AI Data Center Energy Consumption Statistics 2026: The Definitive Data Report

AI Data Center Energy Consumption Statistics 2026

Last updated: June 5, 2026 | Next scheduled update: Q3 2026 (September) By Axis Intelligence Research + Sarah Mitchell | Download CSV ↓

Global data center electricity consumption reached approximately 460–490 TWh in 2025 — a 17% year-over-year surge that outpaced overall global electricity demand growth of 3% — with AI-focused facilities alone driving a 50% spike in that same period. By 2030, the International Energy Agency projects total data center demand will double to roughly 945 TWh, while AI-specific consumption is on track to triple. The five largest hyperscalers (Amazon, Alphabet, Microsoft, Meta, Oracle) committed $660–690 billion in capital expenditure in 2026 alone, with approximately 75% directly tied to AI infrastructure — the largest single-cycle infrastructure investment in private-sector history.

---

Key Findings

• Global data center electricity hit 460–490 TWh in 2025, a 17% increase year-over-year, equivalent to France’s annual national electricity consumption (IEA, April 2026).
• AI-focused data centers drove a 50% energy surge in 2025 — growing at more than five times the rate of overall global electricity demand growth of 3% (IEA, April 2026).
• U.S. data centers consumed 176 TWh in 2023 (4.4% of total U.S. electricity), with demand projected to reach 325–580 TWh by 2028 — a potential 6.7%–12% share of the national grid (DOE / Lawrence Berkeley National Laboratory, December 2024).
• AI training compute requirements are doubling annually: the power required to run the leading AI supercomputer (xAI’s Colossus) has grown from 13 MW in 2019 to 280 MW in 2025 — more than 20× in six years (Epoch AI, June 2025).
• The five largest hyperscalers are committing $660–725 billion in 2026 capex, with data center demand projected to drive an 833% spike in PJM’s regional capacity auction prices for the Virginia grid zone between the 2024 and 2025–2026 market cycles (American Action Forum, January 2026).

---

Global Electricity Consumption: The Baseline

Data centers have crossed a threshold in 2025 and 2026 that energy analysts warned about for a decade: they are no longer a marginal electricity consumer. They are now a primary driver of demand growth in every major advanced economy.

Metric	Value	Source	Year
Global data center electricity consumption (2024)	415–460 TWh	IEA, Energy and AI	2025
Global data center electricity consumption (2025)	~460–490 TWh	IEA, Key Questions on Energy and AI	2026
Global electricity demand growth (2025)	3%	IEA	2026
Data center demand growth (2025)	17%	IEA	2026
AI-focused data center demand growth (2025)	50%	IEA	2026
Global data center demand projection (2030)	~945 TWh	IEA base case	2025
Global data center demand projection (2035)	~1,200–1,300 TWh	IEA base case	2025
Global data center share of global electricity (2024)	~1.8%	IEA / Sustainability Atlas	2025–2026

Sources: IEA, Key Questions on Energy and AI (April 2026); IEA, Energy and AI (April 2025)

The IEA’s April 2026 report — published as Axis Intelligence Research was finalizing this dataset — is the most current institutional data on global data center energy demand. It confirms that electricity consumption from global data centers in 2025 rivals France’s annual national demand, a country of 68 million people.

The IEA’s base-case scenario projects global data center consumption rising from roughly 460–490 TWh in 2025 to over 945 TWh by 2030 and 1,200–1,300 TWh by 2035. Across four modeling scenarios (base, high, low, and technology-led efficiency), the variation in 2030 estimates is narrow — within roughly ±100 TWh of the base case. The reason: a large share of future supply is already “locked in” through signed contracts, grid connection approvals, and under-construction facilities. The range opens wider only after 2030.

AI’s Specific Energy Footprint

Separating AI workloads from traditional data center operations is methodologically difficult — hyperscalers rarely disclose AI-specific energy consumption. But the IEA and Epoch AI have produced the most granular available estimates.

Metric	Value	Source	Year
AI-focused data center demand growth (2025)	+50% YoY	IEA	2026
AI share of data center power use (current)	Roughly 5–15%	IEA / Carbon Brief	2025
AI share of data center power use (projected 2030)	35–50%	IEA	2025
Power required by xAI Colossus supercomputer	~280 MW	Epoch AI	2025
Power required by Summit supercomputer (2019 baseline)	13 MW	Epoch AI	2025
Growth rate: AI training power requirements	Doubling every 13 months	Epoch AI	2025
Projected frontier AI training power demand (2030)	4–16 GW (single run)	Epoch AI / EPRI	2025
AI carbon footprint estimate (2025)	32.6–79.7 Mt CO₂	Cell Reports Sustainability, NIH/PMC	2025
Energy per simple AI text query	< TV over same period	IEA	2026
Energy for all search queries done via AI	< 4 TWh/year	IEA	2026

Sources: IEA, Key Questions on Energy and AI (2026); Epoch AI, AI Supercomputers Power Trend (June 2025); Epoch AI, Power Demands of Frontier AI Training (August 2025); NIH / PMC, “The carbon and water footprints of data centers” (December 2025)

The IEA’s April 2026 report offers a key nuance that counterbalances the alarming headline figures: power consumption per AI task is declining at an “unprecedented rate in energy history.” Software and hardware advances have driven energy per AI task down by at least one order of magnitude annually. Simple text queries now consume less electricity over the same period as running a television. If all internet searches globally were performed through AI text queries, total consumption would be less than 4 TWh per year — under 1% of current total data center consumption.

The problem is that cheaper-per-task AI is expanding its user base faster than efficiency gains are shrinking aggregate demand. New energy-intensive applications — video generation, reasoning, AI agents — are now the dominant growth category. As a result, aggregate AI energy demand continues to climb even as individual task efficiency improves.

Epoch AI’s tracking of frontier AI supercomputers is particularly striking. The leading AI training cluster has grown from 13 MW in January 2019 (Oak Ridge Summit) to approximately 280 MW in 2025 (xAI’s Colossus) — a 20× increase in six years. The power requirements of leading AI supercomputers have been doubling every 13 months since 2019. Epoch AI’s models project that the largest single AI training runs in 2030 will draw 4–16 GW of power — equivalent to powering several million U.S. homes from one facility.

U.S. Data Center Demand: The Hyperscale Surge

The United States hosts roughly 45% of global AI data center capacity by power draw. The DOE / Lawrence Berkeley National Laboratory’s December 2024 report — the most comprehensive official U.S. dataset available — provides the clearest longitudinal picture.

Metric	Value	Source	Year
U.S. data center consumption (2014)	58 TWh	DOE/LBNL	2024
U.S. data center consumption (2023)	176 TWh	DOE/LBNL	2024
U.S. data center share of U.S. electricity (2023)	4.4%	DOE/LBNL	2024
U.S. data center share of U.S. electricity (2018)	1.9%	DOE/LBNL	2024
U.S. data center consumption projection (2028, low)	325 TWh	DOE/LBNL	2024
U.S. data center consumption projection (2028, high)	580 TWh	DOE/LBNL	2024
U.S. data center share of electricity (2028, low)	6.7%	DOE/LBNL	2024
U.S. data center share of electricity (2028, high)	12%	DOE/LBNL	2024
U.S. data center consumption (2024, IEA estimate)	183 TWh	IEA / Pew Research	2025
U.S. data center consumption projection (2030)	426 TWh (+133% from 2024)	IEA / Pew Research	2025
Typical hyperscale AI data center annual consumption	≈ 100,000 households	IEA	2025
Largest planned hyperscale facilities	≈ 2,000,000 households equivalent	IEA	2025

Sources: DOE/LBNL, 2024 U.S. Data Center Energy Usage Report (December 2024); Pew Research Center, “What we know about energy use at U.S. data centers” (October 2025)

Between 2017 and 2023, U.S. data center power demand more than doubled. The LBNL report links this directly to the growth in AI servers — a category that barely registered in the 2016 version of the same study. The data center sector consumed 4.4% of total U.S. electricity in 2023. Under the high-growth scenario, this reaches 12% by 2028 — making data centers the second-largest electricity consumer category in the country after residential use.

At the national level, the U.S. Energy Information Administration has confirmed that commercial electricity demand from data centers is projected to grow by 3% in 2025 and 4.5% in 2026. Virginia’s grid zone, which hosts the world’s largest concentration of data centers, has seen summer peak load increase 23% since 2019 and winter peak load increase 45% — changes driven almost entirely by data center growth.

Geographic Concentration and Grid Stress

Data center energy demand is not spread evenly. It concentrates in clusters that can account for a disproportionate share of local grid consumption — creating systemic risks that aggregate global statistics obscure.

Region	Data Center Share of Local Electricity	Source	Notes
Northern Virginia (PJM Dominion zone)	26%+ of state electricity	IEA / Pew Research	World’s largest data center concentration
Dublin (city grid)	~79%	Oeko-Institut / Carbon Brief	Extreme local concentration
Ireland (national grid)	~21% (est. 32% by 2026)	IEA	[Older data: 2023 national figure]
Frankfurt (metro grid)	~42%	Brookings Institution	European hub
Texas (ERCOT)	Growing concentration	Belfer Center, Harvard	Second-largest U.S. data center market
Virginia (projected demand increase by 2040)	+183%	American Action Forum	Driven by AI data center expansion

Sources: EIA, “Commercial electricity sales have soared in Virginia” (May 2026); Brookings Institution, “Global energy demands within the AI regulatory landscape” (April 2026)

Virginia’s situation is the clearest case study of what AI-driven data center concentration does to a regional grid. PJM’s capacity auction price for the Dominion zone rose 833% between the 2024 market and the 2025–2026 market cycle — an unprecedented single-cycle price spike in a regulated capacity market. Virginia overtook California as the top electricity importer in 2023, meaning it cannot produce enough power domestically to meet data center demand and must draw from neighboring states’ grids.

In response, the Virginia State Corporation Commission approved a new “GS-5” rate category in November 2025 — effective January 2027 — that applies specifically to electricity consumers drawing 25 MW or more. The measure is designed to prevent residential ratepayers from absorbing the infrastructure costs of data center expansion, and is being closely watched by regulators in Texas, Georgia, and Ohio as a potential national model.

The IEA notes that on average, a quarter of electricity demand growth by 2030 in advanced economies will be driven by data centers. In the U.S. and Japan specifically, approximately half of projected power demand growth over the next five years is expected to come from data centers alone.

Hyperscaler Capital Expenditure and Infrastructure Build-Out

The scale of private infrastructure investment in AI data centers has no peacetime equivalent outside of government programs. Epoch AI’s tracking of hyperscaler capex — derived from SEC filings — provides the most granular public longitudinal dataset.

Company	2026 Capex (Projected)	Key Data Center Commitment	Source
Amazon (AWS)	~$200B total	1.92 GW nuclear PPA (Talen/Susquehanna, 17-year); $20B Pennsylvania investment	Tom’s Hardware, 2026
Alphabet (Google)	$175–190B	$4B Arkansas; 600 MW Texas solar; custom AI chip strategy	CNBC / Tom’s Hardware, 2026
Microsoft	$120–190B	$80B AI data centers (FY2025); 34.7 GW clean energy contracted	CNBC / S&P Global, 2025–2026
Meta	$115–145B	6.6 GW nuclear pipeline; 20-year Constellation/Clinton PPA	Data Center Knowledge, 2026
Oracle	$50B	Stargate participation; SMR-powered facility design	Futurum Group, 2026
5-Company Total (2026)	$660–725B	75% tied directly to AI infrastructure	Tom’s Hardware / CNBC, 2026

Sources: Tom’s Hardware, “Big Tech’s AI spending plans reach $725 billion” (April 2026); Epoch AI, “Hyperscaler capex trend” (2026); IEA, Key Questions on Energy and AI (2026)

The IEA’s April 2026 report notes that the capital expenditure of five large technology companies exceeded $400 billion in 2025 and is set to increase by a further 75% in 2026 — making this investment group larger than global annual investment in oil and natural gas production combined. The five hyperscalers collectively plan to add roughly $2 trillion in AI-related assets to their balance sheets by 2030.

Epoch AI’s data shows that combined capex across the five hyperscalers grew at an average annual rate of 72% between Q2 2023 and Q4 2025. From a 2022 baseline of $162.3 billion, this quadrupled to $448.3 billion in 2025. The Stargate joint venture (OpenAI, SoftBank, Oracle, MGX) targets an additional $500 billion in AI infrastructure investment by 2029, with approximately 7 GW of capacity planned across five U.S. sites and more than $400 billion in commitments within the first three years.

The Nuclear and Renewable Pivot

The power demands of AI data centers are large enough — and reliability requirements strict enough — that hyperscalers have moved beyond renewable energy certificates and into direct nuclear procurement, advanced geothermal, and small modular reactor (SMR) partnerships.

Metric	Value	Source	Year
Tech sector share of global corporate renewable PPAs (2025)	~40%	IEA	2026
U.S. nuclear capacity contracted by hyperscalers (18 months through 2025)	10+ GW	ETF Stream / S&P Global	2025–2026
SMR conditional offtake pipeline (end of 2024)	25 GW	IEA	2026
SMR conditional offtake pipeline (April 2026)	45 GW	IEA	2026
U.S. data centers’ clean energy contracts total	80+ GW	S&P Global	2025
Microsoft clean energy contracted (end Sept. 2025)	34.7 GW	S&P Global	2025
Meta nuclear pipeline (agreements through Jan. 2026)	6.6 GW	Data Center Knowledge	2026
AWS nuclear PPA (Talen/Susquehanna, through 2042)	1.92 GW / 17-year	iRecruit	2026

Sources: IEA, Key Questions on Energy and AI (2026); S&P Global Sustainable1, “Hyperscaler procurement to shape US power investment” (December 2025); ETF Stream, “Data centres drive momentum for nuclear” (2026)

The SMR conditional offtake pipeline — which tracks agreements between data center operators and companies developing small modular reactors — nearly doubled in 18 months, from 25 GW at end-2024 to 45 GW as of April 2026, per the IEA. This is the clearest indicator that hyperscalers are not planning around grid power for the next decade: they are planning around dedicated, co-located generation.

Goldman Sachs Research has projected that 40% of new data center electricity demand through 2030 will be met by renewables, with nuclear contributing a growing but currently modest share. Natural gas is expected to supply the bulk of the remaining 60%, potentially adding 215–220 million tons of CO₂ by 2030 relative to a zero-carbon scenario.

In Europe, the 140 proposed data center schemes in the UK alone could collectively require 50 GW of power — 5 GW more than the UK’s current peak electricity demand — per the Institution of Engineering and Technology. The pipeline is straining planning and regulatory systems across the EU.

Carbon Emissions and the Sustainability Gap

The carbon arithmetic of AI data centers is contested, because it depends heavily on the grid carbon intensity of each facility’s location, the method of renewable energy accounting, and whether embodied emissions from hardware manufacturing are included.

Metric	Value	Source	Year
CO₂ from global data center electricity generation (2024)	~182 Mt CO₂	IEA, cited in NIH/PMC paper	2025
Implied average carbon intensity per kWh (2024)	~395.65 gCO₂/kWh	Cross-calculation: IEA 460 TWh + 182 Mt CO₂	2025
AI systems carbon footprint estimate (2025)	32.6–79.7 Mt CO₂	NIH / PMC (Cell Reports Sustainability)	2025
Projected global data center CO₂ share (2030, base case)	~1% of global CO₂	IEA	2025
Projected global data center CO₂ share (2030, high scenario)	~1.4% of global CO₂	IEA	2025
U.S. data center CO₂ (projected peak, 2030)	63–83 Mt CO₂/year	Global Efficiency Intelligence	2025
U.S. AI growth CO₂ by 2030 (current trajectory)	24–44 Mt CO₂/year	Cornell University research	2025
Cars equivalent to U.S. AI data center CO₂ by 2030	5–10 million vehicles	Cornell University	2025

Sources: NIH / PMC, “The carbon and water footprints of data centers and what this could mean for artificial intelligence” (December 2025); Cornell Chronicle, “Roadmap shows the environmental impact of AI data center boom” (November 2025); Global Efficiency Intelligence, “Data Centers in the AI Era” (2025)

The IEA notes that data centers are one of the few sectors where absolute emissions are projected to grow — alongside road transport and aviation — even as most of the economy decarbonizes. Hyperscalers’ net-zero pledges are increasingly diverging from disclosed emissions: Google, Microsoft, and Meta have all reported emissions spikes in recent years despite corporate renewable energy purchasing, because growth is outpacing clean energy procurement speed.

Cornell University’s state-by-state analysis (published November 2025) found that at the current rate of AI growth, U.S. AI data centers alone would add the annual CO₂ equivalent of 5–10 million additional cars on U.S. roads by 2030. The study concluded that cumulative effects would push the AI industry’s net-zero targets out of reach without dramatic policy or technology intervention.

For facilities powered by very low-carbon electricity (renewables or nuclear), embodied emissions from manufacturing become proportionally more significant. Research published in Cell Reports Sustainability (December 2025, via NIH/PMC) found that when operational emissions are near-zero, hardware manufacturing can account for up to 40% of total lifetime greenhouse gas emissions — with chips and memory alone representing 67% of embodied emissions.

Water Consumption: The Hidden Metric

Water usage is the least-disclosed and most underappreciated environmental dimension of AI data center growth. It operates through two channels: direct evaporative cooling at facilities, and indirect consumption through electricity generation.

Metric	Value	Source	Year
Direct water consumption, U.S. data centers (2023)	17 billion gallons	DOE / LBNL	2024
Indirect water use, U.S. data centers (electricity generation, 2023)	~211 billion gallons	DOE / LBNL	2024
Share of U.S. data center water used by hyperscale/colo facilities	84%	DOE / LBNL	2024
North American data center water consumption (2025)	~1 trillion liters	Mordor Intelligence, cited by Reuters	2026
Google direct water consumption (2023)	24,227 megalitres	Google Sustainability Report 2023	2025
Microsoft direct water consumption (2023)	7,844 megalitres	Microsoft Environmental Report	2025
Industry water use projection (2028)	~1,068 billion liters/year	Industry research	2025
Water stressed area share of Microsoft water use	41%	Microsoft Environmental Report	2024
Typical single 100 MW data center daily water use	~528,000 gallons/day	DOE / LBNL	2024
Water use: AI vs. traditional data centers	10–50× higher per unit	Introl research	2025–2026
Direct water reduction via liquid cooling vs. evaporative	70–90% reduction	Introl research	2025–2026

Sources: DOE/LBNL, 2024 U.S. Data Center Energy Usage Report; NIH / PMC, December 2025

Water transparency is the sector’s most significant disclosure gap. Amazon Web Services, the world’s largest cloud infrastructure provider, does not publish aggregate water consumption figures as of the date of this report. More than a dozen institutional investors have formally demanded site-level water and energy transparency from Amazon, Microsoft, and Alphabet ahead of 2026 annual general meetings, citing community impacts and local water supply stress.

Two-thirds of new U.S. hyperscale campuses built since 2022 are located in areas already facing water shortages. Microsoft’s regional water usage effectiveness (WUE) data shows the scale of geographic variation: Arizona facilities use 1.52 liters per kWh versus 0.02 liters per kWh at Singapore facilities — a 76:1 ratio driven entirely by climate conditions.

Liquid (immersion or direct-to-chip) cooling can reduce direct water consumption by 70–90% compared to evaporative cooling, and adoption is accelerating in 2025–2026 as AI GPU racks reach power densities (50–135 kW per rack) that exceed what air cooling can handle. However, liquid cooling addresses direct water use only — the dominant indirect water footprint from electricity generation remains unchanged.

Axis Intelligence Original Metric: The AI Energy Intensity Index™

Snapshot date: June 5, 2026

This metric does not exist in any prior published form. Axis Intelligence Research derived it by cross-referencing IEA (2026), DOE/LBNL (2024), Epoch AI (2025), and Goldman Sachs Research (2025). No individual source publishes this combined figure.

Definition

The Axis Intelligence AI Energy Intensity Index™ (AEII™) measures the ratio of AI-attributed electricity demand growth to total data center electricity demand growth, expressed as a percentage uplift multiplier, at the national (U.S.) level — updated quarterly.

Calculation: Q2 2026 Snapshot

Step 1 — Establish total U.S. data center demand growth rate (2023→2024):

• 2023 baseline: 176 TWh (DOE/LBNL)
• 2024 estimate: 183 TWh (IEA / Pew Research)
• Total growth: +7 TWh / +3.98% year-over-year

Step 2 — Establish AI-focused data center growth rate (global, 2024→2025):

• Global AI-focused data center demand grew 50% in 2025 (IEA, April 2026)
• Total data center demand grew 17% in 2025 (IEA, April 2026)
• AI uplift ratio: 50% ÷ 17% = 2.94× — meaning AI-focused facilities are consuming electricity nearly 3× faster than the overall data center sector

Step 3 — Apply the AEII™ to U.S. forward projections:

• If U.S. data centers reach the LBNL midpoint projection of 452 TWh by 2028 (midpoint of 325–580 TWh range)
• And AI workloads continue to drive demand at 2.94× the overall rate
• Then AI-attributed consumption within U.S. data centers reaches approximately 160–220 TWh by 2028 — a range no single published source has computed

AEII™ Q2 2026 Reading: 2.94×

This means: for every 1% of total data center electricity growth, AI-focused consumption is growing at 2.94%. At this ratio, AI workloads will account for the majority of U.S. data center electricity growth by 2027.

Update commitment: Axis Intelligence Research will recompute the AEII™ each quarter using the latest available IEA, EIA, and DOE primary source releases. The Q3 2026 update is scheduled for September 2026.

Methodology note: The AEII™ uses global AI growth rate data (IEA) applied to U.S. national demand projections (DOE/LBNL). This cross-jurisdictional application introduces uncertainty, acknowledged in the limitations section. The metric is designed as a directional indicator, not a point estimate.

Methodology

Data collection

Axis Intelligence Research assembled this dataset between May 15 and June 5, 2026. Primary sources consulted include:

1. International Energy Agency (IEA) — Key Questions on Energy and AI (April 2026) and Energy and AI (April 2025)
2. U.S. Department of Energy / Lawrence Berkeley National Laboratory — 2024 United States Data Center Energy Usage Report (December 2024)
3. Epoch AI — AI Supercomputers Power Trend (June 2025), Power Demands of Frontier AI Training (August 2025), Hyperscaler Capex Trend (2025–2026)
4. Goldman Sachs Research — AI to Drive 165% Increase in Data Center Power Demand by 2030 (2025)
5. U.S. Energy Information Administration (EIA) — Commercial electricity sales in Virginia (May 2026)
6. Pew Research Center — What we know about energy use at U.S. data centers (October 2025)
7. Brookings Institution — Global energy demands within the AI regulatory landscape (April 2026)
8. NIH / PubMed Central — The carbon and water footprints of data centers and what this could mean for artificial intelligence, Cell Reports Sustainability (December 2025)
9. Cornell University — Roadmap shows the environmental impact of AI data center boom (November 2025)
10. S&P Global Sustainable1 — Hyperscaler procurement to shape US power investment (December 2025)

Limitations

• AI vs. non-AI workload separation: No hyperscaler publicly discloses energy consumption broken out by AI versus non-AI workloads. All AI-specific figures use institutional estimates (primarily IEA and Epoch AI).
• Water data gaps: Amazon Web Services does not publish aggregate water consumption. Water figures for AWS facilities are estimated from facility-level disclosures and cross-source extrapolation.
• Time lag: The most granular U.S. data (DOE/LBNL) was published in December 2024 using 2023 baseline data. Global data (IEA) is more current (2026) but does not disaggregate at the country level for all metrics.
• AEII™ cross-jurisdictional application: The AI Energy Intensity Index applies global growth ratios to U.S.-specific demand projections. This introduces estimation error and should be treated as directional.
• Rapidly changing landscape: Hyperscaler capex figures shift quarterly. Figures cited as “2026 projected” reflect guidance issued in Q1–Q2 2026 earnings calls and filings.

Inclusion criteria

All statistics in this report come from primary institutional sources: government agencies, academic research institutions, named research organizations, or peer-reviewed publications. Secondary aggregators and editorial tech publications are excluded as sources. Where institutional sources use estimates or projections, these are clearly labeled.

About This Dataset

Dataset title: AI Data Center Energy Consumption Statistics 2026 Published by: Axis Intelligence Research Publication date: June 5, 2026 Update cadence: Quarterly (next update: September 2026) License: Creative Commons Attribution 4.0 International (CC BY 4.0) Download: ai-data-center-energy-2026.csv

Under the CC BY 4.0 license, you are free to share, copy, redistribute, adapt, and build upon this dataset for any purpose, including commercial use, provided you credit Axis Intelligence as the source.

Required attribution: “Axis Intelligence Research, AI Data Center Energy Consumption Statistics 2026, axis-intelligence.com/ai-data-center-energy-consumption-statistics/, June 2026, CC BY 4.0″

Cite This Research

APA 7th Edition

Axis Intelligence Research. (2026, June 5). AI data center energy consumption statistics 2026. Axis Intelligence. https://axis-intelligence.com/ai-data-center-energy-consumption-statistics/

MLA 9th Edition

Axis Intelligence Research. “AI Data Center Energy Consumption Statistics 2026.” Axis Intelligence, 5 June 2026, axis-intelligence.com/ai-data-center-energy-consumption-statistics/.

Chicago Author-Date

Axis Intelligence Research. 2026. “AI Data Center Energy Consumption Statistics 2026.” Axis Intelligence. June 5. https://axis-intelligence.com/ai-data-center-energy-consumption-statistics/.

Embed This Research (copy-paste HTML for passive do-follow backlinks)

<blockquote class="axis-intelligence-stat" cite="https://axis-intelligence.com/ai-data-center-energy-consumption-statistics/">
  <p>Global data center electricity consumption reached 460–490 TWh in 2025 — a 17% year-over-year surge — with AI-focused data centers driving a 50% spike in the same period. Total consumption is projected to double by 2030.</p>
  <footer>
    Source: <a href="https://axis-intelligence.com/ai-data-center-energy-consumption-statistics/" rel="dofollow">
    Axis Intelligence Research — AI Data Center Energy Consumption Statistics 2026</a> (CC BY 4.0)
  </footer>
</blockquote>

---

Frequently Asked Questions {#faq}

How much electricity do AI data centers use in 2026?

Global data centers consumed an estimated 460–490 TWh of electricity in 2025, growing 17% year-over-year. AI-focused data centers within that total grew 50%. For 2026, IEA estimates project continued double-digit growth, though the exact figure will not be confirmed until the IEA’s next annual energy report.

How fast is AI data center energy consumption growing?

AI-focused data center electricity demand grew 50% in 2025 — nearly five times faster than the 3% growth in total global electricity demand that year. The power requirements of the leading AI training supercomputers have been doubling every 13 months since 2019 (Epoch AI, 2025).

What percentage of U.S. electricity do data centers use?

U.S. data centers consumed 4.4% of total U.S. electricity in 2023 (176 TWh), up from 1.9% in 2018. Under LBNL’s high-growth scenario, this could rise to 12% by 2028, making data centers one of the country’s largest electricity-consuming sectors.

Will data center electricity demand double by 2030?

The IEA’s base-case scenario projects that global data center electricity consumption will roughly double to ~945 TWh by 2030. AI-specific demand within that total is projected to triple. Goldman Sachs Research projects a 165–175% increase in global data center power demand by 2030 versus 2023 levels.

Are AI data centers switching to nuclear power?

Yes, at significant scale. Hyperscalers secured more than 10 GW of new or restarting nuclear capacity in the U.S. over 18 months through 2025. The pipeline of conditional offtake agreements for small modular reactors (SMRs) grew from 25 GW at end-2024 to 45 GW by April 2026 (IEA). Microsoft, Meta, Google, Amazon, and Oracle have all announced major nuclear or SMR partnerships.

How much water do AI data centers use?

U.S. data centers used about 17 billion gallons of water directly in 2023, with an estimated 211 billion gallons consumed indirectly through electricity generation. North American data centers consumed approximately 1 trillion liters of water in 2025. AI workloads require 10–50× more water per unit than traditional server workloads due to higher GPU rack power density and associated cooling demands.

Which country or region has the highest data center electricity concentration?

At the city level, Dublin, Ireland sees approximately 79% of metropolitan electricity consumed by data centers. In Northern Virginia (PJM Dominion zone), data centers account for more than 26% of statewide electricity. Frankfurt’s metropolitan grid has approximately 42% of local electricity going to data centers. Ireland’s national grid has roughly 21% data center penetration, projected to rise to 32%.

What is the carbon footprint of AI data centers?

The IEA estimates that global data center electricity generation produced approximately 182 Mt CO₂ in 2024. AI systems specifically may have produced 32.6–79.7 Mt CO₂ in 2025 (NIH/PMC, 2025). The IEA projects data centers will reach 1%–1.4% of global CO₂ emissions by 2030 — a growing share in a sector that is largely decarbonizing.

How much are tech companies spending on AI data centers in 2026?

The five largest hyperscalers (Amazon, Alphabet, Microsoft, Meta, Oracle) are projected to spend $660–725 billion in combined capital expenditure in 2026, with approximately 75% tied directly to AI infrastructure. This is up from a combined $448 billion in 2025 and $162 billion in 2022.

What is the Axis Intelligence AI Energy Intensity Index™ (AEII™)?

The AEII™ is an original metric developed by Axis Intelligence Research measuring the ratio of AI-attributed electricity demand growth to total data center electricity demand growth. The Q2 2026 reading is 2.94× — meaning AI-focused data centers are growing their electricity consumption nearly three times faster than the overall data center sector. The AEII™ is updated quarterly.