressure, temperature, dispersion, acoustics, and permits

Meaningful test: a weekly no-load startup does not prove the chain. The program must include tests under load, source failure, transfer, stability, operational autonomy, and return, with calibrated instruments and clear criteria.

UPS, batteries, and energy storage

Technology	Advantage	Attention point
Double conversion	Load isolation and broad application	Losses, bypass, and partial efficiency
Modular	Scalability and per-module maintenance	Common control and bus failures
VRLA	Maturity and initial cost	Temperature, life, weight, and monitoring
Lithium-ion	Service life, cycles, and smaller area/weight	BMS, thermal propagation, and fire strategy
Flywheel	High power and many cycles	Short autonomy and integration with generation

To size autonomy, consider the actual discharge curve, constant power, minimum voltage, temperature, aging, manufacturing tolerance, and growth margin. Required autonomy should not be chosen by habit: it covers the interval between the loss of source and generation stabilization, or a controlled shutdown strategy.

Safety: battery rooms and systems require specific evaluation of ventilation, detection, containment, access, PPE, and emergency response according to the technology and applicable regulations.

Protection, selectivity, and power quality

The electrical study must include load flow, short circuit, coordination and selectivity, voltage drop, grounding, surge protection, harmonics, and, where applicable, arc-flash incident energy. The studies need to reflect all operating modes, including on generator.

Problem	Symptom	Design treatment
Harmonics	Heating and distortion	Metering, filters, transformers, and suitable conductors
Low power factor	Occupied capacity and penalties	Coordinated correction and resonance analysis
Transients	Reset or damage	Cascaded surge protection and bonding
Lack of selectivity	Trips a larger area than the fault	Curves, settings, and protection testing
Imbalance	Heating and loss of capacity	Phase distribution and monitoring

Plan meters at the utility, generation, UPS, cooling, distribution, and rack. Without granularity, PUE cannot be calculated with confidence, nor can loss, imbalance, or idle capacity be located.

End-to-end connectivity

Having two carrier contracts does not guarantee diversity. The fibers may share a pole, duct, box, building entrance, or room. The analysis must map physical routes, points of presence, meet-me rooms, cross-connects, active equipment, and power feed.

• Separate and documented physical entrances.
• Segregated meet-me rooms when criticality requires it.
• Internal A/B routes down to rooms and racks.
• Sized, grounded, and protected trays and ducts.
• Permanent labeling and as-built documentation.
• Optical and copper testing with acceptance reports.

Cabling and network architecture

Layer	Options	Criteria
Campus and carrier	Single-mode fiber	Distance, route, splicing, availability, and SLA
Internal backbone	OS2, OM4, or OM5 depending on application	Speed, distance, and evolution
Horizontal	Copper or fiber	Length, density, PoE, EMI, and management
Core/spine	High capacity and redundancy	Domain failure, convergence, and automation
Leaf/ToR/EoR	Server connection	Cabling, ports, oversubscription, and maintenance
OOB	Out-of-band network	Independence, security, and access during failures

Size fibers, ports, trays, and patch panels with explicit growth. The data rate depends on the standard, transceiver, distance, optical loss, and connectors. The design must also separate production, management, storage, security, and building automation networks.

Fire protection

The strategy combines prevention, early detection, compartmentalization, material control, suppression, and response. No single technology eliminates the risk. The design must be approved by the relevant authorities and coordinated with architecture, electrical, cooling, and operations.

Layer	Examples	Objective
Prevention	Housekeeping, thermography, maintenance, materials	Reduce probability
Detection	Spot, aspirating, thermal, and alarms	Identify early and locate
Passive	Compartmentalization, sealing, and doors	Limit spread
Suppression	Sprinkler/pre-action, water mist, clean agent	Control or extinguish
Response	Defined EPO, brigade, fire department, continuity	Protect people and service

Sealing: when agent effectiveness depends on retention within the enclosure, the design must provide integrity, sealing, pressure relief, and a leak-tightness test per applicable criteria.

Physical security and asset protection

Zone	Typical control
Perimeter	Barriers, lighting, detection, and surveillance
Reception	Identification, authorization, logging, and escort
Technical areas	Access profiles, two-factor authentication, and interlocking
IT room	Least privilege, logs, and customer segregation
Racks and cages	Lock, seal, sensor, and inventory
Media and freight	Chain of custody, inspection, and secure disposal

For CCTV and access control, define coverage, resolution, retention, time synchronism, redundancy, privacy, and investigation. Access events must correlate person, credential, door, camera, and work order. Anti-passback and mantraps need contingency and life-safety procedures.

Controllers, cameras, BMS, and DCIM are digital assets: they require segmented networks, individual accounts, MFA where supported, hardening, patch management, tested backups, and logs sent to a protected repository.

BMS, EPMS, and DCIM

System	Focus	Examples
BMS	Building and environmental automation	Cooling, leaks, pressure, doors
EPMS	Power and power quality	Meters, breakers, UPS, GMG, trends
DCIM	Data center capacity and assets	Rack, port, power, space, workflow
ITSM/NOC	Incidents, changes, and service	Tickets, escalation, SLA, and knowledge

• Each alarm needs priority, delay, hysteresis, and procedure.
• Consequential alarms should be grouped to avoid alarm storms.
• Time synchronism is mandatory for causal analysis.
• The operator needs to know what to do, not just what happened.
• Periodic tests must prove sensor, communication, logic, and notification.

Useful data: historical trending turns monitoring into management — anticipating efficiency loss, battery degradation, thermal saturation, and load imbalance.

Construction models

Model	Characteristic	Risk to manage
Design-bid-build	Design completed before construction contracting	Interfaces and late changes
Design-build	Integrated responsibility	Requirements governance and verification independence
EPC/turnkey	Delivery concentrated in a single contractor	Scope, performance, and cost transparency
Construction management	Packages coordinated by the manager	Integration and responsibility across contracts
Modular/prefabricated	Off-site manufacturing	FAT, transport, interfaces, and assembly

The design must anticipate construction sequencing, lifting, equipment entry, storage, temporary areas, energization, cleaning, asset protection, and future replacements. In an active environment, method of procedure, window, rollback, and isolation become part of the design.

Materials, execution, and QA/QC

Materials must meet mechanical, electrical, thermal, fire, corrosion, and maintenance performance. "Equivalent" substitutions can only be accepted after formal comparison of specification, interfaces, certifications, pressure drop, efficiency, curves, and warranty.

Discipline	Inspection points
Civil and architecture	Levels, load, waterproofing, sealing, finish, and cleaning
Electrical	Torque, labeling, insulation resistance, grounding, protection, and parameterization
Mechanical	Cleaning, flushing, leak-tightness, insulation, balancing, and drainage
Telecom	Bend radius, segregation, certification, labeling, and as-built
Automation	Points, scales, alarms, fail-safe, integration, and history
Fire	Zones, detectors, interfaces, discharge/flow, and legal documentation

Use Inspection and Test Plans with hold points, witness points, criteria, evidence, and responsible parties. Do not leave pending items unclassified: the punch list must record criticality, owner, deadline, acceptance blocking, and closure evidence.

Tiered commissioning

Commissioning is a performance assurance process that begins with the requirements, not a test done on the last day. It verifies documentation, installation, startup, controls, capacity, and interaction between systems.

Level	Example activity
L0 — requirements and design	Review of OPR, BoD, sequences, and testability
L1 — factory	FAT, certificates, inspection, and configuration
L2 — installation	Physical inspection, torque, cables, piping, and labeling
L3 — startup	Individual startup and functional testing
L4 — system	Capacity, alarms, redundancy, and operating modes
L5 — integrated	Chained failures, utility loss, black building, and recovery

An integrated system test (IST) must use an approved script, pre-conditions, calibrated instruments, risk analysis, abort criteria, defined roles, communication, and synchronized data collection. The goal is to demonstrate behavior and recovery, not trigger an uncontrolled failure.

Documentation and handover

Deliverable	Minimum content
As-built	Drawings, diagrams, routes, lists, and final revisions
O&M	Manuals, routines, limits, parts, and contacts
Studies	Short circuit, selectivity, thermal, hydraulic, and risk
Configurations	Setpoints, firmware, backups, and communication matrix
Tests	FAT, SAT, functional tests, IST, and open items
Training	Content, attendance, evaluation, and simulations
Assets	Tag, model, serial, warranty, criticality, and spare parts

Golden rule: do not accept a system that merely "turns on." Accept a system whose capacity, protection, control, alarms, failures, and recovery are documented and demonstrated.

BIM models, the asset base, and DCIM data only add value if there is update governance. Define the official source, responsible party, frequency, integration, and audit.

Critical operation

Design reliability deteriorates when changes, maintenance, and knowledge are not controlled. Mature operation combines governance, procedures, training, communication, and continuous improvement.

Process	Essential control
Change	Risk assessment, approval, MOP, rollback, and post-validation
Maintenance	Plan by criticality, window, spare parts, and evidence
Incident	Command, communication, escalation, logging, and RCA
Capacity	Current, committed, reserved, and forecast per domain
Supplier	SLA, access, competence, parts, and performance
Training	Qualification, simulation, and internal recertification

The SOP describes normal operation; the MOP, a planned intervention; the EOP, the response to an abnormal event. All must have prerequisites, roles, numbered steps, stopping points, validation, and version control.

Indicators and efficiency

PUE is the ratio between the total energy of the data center and the energy of the IT. It is useful when boundaries, period, metering, and conditions are consistent. It does not measure availability, carbon, water, IT productivity, or a single server's isolated efficiency.

Indicator	Use	Caution
PUE	Infrastructure energy efficiency	Compare with equivalent boundaries and periods
WUE	Water use associated with operation	Define methodology and water context
CUE	Emissions associated with energy	Depends on the emission factor
Availability	Time able to provide service	Define exclusions and scope
MTTR	Recovery speed	Do not hide recurrence
Stranded capacity	Idle resource due to another limit	Analyze power, thermal, space, and network

• Establish a baseline and metering quality.
• Address bypass, recirculation, and auxiliary loads before raising setpoints.
• Use sequencing and variable speed.
• Test changes in waves, with limits and rollback.
• Convert savings into avoided cost and released capacity.

AI, HPC, and liquid cooling

AI loads concentrate power, network, and heat into a few racks. The challenge stops being just total capacity and comes to include transients, power quality, physical distribution, flow rate, return temperature, CDU redundancy, and maintenance of liquid circuits close to the electronics.

Topic	Design questions
Power	What peak, ramp, and diversity per cluster?
Network	What topology, latency, cabling, and space for optics?
Liquid	What water class, temperature, pressure, quality, and responsibility?
CDU	Redundancy, heat exchange, pumps, control, and bypass?
Risk	Detection, containment, dripless connectors, and response?
Operation	How to drain, purge, maintain, and expand without stopping?

Even if the first phase is air cooled, reserve routes, technical area, structural load, hydraulic points, electrical capacity, and instrumentation for high-density pods. Improvised retrofits tend to create dependencies and downtime.

Deployment roadmap

Phase	Verifiable outputs
0. Strategy	BIA, build/buy options, site shortlist, and business case
1. Preliminary study	Load, area, risks, concept, schedule, and budget
2. Conceptual design	Architectures, diagrams, criteria, and estimate
3. Basic and executive design	Calculations, details, specifications, BoQ, and test plans
4. Contracting	RFP, equalization, responsibilities, and milestones
5. Construction	QA/QC, safety, changes, and evidence
6. Commissioning	Tests by level, IST, training, and punch list
7. Operation	Handover, baseline, SLA, maintenance, and improvement

Each phase must end with approval criteria. The gate does not exist to add bureaucracy, but to prevent critical uncertainties from advancing and turning into expensive changes. Unmet requirements must be corrected, formally accepted as risk, or removed from scope with the impact recorded.

Checklist for RFP and contracting

Block	Items
Scope	Boundaries, interfaces, inclusions, exclusions, and responsibilities
Performance	Capacity, autonomy, efficiency, availability, and environment
Design	Standards, deliverables, review, BIM/CAD, and change control
Supply	Accepted brands, equivalence, FAT, logistics, warranty, and parts
Construction	Plan, safety, QA/QC, ITP, schedule, and active environment
Testing	SAT, functional tests, load bank, IST, and acceptance criteria
Operation	Training, O&M, as-built, support, and SLA

• Compare technical exceptions, not just total price.
• Normalize efficiency, usable capacity, redundancy, and autonomy.
• Evaluate lead time, obsolescence, and local support.
• Require an item-by-item compliance matrix.
• Tie payment to milestones and performance evidence.

Operational readiness checklist

Question	Expected evidence
Does the team know the sequence of operation?	Recorded training and simulation
Do procedures reflect the as-built?	Reviewed and version-controlled SOP, MOP, and EOP
Do alarms have a defined response?	Alarm matrix and playbooks
Are critical parts available?	Inventory, preservation, and replenishment
Have configuration backups been tested?	Proven restoration
Do contracts cover the 24x7 scenario?	SLA, contacts, and escalation
Is there a baseline before occupancy?	Power, thermal, network, and environment recorded

Result: the data center is ready when people, processes, documentation, tools, and suppliers can operate and recover the infrastructure — not merely when construction ends.

Quick formulas and conversions

Topic	Reference relation
Three-phase power	P ≈ √3 × V × I × PF × efficiency
Energy	kWh = kW × hours
Thermal load	1 kW ≈ 3,412 BTU/h
Ton of refrigeration	1 TR ≈ 3.517 kW thermal
PUE	total data center energy ÷ IT energy
Simple availability	MTBF ÷ (MTBF + MTTR)
Compound growth	future capacity = current × (1 + rate)^years

The relations above are simplifications for estimation. Executive design requires curves, correction factors, environmental conditions, losses, harmonics, coincidence, and normative criteria.

Conceptual sizing example: IT load of 600 kW, growth to 900 kW in five years, and modular expansion of 150 kW. The strategy can deploy 4 active modules + 1 redundant in the initial phase, reserving space and primary infrastructure for future modules. The analysis must verify whether the system maintains N during maintenance and failure, and whether the auxiliary loads keep pace with each stage.

Recommended technical references

Always consult the current edition, the contracted scope, and local legal requirements. The references below guide the study but do not replace acquisition, full reading, and application by qualified professionals.

Reference	Application
ISO/IEC 22237	Data centre facilities and infrastructures — concepts, building, power, environment, telecom, security, and operation
ISO/IEC 30134	Key performance indicators for data centers, including PUE
ASHRAE TC 9.9	Thermal Guidelines for Data Processing Environments
Uptime Institute Tier Standard	Topology criteria and certification programs
ANSI/TIA-942	Telecommunications infrastructure for data centers
NFPA 70 / 75 / 76 / 855	Electrical, IT and telecom equipment protection, and energy storage systems, as applicable
ABNT NBR 5410 / 14039 / 5419	LV and MV electrical installations and lightning protection
ABNT NBR 14565	Structured cabling for commercial buildings and data centers
ABNT NBR 17240 and applicable fire standards	Detection and alarm, plus local Fire Department requirements

Brazilian legislation, ABNT standards, Fire Department instructions, and environmental requirements must be confirmed for the location and date of the project.

Conclusion

Designing a data center means managing consequences. The best solution is not the one that accumulates the most redundant equipment, but the one that connects criticality, risk, architecture, execution, testing, and operation in a coherent and demonstrable way.

• Hold a requirements and BIA workshop.
• Consolidate inventory, capacity, and expansion horizon.
• Perform due diligence on the site and utilities.
• Produce a conceptual design with alternatives and TCO.
• Define commissioning criteria before contracting.
• Plan handover and operation from the start.

EnQ Digital: intelligent infrastructure for companies that cannot stop — strategy, engineering, deployment, operation, and evolution oriented toward risk and performance.