Data center megawatts: IT load vs utility load

A data center MW figure can mean IT load, grid capacity requested or generation. How to tell which one a filing reports before you model it.

Megawatts · 7 min ·

Draft copy
On this page9 sections

A megawatt figure next to a data center name usually means one of two things: the IT load the building is designed to serve, or the grid capacity the site asks its utility for. The second is the larger number, and neither one tells you what the site draws today.

Sometimes the figure is not load at all. Before a number goes into a model, find out which of these it is.

What a data center megawatt measures

A data center megawatt is a capacity figure, a rate of power at peak. It is not the energy the site uses over a year, which is counted in megawatt-hours or terawatt-hours.

Lawrence Berkeley National Laboratory (LBNL) says power capacity “refers to either the facility’s nameplate capacity”, meaning “the IT hardware load it is designed to serve”, or “to the grid interconnection capacity requested to supply power to the facility.” Those are the two bases you will meet most often.

Annual energy is a separate figure. LBNL’s 2025 update estimates US data center use in 2024 at 192 TWh, or 4.7% of US electricity. The report also warns that converting between energy and capacity is hard without data on average demand, maximum demand, nameplate capacity and requested interconnection capacity. A TWh total and a MW pipeline answer different questions, so keep them apart.

IT load: what the servers are built for

IT load is the power the computing equipment is designed to draw: servers, storage and network gear. You will also see it called critical IT load or nameplate IT capacity. LBNL ties it to “infrastructure like uninterruptible power supply systems”, which is how a building’s IT capacity gets sized.

IT load is the smaller of the two numbers because it leaves out cooling and power losses. The Congressional Research Service (CRS) puts the split this way: “Roughly one-half or greater of the electric power demand of data centers stems directly from the operation of electronic IT equipment. Much of the rest is for cooling.”

Operator announcements and leasing figures often quote IT capacity, because that is the part a tenant uses. When a release gives one MW number with no basis, treat it as unlabelled until a filing says which.

Utility load: what the grid is asked for

Utility load is the peak power a site requests at its point of connection to the grid. LBNL defines grid interconnection capacity as “the maximum amount of electrical power a data center requests from the utility or grid operator at its point of connection to the grid”, and adds: “This reflects total facility power needs (for IT, cooling, and other systems) plus design margins.”

In Texas, the grid operator’s numbers are on this basis. ERCOT defines a Large Load as “one or more facilities at a single site with an aggregate peak demand greater than or equal to 75 MW behind one or more common points of interconnection or service delivery points.” Its Planning Guide asks the developer for a schedule for “phased energization of the requested or contracted peak demand expressed in MW.”

State law uses the same threshold. Senate Bill 6 (2025) tells the Public Utility Commission of Texas (PUCT) to set interconnection standards for sites whose total load “would exceed a demand threshold”, and to set that threshold at 75 MW unless it decides a lower one is needed. ERCOT’s wording is “greater than or equal to”; the statute says “exceed”. At the boundary, check which rule applies.

Power usage effectiveness (PUE) is the ratio between them. LBNL defines it as “the ratio of total facility energy use (including cooling and power distribution) to IT equipment energy use (excluding facility infrastructure).” A PUE of 1.2 means the whole facility uses 1.2 units of energy for every unit the IT equipment uses.

LBNL estimates an average PUE of 1.145 in 2024 for facilities serving AI equipment, and a national average across all data centers that improved from 1.55 in 2018 to 1.45 in 2024.

PUE is an energy ratio averaged over time. It does not convert IT capacity into an interconnection request on its own, because the request also carries design margin and redundancy. LBNL says utilization of interconnection capacity “is not well documented but is estimated to be around 50% due to high redundancy requirements and maintenance needs.”

Grid interconnection request200 MW
Facility peak at PUE 1.2120 MW
IT load (nameplate)100 MW
Illustrative figures. Not a real site. One building, three bases: never add them together.

Requested, approved and used are three numbers

A grid operator’s large-load figures split into what was requested, what was approved to energize and what is actually drawn. The gap between them is large.

ERCOT told its board in June 2026 that it was tracking about 438 GW of large-load interconnection requests. Its September 2026 operational overview reports 9,456 MW of load that has received approval to energize, and an observed peak of 4,132 MW from those loads that month.

Approved to energize9,456 MW
Observed peak, Sep 20264,132 MW
ERCOT Monthly Operational Overview, September 2026, page 8. All large loads in ERCOT, not data centers alone.

ERCOT calculates that peak as “the sum of the maximum value for each individual load per month”, and says it “represents how much approved load ERCOT believes is now operational.” The figure covers every large load in ERCOT, so it is not a data center count. It still shows the pattern: an approved megawatt is not a consumed one. The queue and the stage of each project are covered in Phantom data centers: the record test.

Generation is not load

Some MW figures attached to a data center describe a power plant. Texas allows a large load to share a meter with an existing generator under a net metering arrangement (PURA §39.169), so docket titles can carry both.

PUCT Docket 58881 is one. The order describes a wind facility in Armstrong County “that provide a total of 265.5 megawatts (MW) of generating capacity”, and a co-located AI data center “developed in two phases, an initial phase of 180 MW and a second phase of 85 MW.” The load phases add up to 265 MW. The generation is 265.5 MW. Two numbers that close are easy to confuse.

PUCT 58881 · Order · 2026-04-23

265.5 MW is wind generating capacity. The load is 180 MW plus 85 MW.

Texas Public Utility Commission, Docket 58881, Order page 2. The Commission approved the arrangement subject to conditions; Ensign and Crusoe each filed a motion for rehearing on 2026-05-18.

An approval like this one is also a permit stage, not proof of power flowing. Announced isn’t built sets out which document proves each stage.

What a 500 MW data center means

A “500 MW data center” is a headline figure with no basis attached. It could be 500 MW of IT load, a 500 MW interconnection request, the total for every phase of a campus, or a generator’s capacity.

For scale, CRS describes a hyperscale facility as having “an electric power rating, sometimes referred to as power draw, exceeding 100 megawatts”. A 500 MW campus is several of those. If the figure is IT load, the facility will need more at the meter. If it is the interconnection request, the IT load is smaller. Either way it rarely arrives at once, and how long it takes to build a data center depends on which phase you are counting.

Questions for every MW figure

  1. Is it load or generation?
  2. Is it IT load or the grid request at the point of interconnection?
  3. Is it requested, approved to energize or observed?
  4. Is it one phase or the whole campus?
  5. Which document states it, and on what date?

If a filing doesn’t answer the first two, record the figure with its basis marked unknown and keep it out of any total.

Sources

Accessed 2026-11-07.

Spotted an error? Corrections are logged on the article.

Send a correction