There is no single universal number for an AI data centre’s energy or water use. A defensible estimate states the facility boundary and period, calculates or measures IT electricity, applies power usage effectiveness (PUE) to account for non-IT equipment, and reports two water quantities separately: on-site cooling water and water consumed to generate the electricity. The core equations are:
- Facility electricity = IT electricity × PUE
- Direct site water = IT electricity × site WUE (litres per kWh)
- Indirect grid water = facility electricity × grid water-consumption intensity (litres per kWh)
The result is only meaningful when the accounting boundary, workload assumptions, cooling system, climate, electricity supply and water metric are disclosed. The steps below provide a reproducible method for a single building, a portfolio or a modeled national estimate.
Define exactly what you are measuring
Write the boundary and accounting period before collecting numbers. “AI data centre use” might mean accelerator servers only, all IT equipment in one building, the whole facility including cooling and power infrastructure, or the facility plus water consumed at electricity-generation plants. Those are different quantities and must not be blended silently.
Set the physical boundary
- IT-only: servers, accelerators, storage and networking that perform the computing work.
- Whole facility: IT equipment plus cooling, uninterruptible power supplies, power conversion and distribution, pumps, fans, lighting and other building services.
- Facility plus electricity supply: the whole-facility electricity figure plus water consumed in producing that electricity. Keep the two water locations visible even when you show a combined total.
If the site hosts general-purpose workloads as well as AI, allocate only the equipment or energy that belongs to the stated AI boundary. Do not describe the entire building’s electricity as AI electricity unless every workload is in scope.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitches#1 Best Overall
- HD streaming made simple: With America’s number 1 TV streaming platform,* exploring popular apps—plus tons of free movies, shows, and live TV—is as easy as it is fun. *Based on hours streamed—Hypothesis Group
- Compact without compromises: The sleek design of Roku Streaming Stick won’t block neighboring HDMI ports, and it even powers from your TV alone, plugging into the back and staying out of sight. No wall outlet, no extra cords, no clutter.
- No more juggling remotes: Power up your TV, adjust the volume, and control your Roku device with one remote. Use your voice to quickly search, play entertainment, and more.
- Shows on the go: Take your TV to-go when traveling—without needing to log into someone else’s device.
- TV, simplified: With setup that only takes minutes, a simple-to-navigate Home Screen, and an uncluttered remote control that does all you need—Roku makes it easier to watch the TV you love.
Choose the period and reporting basis
Annual operation is common for planning and public reporting, but a month, season, training run or inference campaign can be valid if the period is stated. Use the same period for IT meters, facility meters, PUE and water records. Distinguish a measured result from a modeled estimate, a scenario and an upper-bound assumption.
Choose the water metric
Water withdrawal is water taken from a source. Water input is water supplied to a system. Water consumption is water removed from the immediate water cycle, such as through evaporation or another effectively irreversible process. Two figures both labeled “water use” may therefore be incomparable. State the metric beside every result.
Measure or estimate IT electricity
The preferred input is metered electricity for the IT boundary and the same period used for the facility calculation. A dedicated IT-meter reading avoids assumptions about equipment counts, utilization and power-management behavior.
Build a bottom-up estimate when meters are unavailable
For each equipment class, estimate average operating power under the relevant workload and multiply by the number of devices and operating hours:
IT energy (kWh) = Σ [device count × average operating power (kW) × operating hours]
Include storage and networking when they are inside the stated IT boundary. Separate training, inference and other workloads when their operating profiles are known; otherwise document the blended utilization assumption. A server’s average power should reflect the workload, accelerator duty cycle, memory, CPU activity and power-management settings, not just a catalog rating.
Do not turn nameplate power into a false measurement
Multiplying a maximum or nameplate rating by 8,760 hours produces a scenario or conservative upper bound, not observed annual consumption. Label it accordingly. Utilization, scheduled shutdowns, throttling, redundancy and the difference between peak and average load can materially reduce actual energy.
Lawrence Berkeley National Laboratory’s 2024 United States Data Center Energy Usage Report uses a bottom-up installed-equipment model and notes that shipment, equipment and operating assumptions introduce uncertainty. Its approach is useful when direct data are unavailable, but a facility meter remains preferable for a facility-specific claim.
Free tools Windows power users keep installed
One-click scans. No signup required.
Convert IT energy to whole-facility electricity with PUE
Power usage effectiveness is defined as total data-centre electricity divided by IT-equipment electricity. It is dimensionless:
Rank #2
- Ultra-speedy streaming: Roku Ultra is 30% faster than any other Roku player, delivering a lightning-fast interface and apps that launch in a snap.
- Cinematic streaming: This TV streaming device brings the movie theater to your living room with spectacular 4K, HDR10+, and Dolby Vision picture alongside immersive Dolby Atmos audio.
- The ultimate Roku remote: The rechargeable Roku Voice Remote Pro offers backlit buttons, hands-free voice controls, and a lost remote finder.
- No more fumbling in the dark: See what you’re pressing with backlit buttons.
- Say goodbye to batteries: Keep your remote powered for months on a single charge.
facility electricity (kWh) = IT electricity (kWh) × PUE
A PUE of 1.00 would mean no energy beyond IT equipment, which is not achievable in an operating facility. The difference between the IT figure and the facility figure covers cooling and heat rejection, power conversion and distribution, UPS losses, fans, pumps and other non-IT services.
Select a matching PUE
Use a measured PUE for the same building and period whenever possible. If you must model it, choose a value or range that matches the facility type, climate, redundancy design and cooling system. A warm season, a high-density accelerator hall and a lightly loaded building can have different PUEs. Berkeley Lab’s modeling incorporates cooling systems, UPS equipment, power transformation and distribution, fans, pumps, facility type and weather conditions.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
When the value is uncertain, publish low, central and high cases rather than a single decimal that implies precision. Apply the same PUE definition to every facility in a comparison; otherwise an IT-only number may be mistaken for a whole-facility number.
Estimate direct, on-site water with site WUE
Site water usage effectiveness (WUE) expresses on-site water consumption per unit of IT electricity, in litres per kWh. With a site-WUE value matched to the facility and period:
direct site water (L) = IT electricity (kWh) × site WUE (L/kWh)
Use water meters and IT-energy records together when available. If the site records withdrawals or potable-water input instead of consumption, report that measure rather than converting it without a documented method.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Account for the complete heat-rejection system
Cooling technology alone does not determine water use. Evaporative towers, dry coolers, chilled-water plants, hybrid systems and liquid-cooled servers can have very different water profiles. Liquid cooling may reduce water at the server interface while the plant still rejects heat through an evaporative or hybrid system. Operating mode, cycles of concentration, ambient temperature and humidity, maintenance practices and load all affect the result.
Berkeley Lab evaluated nine common cooling-system configurations with thermodynamic simulations across facility types and U.S. weather-station conditions. A site-WUE estimate should therefore identify the cooling configuration and climate rather than borrow an unqualified industry average.
Rank #3
- 4K streaming made simple:With America’s number 1 TV streaming platform,* exploring popular apps—plus tons of free movies, shows, and live TV—is as easy as it is fun. *Based on hours streamed—Hypothesis Group
- 4K picture quality: With Roku Streaming Stick Plus, watch your favorites with brilliant 4K picture and vivid HDR color.
- Compact without compromises: Our sleek design won’t block neighboring HDMI ports, and it even powers from your TV alone, plugging into the back and staying out of sight. No wall outlet, no extra cords, no clutter.
- No more juggling remotes: Power up your TV, adjust the volume, and control your Roku device with one remote. Use your voice to quickly search, play entertainment, and more.
- Shows on the go: Take your TV to-go when traveling—without needing to log into someone else’s device.
Report seasonal or operating variation when it matters
Annual WUE can conceal a dry, hot season in which evaporation rises or a cool season in which a dry cooler operates with little water. If the facility has monthly records, show seasonal values or at least explain how the annual figure was formed. A modeled annual value should identify the weather data and operating schedule used.
Estimate water consumed to supply electricity
Electricity generation consumes water at power plants and other generation facilities. This indirect quantity is separate from water used at the data-centre site:
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →indirect grid water (L) = facility electricity (kWh) × grid water-consumption intensity (L/kWh)
Match the factor to the electricity supply
Use a generation mix and water-consumption factor that match the facility’s location and electricity procurement. A regional factor is better than a national average when the grid region is known. On-site generation, dedicated contracts, storage and behind-the-meter resources can change the attribution. If those details are unavailable, say that a grid-average factor is being used as a fallback rather than presenting it as a site-specific measurement.
Berkeley Lab’s national estimate combines regional electricity data with plant-water data. It did not incorporate facility-specific power-purchase agreements or behind-the-meter generation because detailed information was unavailable. That limitation matters when applying a national factor to an individual building.
Keep direct and indirect water visible
Show direct site water and indirect grid water in separate rows, identify where each occurs and state that both are consumption figures if you add them. A combined total is defensible only when the accounting definitions, period and boundary are compatible. Otherwise, combining them can imply that water consumed at a power plant occurred at the data-centre site.
Worked example with explicit assumptions
The following is an illustrative calculation, not a measurement of a real facility. Assume a defined AI IT boundary uses 10,000,000 kWh in one year, the modeled PUE is 1.30, site WUE is 0.50 L/kWh, and the grid-water factor is 4.52 L/kWh. The last factor is the 2023 U.S. data-centre electricity average reported by Berkeley Lab and is used here only to demonstrate the arithmetic; it is not a universal value.
| Quantity | Calculation | Illustrative result |
|---|---|---|
| IT electricity | Given assumption | 10,000,000 kWh |
| Whole-facility electricity | 10,000,000 × 1.30 | 13,000,000 kWh |
| Direct site water consumption | 10,000,000 × 0.50 | 5,000,000 L |
| Indirect grid water consumption | 13,000,000 × 4.52 | 58,760,000 L |
| Combined consumption | 5,000,000 + 58,760,000 | 63,760,000 L, only if both definitions are compatible |
The example shows why the choice of water boundary matters: the grid-related amount can exceed on-site cooling water even though it is consumed away from the data centre. Replacing any assumption with a meter reading or a location-specific factor changes the result.
Identify the variables that can move the estimate
Hardware and workload
- Accelerator type, server configuration, memory and networking determine power at a given utilization.
- Training can sustain high accelerator utilization for long periods; inference may be bursty or distributed across many smaller requests.
- Over-provisioning, redundancy and idle capacity consume energy even when user demand is low.
- Storage, cooling controls and network equipment can remain active outside the main compute workload.
Climate and cooling design
- Outdoor temperature and humidity alter chiller, fan, pump and evaporative-system operation.
- Dry cooling can reduce on-site water while increasing electricity in some conditions; evaporative systems can have the opposite trade-off.
- Liquid cooling changes heat-transfer hardware but does not, by itself, establish low site WUE.
- Water availability, treatment requirements and discharge rules can constrain operating modes.
Electricity source
- Regional generation mix changes the water consumed per kWh.
- Contracts for specific generation, on-site generation and storage may change how indirect water is attributed.
- A national-average factor can be useful for context but should not be presented as the facility’s actual supply profile.
Build a low, central and high scenario
Scenario ranges are more honest than false precision when the equipment inventory, utilization, PUE, WUE or electricity supply is uncertain. Keep the variables visible so a reader can see what drives the spread.
Rank #4
- Stunning 4K and Dolby Vision streaming made simple: With America’s number 1 TV streaming platform,* exploring popular apps—plus tons of free movies, shows, and live TV—is as easy as it is fun. *Based on hours streamed—Hypothesis Group
- Breathtaking picture quality: Stunningly sharp 4K picture brings out rich detail in your entertainment with four times the resolution of HD. Watch as colors pop off your screen and enjoy lifelike clarity with Dolby Vision and HDR10+.
- Seamless streaming for any room: With Roku Streaming Stick 4K, watch your favorite entertainment on any TV in the house, even in rooms farther from your router thanks to the long-range Wi-Fi receiver.
- Shows on the go: Take your TV to-go when traveling—without needing to log into someone else’s device.
- Compact without compromises: Our sleek design won’t block neighboring HDMI ports, so you can switch from streaming to gaming with ease. Plus, it’s designed to stay hidden behind your TV, keeping wires neatly out of sight
- Low case: use the lower credible IT load, efficient operating profile, lower matched PUE and lower matched site-WUE value.
- Central case: use the best-supported meter readings or modeled values for the expected workload and operating year.
- High case: use a higher credible load, less favorable operating conditions and an appropriate upper factor; do not silently substitute nameplate maximums for measured demand.
- Vary one assumption at a time: show how the result changes when PUE, site WUE or the grid factor moves while the other inputs stay fixed.
- Label evidence quality: mark each input as measured, modeled, benchmarked or scenario-based.
At minimum, disclose the IT inventory or measured IT energy, workload and utilization assumptions, operating hours, PUE source and period, cooling configuration, site climate, WUE definition, grid-factor geography and source, treatment of power contracts and on-site generation, and all exclusions.
Berkeley Lab’s authors summarize the challenge this way: “The lack of direct energy data available in a sector with relatively little transparency introduces a significant amount of uncertainty into these estimates.” That uncertainty should appear in the reported range, not disappear behind extra decimal places.
Put national and global figures in the right context
Large published totals describe fleets or scenarios, not the footprint of one AI data centre. They are useful for scale and trend context only.
| Scope and source | Figure | What it means |
|---|---|---|
| United States, Lawrence Berkeley National Laboratory, 2023 | 176 TWh of data-centre electricity | Estimated national data-centre electricity use for 2023, not an AI-facility meter reading. |
| United States, Lawrence Berkeley National Laboratory, 2023 | 66 billion L of direct water consumption | On-site U.S. data-centre water consumption under the report’s definitions. |
| United States, Lawrence Berkeley National Laboratory, 2023 | Nearly 800 billion L of indirect water consumption | Water consumed in electricity generation; the report gives an estimated national-average intensity of 4.52 L/kWh for data-centre electricity. |
| Global, International Energy Agency, 2025 | About 415 TWh in 2024 | Estimated global data-centre electricity consumption. |
| Global, International Energy Agency, 2025 Base Case | Around 945 TWh in 2030 | A scenario projection, not a guaranteed outcome or a measurement of AI-only demand. |
The U.S. water figures illustrate the importance of location: the larger indirect estimate occurs at electricity-generation sources, while direct water is associated with data-centre sites. Adding them without labeling their locations and definitions obscures rather than clarifies impact.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Compare two facilities or estimates on equal terms
A lower number is not automatically a more efficient facility. Align these dimensions before drawing a comparison:
Outdated 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 matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstall| Comparison dimension | Question to answer |
|---|---|
| Boundary | Is the number IT-only, whole-facility or facility plus electricity generation? |
| Period and geography | Do both figures cover the same months or year and a comparable climate or grid region? |
| Workload and hardware | Are accelerator mix, utilization, training, inference and non-AI workloads treated similarly? |
| PUE | Is PUE measured for the same facility and period, or modeled from a benchmark? |
| Site WUE | Do both values use the same definition of water and the same on-site boundary? |
| Cooling and climate | Are the cooling configuration and weather conditions comparable? |
| Grid water factor | Is the factor location-specific, grid-average or based on a different electricity supply? |
| Water measure | Are both figures withdrawals, inputs or consumption? |
| Evidence status | Are the values metered, modeled, projected or scenario assumptions? |
Meet reporting requirements without overclaiming
For facilities covered by European Union reporting rules, Commission Delegated Regulation (EU) 2024/1364 specifies measurement of total potable-water input using the EN 50600-4-9 WUE Category 1 methodology or an equivalent. It also requires measurement of heat reused outside the data-centre boundary. Confirm the regulation’s applicability, the reporting year and the facility’s legal scope before describing a result as compliant. A general WUE estimate is not automatically a regulatory submission.
Common estimation mistakes and their fixes
Reporting one “AI water” number
Problem: direct cooling water and water consumed by power generation are merged, or withdrawals are mistaken for consumption.
Fix: publish separate site and source-water rows, name the metric and show a combined total only when the definitions match.
Using a universal water-per-prompt conversion
Problem: a prompt-level figure is treated as the footprint of an entire facility.
Best Value
- The Google TV Streamer (4K) delivers your favorite entertainment quickly, easily, and personalized to you[1,2]
- HDMI 2.1 cable required (sold separately)
- See movies and TV shows from all your services right from your home screen[2]; and find new things to watch with tailored recommendations for everyone in your home based on their interests and viewing habits
- Watch live TV and access over 800 free channels from Pluto TV, Tubi, and more[3]; if you find an interesting show or movie on your TV, mobile app, or Google search, you can easily add it to your watchlist, so it’s ready when you are[2]
- Up to 4K HDR with Dolby Vision delivers captivating, true-to-life detail[4]; and you can connect speakers that support Dolby Atmos for more immersive 3D sound
Fix: use facility energy, PUE, site WUE and a matched grid factor. Prompt-level estimates depend on model, hardware, batching, response length, utilization, cooling and electricity supply, so no universal conversion is established here.
Multiplying maximum power by all hours
Problem: nameplate power times 8,760 is presented as annual measured consumption.
Fix: call it an upper-bound scenario or replace it with metered energy and a workload-based average-power estimate.
Applying a national average as a site measurement
Problem: a national PUE, WUE or grid-water factor is reported as if it described one building.
Recommended Free Tools
Fix: identify the geography and source, use a facility-specific value where available and show a range when it is not.
Ignoring non-AI workloads
Problem: all building electricity is attributed to AI even though other services run in the facility.
Fix: define an allocation method or report the whole facility without calling it AI-only.
Combining incompatible periods
Problem: annual IT energy is multiplied by a seasonal PUE or a water factor from another year without explanation.
Fix: align periods or disclose the mismatch and treat the output as a modeled scenario.
A reproducible reporting layout
A clear estimate can be audited when it presents the following in order:
- Facility name or anonymized identifier, location and physical boundary.
- Accounting period and whether the value is measured, modeled, projected or scenario-based.
- IT-energy meter reading or equipment inventory, average power, utilization and hours.
- PUE value, definition, source, period and uncertainty range.
- Cooling configuration, climate assumptions and site-WUE definition.
- Direct site-water result with its unit and whether it is withdrawal, input or consumption.
- Facility electricity result and the grid-water factor’s geography, source and water metric.
- Indirect grid-water result shown separately from direct site water.
- Low, central and high cases where unknowns materially affect the result.
- Exclusions, such as construction water, fuel extraction, hardware manufacturing, wastewater treatment or power contracts, if they are outside the boundary.
This layout lets a reader replace a benchmark with a meter reading, change the workload or update the grid factor without rebuilding the entire estimate.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →
Say something about this film