Data Center Power Statistics 2026: 30 Numbers With Sources
Compare the latest national electricity estimates with public generator and utility records. Use the definitions and downloadable calculations to size a service territory without confusing planned megawatts with operating equipment.
LBNL's June 2026 update estimates that U.S. data centers used 192 TWh of electricity in 2024, or 4.7% of national consumption. Its 2030 Reference Case projects 649 TWh, with a compounded uncertainty range of 521 to 843 TWh. PorchRocket's separate analysis of Virginia's October 2026 public report totals 13,767 permitted generator sets and 33,132 MW of permitted generator capacity. Those permit figures describe equipment authorization, not measured electricity use.
What matters most
- LBNL's latest update was published in June 2026, despite carrying the subtitle 2025 Update, and its last historical year is 2024.
- PorchRocket calculates that LBNL's 649 TWh Reference Case for 2030 equals 74.1 GW of annual average load, not installed or interconnection capacity.
- PorchRocket's analysis of Virginia's October 2026 report totals 13,767 permitted generator sets and calculates that Tier 4 or Tier 4-equivalent sets represent 25.3% of the count.
- DOE's 12 to 30 month distribution transformer lead-time range describes 2023; its October 2026 update still reports multi-year transformer delays.
U.S. data centers used an estimated 192 TWh of electricity in 2024, according to the June 2026 update from Lawrence Berkeley National Laboratory (LBNL). Its Reference Case projects 649 TWh by 2030. For a generator service company or electrical equipment supplier, the commercial question is where that demand leaves serviceable equipment inside the territory your crew can cover. LBNL’s updated energy report supplies the national baseline.
Updated November 2026. PorchRocket’s analysis below converts the national scenarios into annual average load and calculates Virginia’s permitted generator totals and tier shares from public records. Asset Search by PorchRocket supports the next task: finding visible power equipment in a defined territory, with reviewed evidence and owner information. The figures keep electricity use, utility agreements and generator permits in separate columns.
Key statistics
LBNL, EIA, DOE and Virginia SCC supply the public evidence behind these 30 figures, with PorchRocket named on every derived result.
- LBNL’s June 2026 report estimates that U.S. data centers consumed 192 TWh of electricity in 2024.
- LBNL’s June 2026 report estimates that data centers accounted for 4.7% of total U.S. electricity consumption in 2024.
- LBNL’s June 2026 Reference Case projects 29% growth in U.S. data center electricity consumption from 2025 to 2026.
- LBNL’s June 2026 Reference Case projects U.S. data center electricity consumption of 464 TWh in 2028.
- LBNL’s June 2026 Reference Case projects U.S. data center electricity consumption of 649 TWh in 2030.
- LBNL’s June 2026 Reference Case places data centers at 11.8% of forecast total U.S. electricity consumption in 2030.
- LBNL’s June 2026 compounded uncertainty low scenario projects U.S. data center electricity consumption of 521 TWh in 2030.
- LBNL’s June 2026 compounded uncertainty high scenario projects U.S. data center electricity consumption of 843 TWh in 2030.
- PorchRocket’s analysis of LBNL’s June 2026 Reference Case calculates 74.1 GW of annual average U.S. data center load.
- PorchRocket’s analysis of LBNL’s estimates calculates a 52.2 GW increase in annual average U.S. data center load from historical 2024 to Reference Case 2030.
- PorchRocket’s analysis of LBNL’s rounded figures calculates 238.0% growth in annual U.S. data center electricity use from historical 2024 to Reference Case 2030.
- PorchRocket’s analysis of LBNL’s June 2026 forecasts calculates a 322 TWh spread between the low and high compounded uncertainty cases for U.S. data center electricity use in 2030.
- PorchRocket’s analysis of Virginia’s October 1, 2026 SCC report totals 13,767 permitted data center generator sets across its two reported tier groups.
- PorchRocket’s analysis of Virginia’s October 1, 2026 SCC report totals 33,132 MW of permitted data center generator capacity.
- Virginia’s October 1, 2026 SCC report identifies 10,290 permitted data center generator sets as Tier 2 or earlier.
- Virginia’s October 1, 2026 SCC report identifies 3,477 permitted data center generator sets as Tier 4 or Tier 4-equivalent.
- PorchRocket’s analysis of Virginia’s October 2026 SCC figures calculates that Tier 2 or earlier sets represent 74.7% of the reported permitted generator count.
- PorchRocket’s analysis of Virginia’s October 2026 SCC figures calculates that Tier 4 or Tier 4-equivalent sets represent 25.3% of the reported permitted generator count.
- PorchRocket’s analysis of Virginia’s October 2026 SCC figures calculates that Tier 2 or earlier sets represent 80.6% of the reported permitted generator MW.
- PorchRocket’s analysis of Virginia’s October 2026 SCC figures calculates that Tier 4 or Tier 4-equivalent sets represent 19.4% of the reported permitted generator MW.
- PorchRocket’s analysis of Virginia’s October 2026 SCC figures calculates mean permitted capacity of 2.41 MW per reported data center generator set.
- PorchRocket’s analysis of Virginia’s October 2026 SCC utility figures totals 21,791.4 MW of reported data center service agreements or equivalent contracted demand.
- PorchRocket’s analysis of Virginia’s October 2026 SCC report calculates that permitted generator MW is 1.52 times reported service-agreement or equivalent contracted-demand MW.
- DOE’s transformer supply analysis reports distribution transformer lead times of 3 to 6 months in 2019.
- DOE’s transformer supply analysis reports distribution transformer lead times of 12 to 30 months in 2023.
- DOE’s supply analysis, reviewed in November 2026, reports identifying over 80,000 distribution transformer varieties nationwide.
- EIA’s October 2026 Short-Term Energy Outlook forecasts 2.8% growth in total U.S. commercial-sector electricity demand in 2027.
- EIA’s May 2026 AEO discussion estimates that servers alone accounted for 7% of U.S. commercial-sector electricity consumption in 2025.
- LBNL’s June 2026 report estimates national average data center power usage effectiveness of 1.45 in 2024.
- LBNL’s June 2026 Reference Case projects national average data center power usage effectiveness of 1.36 in 2030.
How much electricity do data centers use in the US?
The newest LBNL historical estimate is 192 TWh for 2024, published in June 2026, with a national electricity share of 4.7%.
The title can cause confusion. The document is called the United States Data Center Energy Usage Report: 2025 Update, but its publication date is June 2026. Its historical period ends in 2024, and the report explains that even its 2025 results contain projections because the underlying IT shipment data extend through only part of that year. Call the 2024 result a modeled historical estimate when describing how it was produced.
LBNL builds electricity consumption from equipment shipments, equipment energy assumptions and facility infrastructure. Its scope includes servers, storage, networking and infrastructure electricity; cryptocurrency mining is excluded. That boundary matters when another report combines computing facilities with mining operations. The report’s overview and results describe the covered equipment and years.
For a 2026 article, publication year and observation year belong beside each other. Reprinting a historical value under a new year changes what the reader thinks was counted. A sentence such as “data centers use 4.7% of U.S. electricity in 2026” would misdate this finding.
EIA uses another boundary in its May 2026 server analysis: servers alone accounted for an estimated 7% of commercial-sector electricity in 2025. That leaves out the broader national denominator used by LBNL and does not describe all facility loads. Put both descriptions in a research note before comparing percentages. The commercial building statistics reference gives the wider building-stock context.
Data center electricity forecasts through 2030
LBNL’s 2030 Reference Case reaches 649 TWh, while its compounded uncertainty cases span 521 to 843 TWh of annual U.S. data center electricity use.
PorchRocket converts each published energy result into an equivalent annual average load. The calculation is TWh multiplied by 1,000, divided by 8,760 hours in 2030. This puts the scenarios into a power unit without assuming a connection utilization rate. Every row below uses LBNL Table 2; the average GW and growth columns are PorchRocket calculations.
| 2030 scenario | Annual electricity, TWh | Published U.S. electricity share | Calculated average load, GW | Calculated growth from 2024 electricity |
|---|---|---|---|---|
| Reference Case | 649 | 11.8% | 74.1 | 238.0% |
| Consolidated Deployment | 578 | 10.5% | 66.0 | 201.0% |
| High ASIC | 664 | 12.1% | 75.8 | 245.8% |
| Low Accelerator Lifetime | 590 | 10.7% | 67.4 | 207.3% |
| High Inference Energy | 782 | 14.2% | 89.3 | 307.3% |
| Compounded uncertainty, low | 521 | 9.5% | 59.5 | 171.4% |
| Compounded uncertainty, high | 843 | 15.3% | 96.2 | 339.1% |
The scenario names describe different equipment and operating assumptions. Consolidated Deployment reduces accelerator shipments; Low Accelerator Lifetime changes how long accelerators remain in use. High Inference Energy changes assumptions about idle power and utilization. The compounded cases combine uncertainties toward their extremes. They are scenario bounds, not a statistical confidence interval.
I would use the Reference Case for a first territory plan and test whether the plan survives the low case, since the branch still needs reachable accounts and qualified technicians if national growth falls below the central forecast. Local purchasing needs a separate check. The forecast does not identify which campus will have a maintenance opening next quarter.
MW, MWh and the difference between load and capacity
EIA defines MW as power and MWh as energy over time; converting annual consumption to average MW does not establish a facility’s rated capacity.
A MW value needs a label. It might describe average demand, peak demand, IT equipment capacity, utility connection rights or generator nameplate capacity. The EIA electricity measurement explanation establishes the unit distinction; the capacity boundary must come from the document containing the figure.
| Measure | What the number describes | What to collect alongside it |
|---|---|---|
| Annual MWh | Electricity consumed during the reporting year | Meter boundary and covered months |
| Annual average MW | Annual MWh divided by hours in that year | Calendar year and calculation method |
| Peak MW | Highest demand within the defined measurement interval | Interval length and whether it is measured or forecast |
| IT capacity MW | Equipment capacity within the stated IT boundary | Installed, available or planned status |
| Interconnection MW | Capacity requested or established at the utility connection | Agreement stage and energization schedule |
| Permitted generator MW | Generator capacity described in a permit record | Permit date, equipment schedule and installation evidence |
Say a facility averages 100 MW of total electrical load throughout 2026. It would consume 100 multiplied by 8,760, or 876,000 MWh, equal to 876 GWh. The arithmetic says nothing about its generator count. A facility’s permitted backup capacity can cover redundancy and equipment arrangements that annual energy does not reveal.
Calendar treatment also matters. PorchRocket uses 8,784 hours for the historical 2024 conversion because it was a leap year, and 8,760 for 2030. The resulting national average loads round to 21.9 GW and 74.1 GW. Subtracting the unrounded values gives an increase of 52.2 GW after rounding. Carry “annual average” into any chart label that uses these results.
Data center capacity by state and the Virginia record
PorchRocket’s analysis of Virginia’s October 2026 SCC report totals 21,791.4 MW of service agreements or equivalent contracted demand across five utility entries.
The state report places those entries under a heading covering electric service agreements or equivalent contracted demand, including electric load. PorchRocket retains that description. The document does not identify their sum as a statewide metered peak, an installed IT capacity total or electricity consumed during the year. The October 1 report, page 2, is the source for the aggregation.
That gives a state-level reference. For a national ranking, I would require matching reporting dates and definitions across states before comparing requested connections in one state with another state’s operational IT capacity, since those figures answer different questions about the equipment and demand. Keep the original metric labels.
For a territory workbook, I would keep separate fields for operating equipment and expansion evidence. Record an agreement in its own column and give the proposed phase its own status. Attach the dated document. Then make a second check against the site, rather than treating an agreement as proof that all the associated equipment is installed.
The same discipline applies to plants supplying the power. A generation plant record describes supply equipment. It does not automatically identify the downstream customer or the customer’s demand. The solar and storage sites by state reference covers those generating assets separately.
This edition omits a 50-state data center capacity ranking. Its reproducible state calculations cover the Virginia report’s defined aggregates, which can be checked directly against the original document.
Virginia data center generator statistics by emissions tier
PorchRocket’s analysis of Virginia’s October 2026 SCC report totals 13,767 permitted generator sets and 33,132 MW, with Tier 2 or earlier equipment contributing most capacity.
PorchRocket adds the two tier groups and divides each group’s count and MW by its corresponding total. Mean MW per set is group capacity divided by group count. All source quantities come from SCC report RD637, item iii; percentages and means are PorchRocket calculations.
| Reported permitted generator group | Generator sets | Share of sets | Capacity, MW | Share of MW | Mean MW per set |
|---|---|---|---|---|---|
| Tier 2 or earlier | 10,290 | 74.7% | 26,718 | 80.6% | 2.60 |
| Tier 4 or Tier 4-equivalent | 3,477 | 25.3% | 6,414 | 19.4% | 1.84 |
| Total | 13,767 | 100.0% | 33,132 | 100.0% | 2.41 |
Unit share and capacity share answer different questions. A maintenance business planning how many sets require documentation will care about count. A supplier reviewing the scale of permitted power equipment may start with MW. The Tier 4 grouping has a larger share of sets than of capacity in this report because its mean permitted MW per set is lower.
Do not use that average to assign a rating to a particular enclosure. The report gives aggregate groups, not the equipment schedule at the address you are researching. Confirm the rating and engine details from the site’s documentation before selecting parts or estimating a test scope.
For a sales plan, confirm the installed equipment and the permit governing the proposed work. Then identify who holds the maintenance agreement and when the owner reviews that scope. Those answers decide whether the site belongs on your near-term call list. The standby generator inventory guide explains the equipment-level search.
What permitted generator capacity tells a service company
The two Virginia aggregates have a ratio of 1.52, calculated by PorchRocket from permitted generation MW divided by reported service-agreement or equivalent contracted-demand MW.
The arithmetic is 33,132 divided by 21,791.4. It is not a redundancy factor for the typical campus: the SCC report does not match each generator group to the individual sites’ agreements or measured loads, and dividing statewide totals cannot supply those missing relationships.
For equipment sales, a permit is an evidence lead. It gives your researcher a document to inspect and a site to qualify. For recurring service, I would ask for installation evidence and the current maintenance responsibility before counting an address as an account your team can pursue. Give each stage its own status field.
Runtime needs another field. Generator capacity alone does not state annual operating hours, fuel consumed or testing performed. A site with backup equipment does not necessarily buy the same amount of fuel as another site with a similar equipment rating. Ask the facilities team for operating records and the approved maintenance scope.
The distinction also protects territory estimates from double counting. A campus may appear in a land-use record, an equipment permit and a utility agreement. Keep the source records, but connect them to the same location and phase when the evidence supports the match. I would count the service account only after identifying who approves its work.
Transformer lead times in 2026
Equipment delays still span multiple years according to DOE’s October 26, 2026 update; the agency’s numerical distribution transformer benchmarks describe earlier periods.
The dates are part of the statistic. DOE’s supply chain analysis says distribution transformer lead times moved from 3 to 6 months in 2019 to 12 to 30 months in 2023. Relabeling the latter range “2026 lead times” would misstate the source.
| DOE evidence | Period described | What it supports for a 2026 buyer |
|---|---|---|
| Distribution transformer lead times of 3 to 6 months | 2019 | Historical comparison before the reported supply disruption |
| Distribution transformer lead times of 12 to 30 months | 2023 | A dated numerical benchmark for distribution equipment |
| Over 80,000 distribution transformer varieties identified | DOE supply analysis reviewed November 2026 | Specification variety is one issue DOE is addressing |
| Multi-year lead times for power and distribution transformers | October 26, 2026 update | Supply constraints remain a current planning issue |
The table combines the DOE supply analysis with its October 2026 equipment shortages update, which identifies continuing material and manufacturing constraints without publishing a universal order-to-delivery range that a buyer could apply to every transformer class.
I would get a dated schedule for the specified unit before promising a commissioning window. Ask where the clock starts: purchase order or approved engineering. Record factory testing and site delivery as separate milestones, then check the supplier’s available production slot against your schedule before relying on a national historical average that describes another specification or an earlier market.
For service territory planning, long procurement schedules also justify keeping an operating-equipment list alongside the expansion list. One supports maintenance outreach; the other supports a later installation conversation. The substation and transformer location guide covers the search fields to carry into that work.
Cooling electricity and power usage effectiveness
DOE defines power usage effectiveness as annual total facility energy divided by annual IT energy, so a PUE of 1.45 includes more than cooling.
The numerator covers the facility boundary. The denominator covers IT equipment. Cooling is part of the difference, along with other infrastructure consumption. DOE’s annual-energy PUE definition is the basis for the comparison below. A cooling contractor should ask for cooling-system measurements before assigning all non-IT electricity to chillers or fans.
For a fixed-load illustration, assume average IT demand of 38 MW throughout 2026 and apply the two national PUE values in LBNL’s report as comparison inputs, holding the IT demand constant to isolate the arithmetic. PUE 1.45 is the report’s modeled 2024 national average; 1.36 is its 2030 Reference Case average.
| Assumed facility performance | Average IT load | Calculated total facility load | Calculated non-IT load | Calculated annual facility electricity |
|---|---|---|---|---|
| PUE 1.45 | 38 MW | 55.10 MW | 17.10 MW | 482,676 MWh |
| PUE 1.36 | 38 MW | 51.68 MW | 13.68 MW | 452,716.8 MWh |
PorchRocket’s calculation multiplies 38 MW by the assumed PUE and then by 8,760 hours. The annual difference is 29,959.2 MWh at fixed IT demand. For a retrofit discussion, replace the comparison inputs with the owner’s metered IT energy and measured facility performance before estimating savings.
National electricity can still grow while average PUE improves. LBNL’s forecast includes both outcomes because the modeled IT load grows. For contractors, I would separate the questions: how much equipment the account has, which systems your crew can maintain, and what the owner has documented about performance.
For adjacent building campaigns, the rooftop HVAC building list and visible storage tank list cover those property features. A data center’s specialized cooling or power-equipment requirement belongs in the custom search scope. The rooftop unit statistics page provides the broader commercial HVAC reference.
What changed for Virginia generator permits in 2026?
Virginia Code section 10.1-1322.6 applies a Tier 4-equivalent emissions benchmark to data center air permit applications submitted on or after July 1, 2026.
The 2026 statutory text covers DEQ air permits, including general permit coverage, and bars issuance unless each gen-set’s established emission limit is equal to or below the emissions achieved by a Tier 4-equivalent gen-set. The section also preserves authority to impose more stringent limits. Application date is the trigger specified in that provision.
The law defines Tier 4-equivalent equipment by its emissions controls: selective catalytic reduction for nitrogen oxides, a diesel oxidation catalyst for carbon monoxide, and a diesel particulate filter for particulate matter, with equivalent controls allowed. Compare those requirements with the approved permit and equipment documentation when qualifying the work.
The implementing regulation is current for this edition: Virginia’s Action 7091 record states that its final stage became effective November 4, 2026. Keep the regulation’s effective date separate from the statute’s application-date trigger when recording a project’s permit history.
The October generator totals describe the reported permitted fleet, spanning the two tier groups. They do not make the earlier group a ready-made replacement backlog. I would record the applicable permit and the owner’s proposed work before deciding which equipment belongs in a sales forecast. A fleet percentage does not establish what a particular owner must buy.
How to map a data center equipment territory, step by step
Build one equipment-and-account file for a defined territory, then use PorchRocket’s national and state statistics to check scale rather than assign demand to individual addresses.
- Set the service boundary. Pick the counties or travel radius your crew can cover and write down the equipment classes you serve. For a generator department, that might mean exterior standby generator enclosures at operating campuses. Put response coverage and the work your technicians can perform in the scope before collecting addresses.
- Choose the evidence fields. Give each site a location, parcel reference and observation date. Add separate columns for observed equipment, permit status, installation evidence and operating status. Preserve the original source beside each field so a researcher can distinguish what was seen from what was reported.
- Commission the territory search in Asset Search. Submit the generator, transformer or substation criteria and geography through the custom asset search service. Ask for a scoped pilot with reviewed candidates, dated views, owner information and dispositions. Define what counts as an accepted result before extending the order.
- Join the public records. Match permit and planning records to the reviewed locations. Keep a proposed campus phase distinct from an operating phase. When records appear to overlap, check the location and equipment description before combining them. Preserve unresolved matches for review instead of silently adding their capacity.
- Qualify the account with the facilities team. Confirm who approves maintenance work and which company holds the current agreement. Ask for equipment ratings, service requirements and the next purchasing window. The asset inventory hub explains how a reviewed site becomes a maintained territory record.
- Build an offer around the equipment you serve. Request an authorized scope review for maintenance, testing support or the other work your department performs. Have the facilities contact identify site access and documentation requirements before dispatching a technician. Count a proposal only when there is a defined scope and an approval path.
- Budget the owner’s account value. Use expected gross profit and the likely paid service period to decide how much research and outreach the account can support. Enter your assumptions in the customer lifetime value calculator. Keep delivery costs and acquisition costs in their respective calculations.
- Refresh the territory on a decision schedule. I would review a sales territory quarterly and use a separate change watch where installations are moving quickly. Record the date an expansion becomes visible and the date the owner confirms a purchasing need. Report accounts qualified and proposals issued alongside the number of sites researched.
The value of one power equipment service account
A service account producing $21,600 in annual gross profit is worth $86,400 over four paid years in the worked example below, before acquisition expense.
Say an electrical service department targets multi-equipment maintenance accounts around Henrico County, Virginia. Assume one signed account brings $67,500 in annual service revenue, with $45,900 in annual delivery costs including field labor, parts and travel. Assume four paid service years with unchanged annual amounts. This is the account model used in the remaining calculations.
| Account calculation | Arithmetic | Result |
|---|---|---|
| Annual gross profit | $67,500 revenue minus $45,900 delivery cost | $21,600 |
| Gross margin | $21,600 divided by $67,500 | 32% |
| Monthly gross profit | $21,600 divided by 12 | $1,800 |
| Four-year gross profit | $21,600 multiplied by 4 | $86,400 |
Customer lifetime value, or CLV, is the gross profit expected from the account over its paid life. Here the calculation uses four years without discounting or price changes. The account value comes from the proposed scope and expected costs. Replace those inputs with your service department’s estimate before setting its pursuit budget.
I would check the travel and access assumptions first, because the paid time a crew needs to service the account belongs in delivery cost even when the estimate did not allow enough hours for the work. Extra sales qualification hours belong in acquisition cost. Moving an expense between columns will not improve the margin.
This is why building fit matters before the call. An account with equipment your team supports and a route you can cover deserves a different pursuit budget from an address that merely has a large power figure in a report. The installed-base mapping guide connects equipment territory work with that account decision.
A territory budget with low, base and high outcomes
Two signed accounts produce $172,800 in four-year gross profit in this model, against $8,640 of internal research and outreach labor before separately quoted vendor fees.
Allocate 144 staff hours at an assumed loaded cost of $60 an hour to research, outreach and qualification across 180 sites. That is $8,640 of staff labor. Hold the same hours and account economics across the scenarios. The only changing input is the number of accounts signed. PorchRocket carries the preceding account model into each scenario. For payback, assume the annual gross profit arrives evenly over the paid service months and measure from the start of service; the sales period comes before that clock. Conversion percentages are rounded to one decimal place.
| Territory outcome | Low | Base | High |
|---|---|---|---|
| Sites researched | 180 | 180 | 180 |
| Accounts signed | 1 | 2 | 3 |
| Sites becoming accounts | 0.6% | 1.1% | 1.7% |
| Internal staff labor | $8,640 | $8,640 | $8,640 |
| Staff labor per account signed | $8,640 | $4,320 | $2,880 |
| Four-year gross profit from signed accounts | $86,400 | $172,800 | $259,200 |
| Gross profit less staff labor | $77,760 | $164,160 | $250,560 |
| Gross profit divided by staff labor | 10 times | 20 times | 30 times |
| Months to recover staff labor from gross profit | 4.8 | 2.4 | 1.6 |
The base case requires 2 accounts from 180 researched sites. It earns twice the account gross profit of the low case with the same allocated staff hours. That is the financial reason to select equipment and service fit before paying people to work an address list. The acquisition assumptions stay modest; the recurring account value creates the larger result.
For an all-in customer acquisition cost, add the accepted research quote and other acquisition spending to staff labor, then divide by accounts signed. Use that total in the lead-list ROI calculator. The gross-profit-to-labor multiples divide four-year gross profit by staff spending. Net return subtracts acquisition spending first. For the full campaign, use (four-year gross profit minus total acquisition spending) divided by total acquisition spending. Enter the accepted vendor quote in that spending total.
My decision rule is to approve the next territory only when the low case covers its full acquisition budget and the delivery margin still works. Track signed accounts against all researched sites to replace the table’s conversion assumption with your own result. Also track conversations, qualified scopes and proposals separately, using each preceding stage as the denominator. That shows whether the next improvement belongs in account selection, outreach or estimating.
Methodology
PorchRocket’s November 2026 analysis combines LBNL energy scenarios and Virginia SCC equipment aggregates with dated EIA electricity context and DOE transformer benchmarks.
The LBNL version is report LBNL-2001758, United States Data Center Energy Usage Report: 2025 Update. The input quantities are its published 2024 historical estimate, 2028 Reference Case value, 2030 Table 2 scenarios and stated PUE values. The PDF identifies a CC BY 4.0 license. PorchRocket downloads the published PDF and reads its numerical tables; it does not redistribute the proprietary shipment microdata underlying LBNL’s model.
The Virginia version is RD637, Report on Information Concerning Data Centers, dated October 1, 2026. The generator filter includes both groups under item iii: Tier 2 or earlier, and Tier 4 or Tier 4-equivalent. The demand aggregation includes the five utility values under item i, retaining that item’s service-agreement or equivalent contracted-demand definition. No operator-level or address-level capacity is inferred.
The reproduction script, compute.py, saves input URLs and file hashes with the downloaded source documents. It extracts the published numbers and writes the statistics CSV, including source, year, geography and calculation method on every row. The principal formulas are:
- Annual average GW = annual TWh multiplied by 1,000, divided by the calendar year’s hours.
- Growth from 2024 = scenario TWh divided by 192 TWh, minus 1, multiplied by 100.
- Tier share = group count or MW divided by its corresponding Virginia total, multiplied by 100.
- Mean permitted generator MW = group MW divided by group generator count.
- The cross-total Virginia ratio = permitted generator MW divided by reported agreement or equivalent contracted-demand MW.
- The service model = stated annual revenue less stated annual delivery cost, multiplied by paid years, with staff acquisition labor calculated separately.
The CSV covers national energy scenarios, Virginia aggregates and the worked models; DOE’s transformer benchmarks are cited directly in the article. The script retains more precision than the displayed tables, which round shares and average loads. It uses the source’s published national electricity shares rather than substituting another denominator. EIA’s October 2026 outlook and May 2026 server discussion provide separate sector context; DOE’s transformer statements retain their original reference dates.
These are published model outputs and aggregate permit records, not a site-level installed-equipment census. Forecasts retain their scenario labels, and the Virginia count is not extrapolated nationwide. The service-account figures are labeled assumptions and reproduce separately from the public-source calculations.
How to cite these statistics
Credit PorchRocket for the calculated average loads, Virginia aggregates and tier shares, and name LBNL or Virginia SCC when using their underlying published estimates.
Suggested citation: PorchRocket, “Data Center Power Statistics 2026: 30 Numbers With Sources,” 2026, updated November 5, 2026. https://porchrocket.com/data-center-power-statistics
You may reuse PorchRocket’s calculations with a link back to this page. Download the data center power statistics CSV for the individual values, units and methods. Keep the source year and geography with a copied row, including its forecast or permitted-equipment label.
For example: “PorchRocket’s analysis of Virginia SCC’s October 2026 report totals 13,767 permitted data center generator sets.” For a national forecast: “LBNL’s June 2026 Reference Case projects 649 TWh of U.S. data center electricity use in 2030.” Those sentences preserve who produced the figure and what period it describes.
Building a power equipment territory with Asset Search
Asset Search finds visible generator enclosures, substations and related equipment in your chosen territory, then returns reviewed site evidence for your account research. Tell us what to find and where in the contiguous United States, then start with a small paid pilot credited toward the full order. I would use per-accepted-result terms to build a service territory, then a quarterly monitoring subscription to track new equipment. The file includes location and parcel, owner information, dated views and observed facts, with fits, rejects and unresolved cases carrying reasons or the field check needed. Choose CSV, Excel or GIS-ready delivery. The search cross-references imagery with property and public records and includes human review. Give your sales team a territory built around the equipment it can serve.
Quick answers
How much electricity do U.S. data centers use?
About 192 TWh in 2024 is LBNL's updated historical estimate, published in June 2026. That equals 4.7% of U.S. electricity consumption. Its 2030 Reference Case reaches 649 TWh. Keep the historical year attached when citing the first figure and identify the second as a forecast.
What percentage of U.S. electricity will data centers use in 2026?
The evidence presented here does not establish an observed full-year 2026 percentage. LBNL's June 2026 report puts the historical 2024 share at 4.7% and projects 29% growth in data center electricity consumption from 2025 to 2026 in its Reference Case. That growth rate is not a national electricity share.
How many backup generators do data centers have?
PorchRocket's analysis totals 13,767 permitted generator sets in Virginia's October 1, 2026 public report. The report covers two emissions tier groups with a combined permitted capacity of 33,132 MW. This is a Virginia permit measure; it does not establish the number installed, their annual runtime or a U.S. total.
Can generator capacity tell me a data center's electricity use?
No. Generator MW describes an equipment capacity, while electricity consumption requires energy measured over time. Backup systems can include redundancy and idle equipment. Use metered MWh for consumption, keep permitted generator MW in an equipment field, and obtain the site's operating information before estimating fuel or maintenance demand.
How long are transformer lead times in 2026?
DOE's October 26, 2026 update still describes multi-year delays for power and distribution transformers. Its separate 12 to 30 month distribution transformer range refers to 2023. Neither statement is a delivery quote for a particular specification. Obtain a dated supplier schedule covering engineering approval, testing, shipment and site delivery.
Why does EIA report a different data center electricity percentage?
Different coverage and denominators can produce different percentages. EIA's AEO2026 discussion estimates servers alone used 7% of commercial-sector electricity in 2025. LBNL's 4.7% figure covers broader data center facility electricity and uses total U.S. electricity in 2024 as its denominator. The two statements describe different measures.
Where can I find data center capacity by state?
Start with public utility commission reports and identify the definition before comparing states. Virginia's October 2026 report provides utility service agreements or equivalent contracted demand; PorchRocket sums those figures to 21,791.4 MW. A state ranking requires comparable dates, operating status and capacity boundaries, which this dataset does not supply.
How can a service company find nearby data center power equipment?
Define the generator enclosures, substations or other visible equipment your crew serves and choose the territory. Asset Search cross-references imagery with property and public records, reviews candidates and returns dated evidence with owner information. Start with a scoped pilot, then confirm equipment ratings and access requirements with the facilities team.
Sources used in this guide
Sources are linked at the claim they support and collected here for auditability. Vendor features and prices can change; verify them before purchasing.
- United States Data Center Energy Usage Report: 2025 UpdateLawrence Berkeley National Laboratory and U.S. Department of Energy: June 18, 2026 report, LBNL-2001758; historical use, Table 2 scenarios, forecast scope and PUE; released under CC BY 4.0. Checked November 5, 2026.
- Report on Information Concerning Data Centers, RD637Virginia State Corporation Commission: October 1, 2026 report; five utility demand figures and two permitted generator tier groups, on page 2. Checked November 5, 2026.
- Supply Chain and Market AnalysisU.S. Department of Energy, Office of Electricity: Distribution transformer lead times for 2019 and 2023 and over 80,000 transformer varieties; the historical dates are retained. Checked November 5, 2026.
- Innovative Solutions for Grid Equipment ShortagesU.S. Department of Energy, Office of Electricity: October 26, 2026 update confirming multi-year transformer lead times and continuing material and manufacturing constraints. Checked November 5, 2026.
- Data center server energy use grows across the commercial building stockU.S. Energy Information Administration: May 19, 2026 AEO explanation; servers-only consumption and the commercial-sector denominator. Checked November 5, 2026.
- Short-Term Energy Outlook, October 2026U.S. Energy Information Administration: Archived October release; 2027 commercial electricity demand growth forecast, page 11. Checked November 5, 2026.
- Measuring electricityU.S. Energy Information Administration: Definitions of power, energy and electricity units used in the conversions. Checked November 5, 2026.
- Cooling Water Efficiency Opportunities for Federal Data CentersU.S. Department of Energy, Federal Energy Management Program: Annual-energy PUE definition and the distinction between IT, cooling and other facility loads. Checked November 5, 2026.
- Code of Virginia section 10.1-1322.6, 2026 updateVirginia General Assembly: Application-date trigger of July 1, 2026, Tier 4-equivalent definition and permit emission limit requirement. Checked November 5, 2026.
- Virginia Regulatory Town Hall, Action 7091Virginia Department of Environmental Quality: Implementing regulation for Chapter 397; final stage became effective November 4, 2026. Checked November 5, 2026.
Find the power equipment your team can service.
Define the equipment and territory. Get a scoped Asset Search pilot with reviewed sites, dated evidence and owner information in your team's format.