Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
Estimating an AI data center’s cost and power needs requires a workload-based model, not a universal “cost per megawatt” multiplier. Start with the servers, storage and networking required for your training or inference workload; convert that IT load into total facility demand with an explicit power usage effectiveness (PUE) assumption; then price the site, utility connection, electrical and cooling systems, building work, hardware and ongoing operations separately.
A defensible first estimate reports average energy, peak demand and engineered capacity as different figures, shows low, base and high cases, and identifies which assumptions still need utility or vendor quotations. The method below gives you an auditable worksheet even when engineering bids are incomplete.
As an Amazon Associate I earn from qualifying purchases.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWhat exactly are you estimating?
Before entering a number, define the boundary. “AI data center cost” can mean an accelerator cluster inside an existing building, a complete greenfield campus, or the total cost of owning and operating a facility. Those are not interchangeable.
Write the boundary at the top of the estimate and keep excluded items visible. A useful scope statement answers these questions:
Popular on Amazon
- 【90 Degree Full Function USB C Adapter】Supports the latest USB4 standard for up to 40Gbps of bandwidth, 8K video support, and up to 240W Fast charging; Support USB PD 3.1 Extended Power Range (EPR) including 140W, 180W and 240W. Your productivity is easily doubled. *Warm Tip: For the best fit, please make sure the USB-C port is fully exposed; thick or bulky cases may affect compatibility.
- 【Reliable Quality】Made from high-quality, scratch-resistant aluminum alloy, delicate metal frame. get excellent wear resistance and heat dissipation. Non-Slip Design for easier insertion and removal.
- 【Small and portable】Ultra thin size, ultra compact design, a petite physique that matches the minimalist design of your modern laptops. Coming with a slim figure that won't block the port next to the one it's plugged into.
- 【Compatibility】Widely work with most usb c, thunderbolt 4/3 devices. Compatibilit for ROG Ally, Steam Deck, Switch, MacBook Pro, Quest 3 VR headset, Surface Book 2, iPhone 18/17/16/15, Samsung Galaxy S23/S22/S21, Note 20/10, Google Pixel, ChromeBook, etc.
- 【Excellence Customer Support】If you experience any issues or cable failure, simply reach out to our dedicated customer service team. We provide a prompt response within 18 hours, providing you with the necessary technical support to resolve any concerns.
- Workload: model training, batch inference, interactive inference, or a mix.
- Performance target: accelerator count, throughput, latency, training time or service-level objective.
- Deployment: on-premises, colocation, cloud access, or edge sites.
- Reliability: the required redundancy, maintenance strategy and uptime target.
- Geography: country, utility territory, climate, land constraints and water availability.
- Cost boundary: IT-only, facility construction, utility upgrades, financing, taxes, operating costs, or all of them.
- Time horizon: initial build, expansion phases and equipment replacement dates.
Keep three outputs separate:
- Average facility energy: the long-run electricity use used to calculate annual bills and emissions.
- Peak facility demand: the highest expected draw that affects utility capacity, demand charges and generation.
- Engineered capacity: the electrical and thermal capacity installed to handle peaks, maintenance and failures.
A facility may have a 10 MW average load, a higher short-duration peak and still require more than 10 MW of installed capacity because of redundancy and expansion headroom. Combining those values produces an estimate that is impossible to audit.
How do you define the workload and sizing assumptions?
Training, inference and mixed workloads create different utilization patterns. Training may run large accelerator fleets for long periods, while interactive inference can be latency-sensitive and uneven. Record the assumptions before selecting equipment.
| Input | What to record | Why it changes the estimate |
|---|---|---|
| Workload type | Training, inference, fine-tuning, batch jobs or mixed | Determines accelerator mix, utilization and network pattern |
| Compute target | Required throughput, latency, training duration or tokens per second | Sets the quantity and class of servers |
| Utilization | Expected average and peak utilization by server class | Controls average energy; it should not be applied blindly to nameplate power |
| Growth | Initial deployment, annual demand growth and expansion dates | Changes the utility connection, building modules and stranded-capacity risk |
| Reliability | Redundancy, maintenance operation and failure tolerance | Adds electrical, cooling and sometimes duplicate IT capacity |
| Location | Utility territory, climate, water constraints and land | Changes tariffs, cooling design, construction and interconnection work |
| Boundary | Included and excluded capital and operating items | Prevents a facility quote from being compared with an IT-only price |
Details checked November 2026.
If a workload can be served from a cloud or colocation provider, model that alternative on the same utilization, geography, service level and time horizon before committing to construction. The comparison may change the build-versus-buy decision.
How do you calculate IT power?
Build an equipment inventory by class, then calculate operating power for each class. The basic calculation is:
IT power (kW) = Σ [quantity of each equipment class × average operating power per unit (kW)]
Use separate columns for average operating power, rated maximum power and the source of each value. Average power is appropriate for annual-energy estimates; rated maximum is needed for electrical and thermal design checks. If a vendor supplies a range, show the range rather than silently choosing the midpoint.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
| Equipment class | Quantity | Average operating power | Rated maximum | Source and notes |
|---|---|---|---|---|
| GPU or other accelerator servers | Number of each server type | Measured or vendor operating figure | Nameplate or design maximum | Record accelerator count and chassis configuration |
| CPU servers | Number by configuration | Operating figure | Maximum | Include control, preprocessing and non-accelerated jobs |
| Storage | Usable and raw capacity, enclosure count | Operating figure | Maximum | Include storage controllers and expansion shelves |
| Network fabric | Switches, routers and optical equipment | Operating figure | Maximum | High-bandwidth training fabrics can materially change the total |
| Management and security | Management nodes and appliances | Operating figure | Maximum | Do not omit “small” infrastructure that runs continuously |
The U.S. Department of Energy’s 2025 data-center energy report models consumption by server category and distinguishes two-, four- and eight-GPU AI servers. That structure is a useful precedent: separate server types instead of applying one multiplier to an entire “AI” label.
Do not multiply nameplate power by a utilization percentage when equipment has significant idle draw. A server that draws substantial power while waiting for work will use more than a simple “nameplate × utilization” formula predicts. Use measured operating power, a vendor workload estimate, or a model with idle and active states:
Average server power = idle power + (active power − idle power) × utilization
Also popular on Amazon
- 5-in-1 Connectivity: Equipped with a 4K HDMI port, a 5 Gbps USB-C data port, two 5 Gbps USB-A ports, and a USB C 100W PD-IN port. Note: The USB C 100W PD-IN port supports only charging and does not support data transfer devices such as headphones or speakers.
- Powerful Pass-Through Charging: Supports up to 85W pass-through charging so you can power up your laptop while you use the hub. Note: Pass-through charging requires a charger (not included). Note: To achieve full power for iPad, we recommend using a 45W wall charger.
- Transfer Files in Seconds: Move files to and from your laptop at speeds of up to 5 Gbps via the USB-C and USB-A data ports. Note: The USB C 5Gbps Data port does not support video output.
- HD Display: Connect to the HDMI port to stream or mirror content to an external monitor in resolutions of up to 4K@30Hz. Note: The USB-C ports do not support video output.
- What You Get: Anker 332 USB-C Hub (5-in-1), welcome guide, our worry-free 18-month warranty, and friendly customer service.
Apply the model separately to accelerator, CPU, storage and network classes. Document whether utilization means the fraction of time busy, the fraction of maximum compute throughput, or another metric; those definitions are not equivalent.
How do you convert IT load into total facility power?
Convert the IT result with an explicit PUE assumption:
Facility power (kW) = IT power (kW) × PUE
PUE, or power usage effectiveness, is the ratio of total facility energy to IT equipment energy. It includes cooling, UPS losses, power distribution, lighting and other overhead; it is not a synonym for cooling efficiency. For annual energy:
Annual facility energy (kWh) = average facility power (kW) × 8,760 hours
For annual electricity cost:
Annual energy cost = annual kWh × delivered electricity tariff
Model demand charges, taxes, standby fees, renewable-contract premiums and other tariff components separately when they apply. A flat energy rate can materially understate a site with a high coincident peak.
For example, an illustrative 1,000 kW average IT load at a PUE of 1.30 produces 1,300 kW of average facility demand and 11,388,000 kWh per year (1,300 × 8,760). Those are arithmetic outputs from the stated assumptions, not a typical AI-facility benchmark. Replace the illustrative PUE and tariff with values from the proposed site.
The International Energy Agency reports that servers account for about 60% of electricity demand in modern data centers on average. Cooling can represent roughly 7% in efficient hyperscale facilities and more than 30% in less-efficient enterprise facilities. The spread is why a single overhead percentage is unsafe.
For context, the IEA estimates global data-center electricity use at about 415 TWh, or 1.5% of global electricity consumption, in 2024 and projects about 945 TWh in its 2030 Base Case. These are global outlook figures, not inputs for a particular site; the agency emphasizes uncertainty around efficiency, adoption and grid constraints.
Recommended Free Tools
Also popular on Amazon
- Sleek 7-in-1 USB-C Hub: Features an HDMI port, two USB-A 3.0 ports, and a USB-C data port, each providing 5Gbps transfer speeds. It also includes a USB-C PD input port for charging up to 100W and dual SD and TF card slots, all in a compact design.
- Flawless 4K@60Hz Video with HDMI: Delivers exceptional clarity and smoothness with its 4K@60Hz HDMI port, making it ideal for high-definition presentations and entertainment. (Note: Only the HDMI port supports video projection; the USB-C port is for data transfer only.)
- Double Up on Efficiency: The two USB-A 3.0 ports and a USB-C port support a fast 5Gbps data rate, significantly boosting your transfer speeds and improving productivity.
- Fast and Reliable 85W Charging: Offers high-capacity, speedy charging for laptops up to 85W, so you spend less time tethered to an outlet and more time being productive.
- What You Get: Anker USB-C Hub (7-in-1), welcome guide, 18-month warranty, and our friendly customer service.
Planning calculators such as Schneider Electric’s TradeOff tools can help explore PUE, capacity, energy and UPS-efficiency choices. If you use a calculator, preserve the input assumptions and version or access date in the worksheet so another person can reproduce the result.
How should you check rack density and cooling feasibility?
Total megawatts do not prove that a building can physically host the equipment. Divide the IT load by a rack-density assumption and check the resulting rack count, floor loading, power distribution and heat-rejection design.
The OECD’s 2026 outlook describes typical AI racks at roughly 100–200 kW, compared with 10–15 kW for standard IT racks, and discusses future designs projected to reach 1 MW per rack. Those figures are context ranges, not a specification for your chosen server platform.
Use the selected rack’s actual power, dimensions, weight, service clearances and cooling method. For each rack group, record:
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →- Average and peak rack power.
- Rack power distribution units (rack PDUs), upstream voltage and phase requirements.
- Air, rear-door heat exchanger, direct-to-chip liquid or other cooling architecture.
- Supply and return temperatures, coolant flow and heat-rejection capacity.
- Floor loading, ceiling height, aisle layout and maintenance access.
- Electrical and cooling redundancy during component failure or maintenance.
A published Vertiv 360AI 500 kW reference design specifies 484 kW of IT load across ten racks: seven at 58 kW and three at 20 kW. It also shows rack PDUs, UPS equipment and cooling as part of the integrated design. Its central lesson is stated plainly: “Design both Power and Cooling together to optimize AI infrastructure.” Treat that document as an example configuration, not a universal design.
If your calculated rack count requires more density than the selected cooling system, the estimate is not merely expensive; it is physically infeasible. Revise the server mix, rack layout, cooling architecture or deployment size before refining costs.
What belongs in the capital-cost estimate?
Use line items instead of a blended dollar-per-megawatt figure. A blended figure hides whether a project is expensive because of land, grid work, cooling, hardware or schedule.
| Capital category | Include | Questions to resolve |
|---|---|---|
| Land and civil work | Site acquisition, grading, roads, drainage, foundations and environmental work | Is land owned, leased or purchased? Are remediation or flood controls required? |
| Building shell and fit-out | Structure, secure rooms, fire protection, loading, offices and commissioning spaces | Is an existing shell reusable, and what modifications are needed for high-density racks? |
| Utility interconnection | Studies, substation, transformers, transmission upgrades, metering and connection fees | What capacity is available, when can it be delivered, and who pays for upgrades? |
| Electrical systems | Switchgear, UPS, batteries, generators, transfer equipment, busway and distribution | What redundancy and maintenance configuration is required? |
| Cooling and heat rejection | Chillers, pumps, cooling distribution, liquid-cooling equipment, heat exchangers and controls | What peak heat load, water supply and climate assumptions apply? |
| IT equipment | Accelerator and CPU servers, storage, network fabric, racks and rack PDUs | What is included in the supplier quote, and what replacement cycle is assumed? |
| Soft costs and risk | Design, engineering, permits, project management, commissioning, contingency, financing, taxes and schedule-related costs | Which items are inside the project boundary and which are owner costs? |
McKinsey’s 2025 global AI-infrastructure analysis groups investment into power, data-center infrastructure and IT equipment. Its scenarios range from $3.7 trillion to $7.9 trillion through 2030, with a $5.2 trillion central case. Those are assumption-dependent global scenarios, not installed-cost quotes for a facility and not a valid per-megawatt benchmark.
When requesting bids, give every supplier the same boundary and equipment schedule. Ask whether prices include freight, taxes, commissioning, spares, software licenses, warranties, integration and escalation. Record quote validity dates and currency; a price without those fields cannot be compared reliably.
Also popular on Amazon
- Portable and powerful USB-C HUB: BENFEI USB Type-C HUB, with super-soft and knot-free silicone woven design cable, meets most mobile office needs. Compact, lightweight, stylish, and powerful portable USB C Hub equipped with 1 x HDMI port, 1 x 100W charging, and 3 x USB ports. 18-month warranty, 24-hour response, to ensure you feel at ease when using our product.
- Design centered on comfort and reliability: Thanks to BENFEI's end-to-end in-house cable production capability, in-house PCBA and assembly capability, using the industry's most advanced silicone woven design and process, 20cm cable in length, no knots, super-soft, the HUB is easy to use in all scenarios: laptop, tablet, stand etc. Super-soft, 25000+ life cycles, to meet your daily carrying and office needs.
- 100W Charging: Support up to 90W USB C pass-through charging via Type-C port to keep your laptop powered. 10W is reserved for other interface operations. No data and video function on the Type-C port.
- 4K HDMI Display: The HDMI port supports media display at resolutions up to 4K 30Hz, keeping every incredible moment detailed and ultra vivid. Please note that the C port of the Host device needs to support video output.
- Transfer Files in Seconds: Transfer files and from your laptop at speeds up to 10 Gbps with USB A 3.2 port. Extra 2 USB A 2.0 ports are perfectly for your keyboards and mouse.
How do you estimate operating costs?
Start with average facility energy and the proposed delivered tariff, then add costs that are often omitted from a headline electricity number:
- Demand charges, capacity reservations, standby service and taxes.
- Maintenance contracts for electrical, cooling and fire systems.
- Staffing for operations, security, facilities and hardware support.
- Network connectivity and data-transfer charges.
- Water, treatment, discharge or heat-rejection fees where applicable.
- Software, monitoring, licenses and managed services.
- Spare parts, accelerator replacement and storage refresh cycles.
- Insurance, property taxes, leases and compliance costs.
Separate fixed and variable costs. Electricity usually varies with utilization, while staffing, leases and many maintenance contracts do not. For a multi-year model, specify inflation, tariff escalation, equipment refresh timing and residual value rather than burying them in one annual percentage.
How do you compare building, colocation and cloud alternatives?
Compare alternatives against the same workload and service level. A facility’s full capital cost should not be compared with a cloud price that excludes equivalent utilization, redundancy, support or data movement.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →| Comparison axis | Questions to ask |
|---|---|
| Capacity and utilization | What average and peak IT loads are paid for, and how much capacity sits idle? |
| Energy | Who pays electricity, demand charges and cooling overhead, and what PUE or equivalent factor is assumed? |
| Reliability | Are duplicate systems, maintenance windows and failure recovery equivalent? |
| Location | Are latency, data residency, taxes and network paths comparable? |
| Expansion | Can capacity be added when needed, or must it be reserved years ahead? |
| Lifecycle | Who funds hardware replacement, upgrades, spares and decommissioning? |
| Contract risk | What happens when energy prices, demand or hardware availability changes? |
For each option, report total cost over the same horizon and show both the average and peak resource commitment. A lower unit price can lose its advantage if utilization is low or if the contract requires paying for reserved capacity.
How should you build low, base and high cases?
Scenario analysis is more informative than a single precise-looking number. Vary the inputs that have the largest effect and state what remains constant.
| Variable | Low case | Base case | High case |
|---|---|---|---|
| Server deployment | Initial fleet only | Planned phase-one fleet | Accelerated expansion or additional server classes |
| Utilization | Lower sustained workload | Expected operating profile | High sustained utilization and larger peaks |
| Server power | Lower operating figure or newer efficient configuration | Supplier operating estimate | Higher draw, throttling or conservative design maximum |
| PUE | Efficient design under stated climate and load | Expected operating PUE | Less-efficient partial-load or legacy operation |
| Electricity | Lower delivered tariff | Contract or utility base rate | Higher tariff, demand charges or escalation |
| Grid and schedule | Existing capacity and on-time connection | Normal interconnection scope | Transmission upgrades, delay and temporary generation |
| Capital | Reusable building and standard fit-out | Planned construction scope | Ground-up work, escalation and larger contingency |
Do not change every variable arbitrarily in the high case. Explain the causal story: for example, faster growth may require more servers, a larger interconnection and an earlier equipment purchase. Keep global outlook figures, such as the IEA and McKinsey scenarios, separate from local design cases.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What does an auditable estimation worksheet look like?
A spreadsheet or model should let another reviewer trace every output to an input. Use one row per equipment class and one sheet or section for each boundary.
Free tools Windows power users keep installed
One-click scans. No signup required.
- Assumptions: workload, location, dates, utilization, reliability, tariff, PUE, growth and scope.
- IT inventory: quantity, average power, maximum power, rack assignment and source for each class.
- Load calculation: average IT kW, peak IT kW and any diversity or concurrency assumptions.
- Facility conversion: PUE, average facility kW, peak facility kW and engineered capacity.
- Energy and tariff: annual kWh, energy charges, demand charges, taxes and escalation.
- Rack and cooling check: rack count, rack kW, heat load, cooling capacity and redundancy.
- Capital: line-item quantities, quotes, currency, escalation, contingency and inclusions.
- Operations: staffing, maintenance, connectivity, water, software and replacement schedule.
- Scenarios: linked low, base and high cases with a change log.
- Review log: owner, source URL, quote date, open question and next action.
Use formulas rather than manually typed totals. Lock or label sourced values, illustrative assumptions and calculated outputs differently. A reviewer should be able to change PUE or server quantity and see the facility demand, annual energy and cost update without editing several unrelated cells.
What are the most common estimation mistakes?
- Using “AI data center = X dollars per MW.” This ignores rack density, utility upgrades, cooling architecture, land, redundancy and schedule. Use a line-item estimate.
- Confusing IT MW with facility MW. Always show the PUE multiplication and identify whether a number is average, peak or engineered capacity.
- Applying utilization to nameplate power. Include idle draw and model each equipment class separately.
- Leaving out networking and storage. Training fabrics, storage controllers and management systems consume power and occupy racks.
- Counting only the first phase. Utility and cooling systems may be sized for expansion, while unused capacity creates carrying cost if growth does not arrive.
- Assuming a rack density without a thermal design. High-density AI racks can exceed conventional air-cooling limits; verify coolant, heat rejection and service clearances.
- Ignoring interconnection timing. A low-tariff site is not useful if grid capacity arrives after the business requirement.
- Comparing unlike prices. Check currency, tax, freight, commissioning, warranties, software, spares and financing.
- Presenting forecasts as quotes. IEA and McKinsey outlooks describe broad scenarios; they do not establish local construction or electricity prices.
- Hiding uncertainty. Show the sensitivity to PUE, utilization, hardware power, energy price, redundancy and schedule.
Which sources and tools should support the estimate?
Use primary or near-primary documentation for each input. The DOE report supports category-based server modeling; the IEA provides global energy context and component shares; the OECD provides rack-density context; and the Vertiv reference design illustrates how rack power, UPS, PDUs and cooling are integrated. Schneider Electric’s TradeOff suite can help explore capacity growth, energy forecasts, capital choices and UPS efficiency, provided its assumptions are recorded.
Also popular on Amazon
- 【Read before You Buy】 Note: This adapter is USB 2.0 version and cannot transmit HDM, VGA, or other video signals. Nevertheless, they are fully capable of charging devices,duplicating files, and can be seamlessly integrated with USB-C earphones, wireless network adapters, and SD card adapters.
- 【Easy to Use and Safe】Thanks to modern manufacturing techniques, we've made these adapters sufficiently compact so that they can remain inserted in your device without noticeable protrusion or inconvenience. This means you don't have to constantly plug and unplug them, reducing the likelihood of wear and misalignment, and making them more convenient to carry around.
- 【Reliable Quality】Engineered for enhanced durability, durable aluminum alloy shell material,get excellent wear resistance and heat dissipation,the male USB C and female USB connectors & the male USB and female USB C connectors are designed to be plugged and unplugged up to 20000 times. Provides data transfer, device charging, memory space expansion, and additional computer peripherals.
- 【Think more for you】Do you need to connect a USB-C-enabled device to an odd device that only has USB-A ports? Then these USB-A male to USB-C female adapters are made for you.Simply insert the adapter into the USB-A port on the device, and you can instantly connect your USB-C device to the device that originally didn't have the correct connection port.
- 【Trusted xiwxi Quality and Service】As a responsible brand seller, you will receive xiwxi's 12-month Quality assurance and lifetime friendly customer service, you're not just buying a product,you're investing in peace of mind with all-round quality and exceptional customer service.
None of these sources supplies a universal installed cost per megawatt, a current local tariff or a project-specific interconnection price. Obtain a utility capacity letter, site and construction estimates, electrical and cooling equipment quotes, and an IT bill of materials before treating the model as a budget.
What should you hand to an engineer or finance team?
Deliver a short assumptions register alongside the model. It should state the workload, server types, operating and maximum power, utilization definition, PUE, rack density, cooling method, redundancy, location, tariff, growth phases, cost boundary and estimate date. Include a one-page summary that reports average IT load, average facility load, peak demand, engineered capacity, annual kWh, annual electricity cost and capital by category.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minuteFlag every unresolved item with an owner and decision date. Examples include the utility’s available capacity, water rights, liquid-cooling selection, equipment delivery schedule, fire-code interpretation and whether off-site generation is in scope. This turns a rough estimate into a controlled estimate that can absorb new bids without changing its meaning.
The result is not one magical dollar-per-megawatt figure. It is a transparent chain from workload to equipment, equipment to IT load, IT load to facility demand, and facility demand to capital and operating cost. That chain is what lets decision-makers explain the estimate, challenge it and update it as engineering information improves.
Frequently Asked Questions
What does 1 MW mean in terms of AI GPUs?
There is no fixed conversion because GPU servers differ in accelerator count, CPU configuration, memory, networking and operating power. Convert a 1 MW target by listing the exact server model and its average and maximum draw, then reserve separate capacity for storage, networking and facility overhead.
Should a backup generator be sized for the full AI load?
That depends on the required reliability architecture, transfer strategy and whether noncritical loads can be shed. The generator and UPS design should be based on the engineered peak, motor-start behavior, redundancy policy and local code rather than the annual average load.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
How early should I contact the utility?
Contact the utility before selecting a final site or publishing a firm schedule. Request written information on available capacity, interconnection studies, upgrade responsibility, energization milestones, voltage level, demand charges and curtailment or standby conditions; these items can dominate both cost and delivery date.
Can a small pilot use the same PUE as the finished data center?
Usually not without qualification. Partial loading, temporary cooling, oversized UPS equipment and a mixed-use building can produce a higher PUE than the stabilized full-scale design, so model pilot, phase-one and mature operation separately.

