A BlackForge Ventures company
BlackForge Data Centers

Power & interconnection

How Much Power Does a Data Center Need?

A data center needs anywhere from about 1 MW for a small edge site to hundreds of megawatts for a hyperscale campus, and the largest AI campuses now plan for a gigawatt or more: the Stargate campus in Abilene, Texas, was planned at about 1.2 GW,1 and Meta has said its Hyperion campus in Louisiana will reach 5 GW.2 The number a utility sees is the IT load multiplied by the facility’s PUE, delivered on a phased ramp that utilities increasingly tie to minimum-payment commitments.3 Getting that number and its timing right is the first step in judging whether a site can work.

Last reviewed · 10 min read · BlackForge Data Centers

Key takeaways

  • Edge sites run roughly 1–10 MW, enterprise and colocation 10–60 MW, hyperscale 100–500 MW, and the largest AI campuses 1 GW and up.12
  • Total facility demand equals critical IT load times PUE, and the industry’s weighted average PUE has hovered just above 1.5 for six straight years.45
  • Utilities plan around peak demand and a year-by-year ramp, and tariffs such as AEP Ohio’s now bill large data centers for at least 85% of contracted demand.3
  • AI racks can draw about 120 kW each, roughly ten times a general-purpose server rack, so the same floor area can need far more power.6
  • Redundancy (N+1, 2N) changes how much equipment you install on site, not how much the utility must serve.
  • Power, not land, is usually the binding constraint on how large a campus can grow.
Public data · Power trackerLarge-load power tracker39 utilities and their data center power rules, from public filings.

01Typical power demand by data center type

Data center power is quoted in megawatts (MW). The range is wide because the facilities are very different. A single-building edge site near a population center has little in common with a multi-building AI training campus, and their power needs differ by two to three orders of magnitude. The sector as a whole is growing fast: the Department of Energy reported that data center load tripled over the past decade and is projected to double or triple again by 2028, based on Lawrence Berkeley National Laboratory’s 2024 report to Congress.78

Indicative demand by facility type (planning ranges, not industry statistics)
Facility typeTypical powerRelative land needUsual service
Edge1–10 MWA few acresDistribution, or a small transmission tap
Enterprise and colocation10–60 MWTens of acresDistribution at the low end; transmission above that
Hyperscale campus100–500 MWHundreds of acresTransmission with a dedicated substation
AI and gigawatt campus1 GW to 5 GW announcedHundreds to thousands of acresHigh-voltage transmission, often several sources and new generation9

These are planning ranges, not rules. A colocation operator can build a 100 MW campus, and a hyperscaler can build a 30 MW facility. The largest projects are moving the top of the scale: Stargate’s first phase in Abilene opened with two buildings at about 200 MW, with a second phase of six more buildings bringing the campus to about 1.2 GW,1 and Meta has described Hyperion as a 5 GW campus on a site of about 4,000 acres.2 Our guides to site needs by data center type and what a gigawatt-scale AI campus requires cover how these profiles differ beyond power.

Fig. 1Two gigawatt-scale AI campuses

Stargate Abilene, first two buildings
~200 MW
Stargate Abilene, eight buildings
~1.2 GW
Meta Hyperion, announced capacity
5 GW
New gas generation approved for Hyperion9
~2.25 GW
Announced and reported figures for two public projects; plans and phasing change, so check current reporting before relying on them.12

02IT load vs. total facility load

Developers often describe a project by its critical IT load: the power consumed by servers, storage and network equipment. The utility has to serve more than that. Cooling systems, fans, pumps, lighting, offices and losses in transformers and UPS systems all add to the total.

The standard measure of that overhead is power usage effectiveness (PUE): total facility energy divided by IT equipment energy. It is standardized internationally as ISO/IEC 30134-2, whose second edition was published in January 2026.10 A PUE of 1.0 would mean zero overhead. Uptime Institute’s 2025 global survey found the weighted average still hovering above 1.5, with little change for the sixth consecutive year, held back by older facilities and regional barriers to efficient cooling.4 Its October 2025 analysis of 681 operators found the global average relatively stable since 2018.5 New hyperscale and AI facilities typically design for much lower figures, so the right number for a site depends on climate, cooling design and how fully the building is loaded. Our guide to PUE and data center energy efficiency goes deeper.

  • Total facility demand ≈ IT load × PUE.
  • Example: 100 MW of IT load at a design PUE of 1.3 is about 130 MW of facility demand.
  • Design PUE (at peak, on the hottest day) is what sizes the utility request. Annual average PUE is lower and matters for energy cost, not interconnection.

Fig. 2From IT load to utility demand

Illustrative
  • Critical IT load100
  • Cooling at peak+22
  • Electrical losses+6
  • Lighting, offices, other+2
  • Facility demand130

MW

Illustrative example of a 100 MW IT load at a peak design PUE of 1.3. The split of overhead varies with climate and cooling design.

Utilities also think in MVA rather than MW. MVA is apparent power, which includes the reactive component. Data centers generally operate at a high power factor, but a utility will still ask for both figures, and transformers and lines are rated in MVA.

03What drives a data center’s power demand

Within a given building footprint, four things set how many megawatts it will draw.

  • Rack density. A general-purpose CPU rack supports up to about 12 kW and an air-cooled H100 GPU rack about 40 kW, while an NVIDIA GB200 NVL72 rack needs roughly 120 kW and liquid cooling.6 The same floor area can therefore need several times more power.
  • Cooling approach. Air cooling, evaporative systems and direct liquid cooling have different energy overheads and different sensitivity to climate. The choice moves PUE and, with it, the facility load.
  • Building count and phasing. Campuses are built one building or one data hall at a time. The full-build number may be years away.
  • Non-IT loads. Offices, security, mechanical yards and on-site water treatment are small relative to IT load but still part of the request.

Fig. 3Power per rack by rack type

  • General-purpose CPU rackup to ~12
  • Air-cooled H100 rack~40
  • GB200 NVL72 (liquid)~120

kW per rack

Approximate figures from SemiAnalysis; actual draw depends on configuration. The GB200 NVL72 is about ten times a general-purpose rack.6

Redundancy is often misunderstood. A 2N electrical design installs two complete power paths, but the IT equipment only draws its load once. Redundancy increases the transformers, switchgear, UPS and generators on site. It does not double the demand placed on the grid. What it can change is the request for two independent utility sources, which affects the substation and transmission design, a topic covered in our guide to redundant utility feeds.

AI workloads add another consideration. Large training clusters can change their power draw quickly and in step. In May 2026 NERC issued a rare Level 3 alert on large loads, citing customer-initiated load reductions and significant oscillations that occur in seconds, and set an August 3, 2026 deadline for registered entities to respond to its seven essential actions.1112 Utilities now ask about load behavior, ride-through during faults and how fast demand can swing, not only how large it is; CenterPoint Energy’s load data request, for example, notes that planners need to understand data center characteristics to model them in studies.13 Our guides to NERC reliability and large loads and AI data center site requirements go further.

04Peak demand, load factor and the ramp schedule

A utility does not plan around one final number. It plans around peak demand in each year and how that demand grows. A large-load request is usually expressed as a ramp schedule: for example, 50 MW in the first year of service, 150 MW two years later and 300 MW at full build.

Fig. 4Illustrative three-phase load ramp

Illustrative
  • Phase 1: 50 MWFirst building energized
  • Phase 2: 150 MW totalSecond and third buildings
  • Phase 3: 300 MW totalFull build
02468years
Illustrative example. Each bar shows the years in which that phase sets the campus peak; real ramps follow the utility’s upgrade schedule.

Data centers are also high-load-factor customers. They run close to their peak around the clock, with little daily or seasonal variation compared with most commercial or industrial loads. That makes them attractive to some utilities, because the energy sold per MW of capacity is high. It also means the grid has to carry that load during its own peak hours, with no natural off-peak relief.

A credible ramp matters because utilities now attach money to it. In July 2025 the Public Utilities Commission of Ohio approved AEP Ohio’s data center tariff for loads above 25 MW: customers pay for at least 85% of the demand they subscribe to, even if they use less, for up to 12 years including a ramp-up period of up to four years, must show financial viability, and owe an exit fee if a project is cancelled.3 A ramp that overstates early demand can therefore create obligations the project cannot meet. Our guides to load ramp schedules and large-load tariffs cover where the ramp becomes binding.

05How power requirements translate to land

Power and land move together, but power is usually the constraint. A 300-acre tract with access to 40 MW is a colocation site, not a hyperscale campus, however large it is. The reverse also happens: a site with a strong transmission path may be held back by too little buildable acreage once setbacks, wetlands and a substation footprint come out.

At the top of the scale, the power plan can reshape a region. Louisiana regulators approved about 2.25 GW of new gas-fired generation and at least 1.5 GW of solar tied to Meta’s Richland Parish campus in August 2025.9 Higher rack densities are pushing more megawatts onto each acre, which raises the importance of electrical yard space, substation pads, cooling equipment and, for some designs, room for on-site generation. Our guide to how much land a data center needs works through the land side of the math.

06How to estimate the power a site should support

  1. 01Start with the target product: edge, colocation, hyperscale or AI campus, and the likely IT load at full build.
  2. 02Estimate rack density and cooling approach, since GPU racks with liquid cooling draw far more per rack than general-purpose racks.6
  3. 03Apply a design PUE suited to the climate and cooling approach to get facility demand in MW, then convert to MVA.
  4. 04Break the full build into phases with target energization years, and check the serving utility’s tariff for minimum-demand terms.3
  5. 05Decide whether the design needs one utility source or two independent sources.
  6. 06Describe how the load behaves, including ride-through and how fast it can change, because utilities need it for their studies.13
  7. 07Compare that ramp with what the nearby transmission system and substations can plausibly deliver, and when.

The output is a load profile you can take to a utility. It also becomes the test for every site on a shortlist: can this location deliver this ramp on this schedule? When we screen a site, that question comes before almost any other. If you have a parcel in mind, you can get a site reviewed.

CalculatorIT load to grid demand →CalculatorBackup generators & fuel →

Common questions

How many megawatts does a hyperscale data center use?

A hyperscale campus typically plans for 100–500 MW at full build, usually spread across several buildings that come online in phases. Individual buildings on the campus often draw a fraction of that; Stargate’s first two buildings in Abilene were reported at about 200 MW.1 The figure the utility serves is the IT load multiplied by the design PUE, so a campus described as 300 MW of IT capacity will need noticeably more than 300 MW from the grid.

How much power does an AI data center need?

The largest AI training campuses are planned at 1 GW or more: Stargate Abilene at about 1.2 GW and Meta’s Hyperion at an announced 5 GW.12 Smaller AI deployments inside colocation or enterprise facilities can be far less. What sets AI apart is density: a GB200 NVL72 rack needs about 120 kW, roughly ten times a general-purpose rack, so AI sites need much more power per acre and more attention to how fast the load changes.611

What is PUE and why does it matter for power requirements?

Power usage effectiveness (PUE) is total facility energy divided by IT equipment energy, as defined in ISO/IEC 30134-2.10 It captures the overhead of cooling, power conversion and building systems; the global average has hovered just above 1.5.4 For siting, the design PUE at peak conditions matters most because it sets the total demand the utility must serve. A facility with 100 MW of IT load and a design PUE of 1.3 needs about 130 MW of utility capacity.

Does 2N redundancy double a data center’s power demand?

No. A 2N design installs two complete electrical paths, but the IT equipment only draws its load once, so the demand on the grid does not double. Redundancy increases the transformers, switchgear, UPS systems and generators on site. It can, however, lead the developer to request two independent utility sources, which shapes the substation and transmission design.

Will the utility charge for capacity a data center does not use?

Increasingly, yes. AEP Ohio’s tariff, approved in July 2025, requires new data centers above 25 MW to pay for at least 85% of their subscribed demand for up to 12 years, with exit fees if a project is cancelled.3 Other utilities use similar minimum-bill and collateral terms, so confirm the terms with the serving utility before fixing a ramp.

Notes

  1. 1.ETCentric, “Meta Power Play Includes Data Centers of Up to 5 Gigawatts,” 2025. etcentric.org
  2. 2.Bisnow, “Meta Ups Economic Investment To $50B For Its Largest Data Center Project,” 2025. bisnow.com
  3. 3.POWER Magazine, “Regulator Approves AEP Ohio’s Landmark Data Center Tariff,” 2025. powermag.com
  4. 4.Uptime Institute, “Uptime Institute Global Data Center Survey 2025 (UII Keynote Report 180),” 2025. intelligence.uptimeinstitute.com
  5. 5.Uptime Institute, “Mapping PUE Trends by Data Center Region, Age and Size (UII Update 420),” 2025. intelligence.uptimeinstitute.com
  6. 6.SemiAnalysis, “GB200 Hardware Architecture and Component Supply Chain,” 2024. newsletter.semianalysis.com
  7. 7.U.S. Department of Energy, “DOE Releases New Report Evaluating Increase in Electricity Demand from Data Centers,” 2024. energy.gov
  8. 8.Lawrence Berkeley National Laboratory, “2024 United States Data Center Energy Usage Report,” 2024. eta-publications.lbl.gov
  9. 9.Industrial Info Resources, “Louisiana Greenlights Massive Power Projects for Meta Data Center,” 2025. industrialinfo.com
  10. 10.International Organization for Standardization, “ISO/IEC 30134-2:2026, Information Technology, Data Centres Key Performance Indicators, Part 2: Power Usage Effectiveness (PUE),” 2026. iso.org
  11. 11.Energy-Storage.news, “NERC’s Level 3 Alert Puts Data Centres on Notice as Grid Faces Gigawatt-Scale Load Swings,” 2026. energy-storage.news
  12. 12.Renewable Energy World, “NERC Issues Level 3 Alert to Address ‘Immediate Risks’ Data Center Loads Pose to the Grid,” 2026. renewableenergyworld.com
  13. 13.CenterPoint Energy Houston Electric, “CEHE Load Interconnection Data Request Form,” n.d. assets.centerpointenergy.com

Have a site in mind?

Get a straight answer on your land.

Send a parcel number, an address, a map pin or a target load. We’ll tell you what it can support and what it would take.

Start a conversation →

This guide is general information about data center site selection. It is not engineering, legal, tax or investment advice. Requirements vary by state, utility and county, so confirm the specifics for any site with the relevant authorities and advisors.

Terms in this guide

Related guides

More in Power & interconnection

Guide 1 of 35 in Power & interconnection · all 242 guides · glossary · calculators