AI-optimized servers will use 175 TWh of electricity in 2026, up from 95 TWh in 2025, according to Gartner’s June 2026 forecast. The IEA reports that AI-focused data centers raised their electricity use 50% in 2025.
This page covers AI datacenter power consumption statistics by hardware segment, campus, country and 2030 forecast, using releases published between April and October 2026.
AI Datacenter Power Consumption Statistics
How Much Power Do AI Servers Use in 2026?
Gartner forecasts AI-optimized servers will draw 175 TWh in 2026, or 31% of the 565 TWh used by data centers worldwide. That is an 84.2% increase on 95 TWh in 2025.
Conventional servers rise 1.2% to 195 TWh over the same year. In Gartner’s forecast, AI-optimized hardware passes them in 2027, at 258 TWh against 200 TWh.
| Segment | 2025 (TWh) | 2026 (TWh) | 2027 (TWh) | 2026 growth |
|---|---|---|---|---|
| Conventional Server | 193 | 195 | 200 | 1.2% |
| AI-Optimized Server | 95 | 175 | 258 | 84.2% |
| Cooling and other Infrastructure | 159 | 195 | 243 | 22.6% |
| All Data Center Power Consumption | 447 | 565 | 702 | 26.4% |
Source: Gartner, Data Center Electricity Consumption by Segment, Worldwide, June 2026 forecast (segments may not sum to totals because of rounding)
Cooling and other infrastructure reach 195 TWh in 2026, the same as the conventional server line. Operating system splits for that conventional fleet appear in our Linux server market share by industry sector data.
Are AI-Focused Data Centers Growing Faster Than the Rest?
Yes. Electricity use at AI-focused data centers rose 50% in 2025, against 17% for all data centers and 3% for global electricity demand.
The IEA puts total 2025 data center use at 485 TWh. Gartner’s separate model puts it at 447 TWh, so the two figures sit side by side here rather than combined.
Source: IEA, Key Questions on Energy and AI, April 2026; Gartner, June 2026
AI Datacenter Power Consumption by Campus, October 2026
| AI data center | Owner | Location | Estimated IT power (MW) |
|---|---|---|---|
| Anthropic-Amazon New Carlisle | Amazon | New Carlisle, Indiana | 1,037 |
| Colossus 2 | SpaceXAI | Memphis, Tennessee | 946 |
| Microsoft Fairwater Atlanta | Microsoft | Fayetteville, Georgia | 636 |
| DayOne Nusajaya | Unknown | Johor Bahru, Malaysia | 473 |
| Meta Prometheus | Meta | New Albany, Ohio | 471 |
Source: Epoch AI, AI data centers hub, updated October 20, 2026 (IT power capacity estimates)
Anthropic-Amazon New Carlisle is now the largest AI data center Epoch tracks by IT power, at 1,037 MW. Earlier versions of the same Epoch page listed it at 910 MW, behind Colossus 2.
Colossus 2 still leads on compute, at an estimated 1,112k H100-equivalents. Software stacks for clusters like these are covered in our Linux ML and AI infrastructure statistics.
Epoch’s 95 tracked sites hold 14.2 GW of IT power and, by its estimate, 44% of global deployed AI compute. Cooling and other overhead lift their footprint to 18.5 GW, above New York City’s 11 GW peak demand.
These are capacity estimates, not energy use; Epoch says average draw typically runs at 60% to 80% of capacity. AI inference outside large campuses shows up in our edge computing deployment figures.
Ireland’s Data Centres Took 23% of Metered Electricity in 2025
Data centres in Ireland used 7,663 GWh in 2025, up 10% from 6,973 GWh in 2024. Consumption by all other metered customers grew 2%.
The 23% share exceeds urban households at 18% and rural households at 9%, though combined residential use remains higher at 28%. Quarterly data centre use climbed from 291 GWh in Q1 2015 to 1,991 GWh in Q4 2025.
| Year | Data Centre use (GWh) | % Data Centre of metered total |
|---|---|---|
| 2021 | 4,012 | 14% |
| 2022 | 5,273 | 18% |
| 2023 | 6,339 | 21% |
| 2024 | 6,973 | 22% |
| 2025 | 7,663 | 23% |
Source: Central Statistics Office Ireland, Data Centres Metered Electricity Consumption 2025, released July 7, 2026
The CSO identifies data centres from meter data and does not separate AI sites. These totals cover all data centre activity in Ireland.
What Will U.S. Data Centers Draw by 2030?
Lawrence Berkeley National Laboratory’s June 2026 update puts U.S. data centers at 11.8% of national electricity in 2030. Its reference case equals 649 TWh.
Scenario bounds run from 521 TWh to 843 TWh, or 9.5% to 15.3% of U.S. electricity. Higher AI server idle power and utilization gives the largest of four single-factor results, at 782 TWh.
| Berkeley Lab scenario | U.S. data center use in 2030 (TWh) |
|---|---|
| Low Compounded Uncertainty bound | 521 |
| Lower forecast IT equipment installations | 578 |
| Shorter assumed AI chip lifetime | 590 |
| Reference Case | 649 |
| More specialized graphics chips installed | 664 |
| Higher AI server idle power and utilization | 782 |
| High Compounded Uncertainty bound | 843 |
Source: Lawrence Berkeley National Laboratory, United States Data Center Energy Usage Report: 2025 Update, June 2026
The lab’s 2024 report had put the range at 6.7% to 12.0% of U.S. electricity, with 2028 as its end year.
IEA vs Gartner: Two AI Datacenter Power Consumption Forecasts for 2030
The IEA projects global data center use rising from 485 TWh in 2025 to 950 TWh in 2030, around 3% of world electricity demand. AI-focused data centers triple their consumption over the same period.
Bottlenecks across the value chain make faster near-term scenarios less likely, the agency says. It also flags possible upside after 2030.
Source: IEA, Key Questions on Energy and AI, April 2026
Gartner expects consumption above 1,200 TWh by 2030 and says grid supply will fall short of planned construction. Its power demand forecast climbs from 104 GW in 2025 to 132 GW in 2026 and 290 GW in 2030.
Source: Gartner, June 2026
The two outlooks start from different 2025 baselines, 485 TWh and 447 TWh. Each uses its own model, so neither revises the other.
Why Do AI Racks Strain the Grid?
AI server power density rose 11 times between 2020 and 2025, and the IEA expects another fourfold rise by 2027. By then, one refrigerator-sized rack could carry the peak demand of 65 households.
Server load in AI data centers can swing by more than 50% of rated capacity within one second. The IEA sees 20 to 25 GW of battery storage potentially installed in data centers by 2030 to manage those swings.
Slow grid connections have pushed U.S. developers toward onsite gas. About one-fifth of tracked projects have started land clearing or construction.
Reliable onsite supply needs 30% to 70% more generation than the load it serves. Around 15 to 27 GW of onsite gas could power data centers by 2030, mostly in the United States.
Gas turbine orders rose 70% in 2025, and a high-bandwidth memory shortage is expected to last through at least the end of 2027. Capital spending by five large technology companies topped $400 billion in 2025, with a further 75% rise expected in 2026.
Tech firms signed around 40% of corporate renewable power purchase agreements in 2025. Conditional offtake deals with small modular reactor projects grew from 25 GW at the end of 2024 to 45 GW by April 2026.
Source: IEA, Key Questions on Energy and AI, April 2026
Does One AI Query Use Much Electricity?
A simple AI text query now uses less electricity than running a television for the same length of time, the IEA reported in April 2026. Energy per AI task has fallen by at least an order of magnitude each year in recent years.
Video generation, reasoning and agentic tasks can use hundreds or thousands of times more energy per query. Company rollouts of agentic tools appear in our enterprise LLM adoption numbers.
Running every conventional web search as a simple AI text query would use under 4 TWh a year. That equals less than 1% of current data center consumption, per the IEA.
Source: IEA, Key Questions on Energy and AI, April 2026
Power Equipment Sales Tied to AI Datacenter Power Consumption
The figures below measure company revenue, not electricity. They track spending on chips, power gear and onsite generation that set or serve data center load.
NVIDIA reported Data Center revenue of $89.0 billion for the quarter ended July 26, 2026, up 117% from a year earlier. AI accelerator vendor shares are compared in our NVIDIA vs AMD vs Intel AI GPU breakdown.
Source: NVIDIA, second quarter fiscal 2027 results, August 26, 2026
Vertiv, a critical digital infrastructure supplier, posted Q2 2026 net sales of $3,274 million, up 24%. It guides to $14,000 million for 2026, with 31% organic growth at the midpoint.
Source: Vertiv, second quarter 2026 results, July 29, 2026
Bloom Energy reported Q2 2026 revenue of $1,065.4 million, up 165.5%, with product revenue up 215.4%. The fuel cell maker raised its 2026 revenue guidance to $3.9 billion to $4.2 billion.
Source: Bloom Energy, second quarter 2026 results, July 28, 2026
FAQs
How much electricity do AI data centers use?
Gartner forecasts AI-optimized servers will use 175 TWh in 2026, or 31% of the 565 TWh consumed by data centers worldwide. Cooling and other infrastructure, counted separately, reach 195 TWh.
What is the biggest AI data center by power?
Anthropic-Amazon New Carlisle in Indiana, at an estimated 1,037 MW of IT power capacity, per Epoch AI’s October 20, 2026 update. Colossus 2 in Memphis follows at 946 MW.
How much energy does one AI query use?
Less than running a television for the same length of time, for a simple text query, per the IEA’s April 2026 report. Video, reasoning and agentic tasks can use hundreds or thousands of times more.
What percentage of U.S. electricity will data centers use by 2030?
Berkeley Lab’s June 2026 reference case puts it at 11.8%, or 649 TWh. Its scenario range runs from 9.5% to 15.3% of total U.S. electricity use.
Do Reddit claims that AI will overload the power grid match the data?
It depends on scale. The IEA projects data centers at around 3% of global electricity demand by 2030, but Ireland’s data centres already used 23% of metered electricity in 2025, per the CSO.
Sources
https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary
https://epoch.ai/data/ai-data-centers
https://eta-publications.lbl.gov/publications/united-states-data-center-energy-2025
https://www.aboutchromebooks.com/ai-energy-consumption-statistics/
