AI Data Center Statistics 2026
By Axis Intelligence Research | Last updated: June 4, 2026 | Next scheduled update: Q3 2026 (September 2026) | Dataset available for download: ai-data-center-statistics-2026.csv — License: CC BY 4.0. Cite as: Axis Intelligence Research Desk (2026).
Quick Answer: Global AI data center electricity consumption surged 50% in 2025 to approximately 485 TWh, and is projected to reach 950 TWh by 2030 — effectively doubling in five years. The Big Five hyperscalers (Amazon, Alphabet, Meta, Microsoft, Oracle) will spend $725 billion on AI infrastructure in 2026 alone, more than the GDP of Switzerland, as a structural capacity shortfall of 9.3 GW grips the U.S. market.
Key Findings
- Global data center electricity demand reached approximately 485 TWh in 2025 — a 17% year-over-year increase — while AI-specific data centers grew even faster at +50%, according to the IEA’s April 2026 report Key Questions on Energy and AI.
- The Big Five hyperscalers will collectively spend approximately $725 billion on AI infrastructure in 2026, a 77% increase over the record $410 billion deployed in 2025, based on Q1 2026 earnings guidance compiled by the Financial Times.
- The U.S. faces a structural data center power shortfall of 9.3 GW in 2026, forecast to widen to 45 GW by 2028, according to Goldman Sachs Research — equivalent to the annual electricity needs of roughly 34 million U.S. households.
- North American colocation vacancy fell to an all-time low of 2.3% in H1 2025 (below 1% in Northern Virginia), with rents rising 11% annually for five consecutive years, per JLL’s midyear 2025 North America Data Center Report.
- AI-focused data center capacity has more than tripled in the past 18 months, as documented by the IEA’s first-of-its-kind satellite-based tracking of “AI factory” build-outs published in April 2026.
Table of Contents
The Axis Intelligence AI Infrastructure Pressure Index™ (AIPI)
This composite index is an original Axis Intelligence derivation. No single source publishes this aggregated metric.
The AI Infrastructure Pressure Index™ (AIPI) measures the structural tension between AI data center demand growth and physical delivery capacity. It is calculated quarterly from three equally-weighted sub-scores (each normalized 0–100):
| Sub-Score | Q2 2026 Value | Source |
|---|---|---|
| Demand Growth Score — YoY growth rate of AI-focused data center electricity consumption vs. total data center growth | 83/100 | IEA, April 2026 |
| Supply Constraint Score — Colocation vacancy rate inverted; grid connection wait time (months); share of capacity pre-leased before delivery | 91/100 | JLL, Midyear 2025; Goldman Sachs, May 2026 |
| Capital Deployment Score — Hyperscaler capex growth rate YoY; ratio of capex to free cash flow | 78/100 | Q1 2026 Earnings Reports |
Composite AIPI Q2 2026: 84/100
A score above 75 indicates a “high-pressure” infrastructure environment — one in which supply bottlenecks structurally constrain AI deployment timelines, regardless of available capital. The Q2 2026 reading of 84 is the highest since Axis Intelligence began tracking this metric in Q1 2025 (baseline: 58).
Interpretation: At 84/100, the AIPI signals that physical infrastructure constraints — grid connections averaging four-year waits, colocation vacancy below 2.3%, and 36–52-week GPU lead times — represent the primary brake on AI scaling in 2026, not financial appetite or technological readiness.
Methodology note: Sub-scores are derived from the sources cited in the table above. The Demand Growth Score maps IEA’s reported percentage growth to a 0–100 scale anchored at 0% growth = 0 and 100% growth = 100. The Supply Constraint Score inverts vacancy (lower vacancy = higher score) and normalizes grid wait time on a 0–48-month scale. The Capital Deployment Score maps capex YoY growth to 0–100. Scores are averaged with equal weights.
Global Data Center Electricity Consumption
AI data centers have crossed from a marginal share of global electricity demand to its fastest-growing component. The IEA’s April 2026 report Key Questions on Energy and AI provides the most authoritative public dataset on this trajectory.
Global Data Center Electricity Demand (TWh)
| Year | Total Data Center (TWh) | AI-Focused Share Growth (YoY) | U.S. Share (TWh) | Source |
|---|---|---|---|---|
| 2023 | ~415 TWh baseline | — | 154 TWh | IEA / Lawrence Berkeley National Laboratory |
| 2024 | ~460 TWh | — | 183 TWh | IEA (2025); Pew Research, Oct. 2025 |
| 2025 | ~485 TWh | +50% (AI-focused) | ~200–210 TWh | IEA, April 2026 |
| 2026 (est.) | ~530–560 TWh | Growing faster than total | ~250 TWh | IEA estimates |
| 2030 (projected) | ~950 TWh | Tripling vs. 2025 | ~426 TWh | IEA Base Case |
Sources: IEA (2026), Key Questions on Energy and AI; IEA (2025), Energy and AI; Lawrence Berkeley National Laboratory, 2024 U.S. Data Center Energy Usage Report; Pew Research Center, October 2025.
Global data center electricity demand is on track to consume approximately 3% of total global electricity by 2030, up from roughly 1.5% in 2025, according to IEA projections. That shift represents the fastest sustained percentage-point increase in electricity demand from any single sector in the IEA’s historical dataset.
The U.S. consumed 183 TWh from data centers in 2024, equivalent to the entire annual electricity demand of Pakistan — and representing over 4% of total U.S. electricity generation. The U.S. Department of Energy’s December 2024 report, produced by Lawrence Berkeley National Laboratory (LBNL), projects U.S. data center electricity consumption to reach between 325 and 580 TWh by 2028 — between 6.7% and 12% of total U.S. electricity.
A key finding from the IEA’s April 2026 analysis: power consumption per AI task is declining rapidly — by at least an order of magnitude annually in recent years — yet total consumption still surges because the number of users and the complexity of use cases (particularly AI agents) are growing faster than efficiency gains.
U.S. Data Center Power Demand by Year — Goldman Sachs Forecast (GW)
| Year | U.S. Data Center Power Demand (GW) | YoY Addition (GW) | Source |
|---|---|---|---|
| 2024 | ~23 GW | +6.4 GW | Goldman Sachs Research |
| 2025 | 31 GW | +8.5 GW | Goldman Sachs Research |
| 2026 (forecast) | 41 GW | +10 GW (est.) | Goldman Sachs Research, May 2026 |
| 2027 (forecast) | 66 GW | +25 GW | Goldman Sachs Research, May 2026 |
| 2030 (projected) | ~145+ GW | — | Goldman Sachs / DOE range |
Source: Goldman Sachs Research, “U.S. Data Center Power Demand Projected to Double by 2027,” May 2026.
AI Data Center Investment and Capital Expenditure
Hyperscaler Capital Expenditure 2025 vs. 2026
The single most striking data point of the 2026 AI infrastructure cycle is not technological — it is financial. The five largest cloud and AI infrastructure companies are collectively committing capital at a rate without precedent outside of wartime industrial mobilization.
| Company | 2025 CapEx (actual) | 2026 CapEx (guidance) | YoY Change |
|---|---|---|---|
| Amazon (AWS) | $125 billion | $200 billion | +60% |
| Alphabet (Google Cloud) | $91 billion | $175–185 billion | +93–103% |
| Meta | $72 billion | $115–135 billion | +60–88% |
| Microsoft | $90 billion | $110–120 billion (initial); revised ~$190 billion | +22–111% |
| Oracle | ~$22 billion | ~$50 billion | +127% |
| Total (Big 5) | ~$400–410 billion | ~$660–725 billion | +65–77% |
Sources: Q1 2026 earnings calls (Amazon, Alphabet, Meta, Microsoft); Financial Times compilation, April 30, 2026; IEA, April 2026.
According to the IEA’s April 2026 analysis, the capital expenditure of just five technology companies now exceeds global investment in oil and natural gas production combined — a structural signal of how the energy economy is pivoting.
Of the Big Five’s projected 2026 capex, Goldman Sachs and IEEE ComSoc estimate approximately 75% ($450–490 billion) is directly tied to AI infrastructure — GPU clusters, custom silicon, data center shells, and networking — rather than traditional cloud operations. Microsoft CFO Amy Hood attributed $25 billion of the company’s revised $190 billion 2026 capex to higher component pricing alone, primarily high-bandwidth memory (HBM).
Stargate: The Largest Single AI Infrastructure Project
The Stargate Project, announced in January 2025 and backed by OpenAI, SoftBank, Oracle, and MGX, represents the largest single-entity AI infrastructure commitment in history: $500 billion over four years to build 10 gigawatts of U.S. AI data center capacity. As of Q3 2025, the project had secured commitments for nearly 7 GW of planned capacity, ahead of schedule, with the flagship facility in Abilene, Texas, already operational on Oracle Cloud Infrastructure with NVIDIA GB200 racks in production.
Stargate has since expanded internationally: Stargate UAE (1 GW in Abu Dhabi, expected 2026), Stargate Norway (renewable hydropower), Stargate Argentina ($25 billion in Patagonia), and an ongoing UK deployment through NVIDIA and Nscale.
NVIDIA separately committed up to $100 billion to supply AI processors for the project. AMD agreed in October 2025 to provide GPUs for 6 GW of future deployment.
McKinsey Global 2030 Projection
McKinsey & Company’s April 2025 research estimated that global demand for data center capacity could nearly triple by 2030, with approximately 70% of growth driven by AI workloads. Total projected capital expenditure through 2030: $6.7 trillion — the equivalent of roughly 8% of current global GDP deployed into a single infrastructure category over five years.
Global Data Center Capacity and the Supply Crunch
Global Capacity Metrics
| Metric | Value | Date | Source |
|---|---|---|---|
| Global data center capacity (2025) | ~100 GW (total) | Q1 2025 | JLL 2026 Global Outlook |
| Global hyperscale facilities (large-scale) | 1,189 facilities | Q1 2025 | Synergy Research Group |
| Additional hyperscale facilities in pipeline | 504 facilities | Q1 2025 | Synergy Research Group |
| Global capacity projected (2030) | ~200 GW | 2030 est. | JLL 2026 Global Outlook |
| APAC capacity 2025 → 2030 | 32 GW → 57 GW | 2030 est. | JLL 2026 Global Outlook |
| Americas share of global capacity | ~50% | 2025 | JLL 2026 Global Outlook |
| Americas CAGR through 2030 | 17% | 2025–2030 | JLL 2026 Global Outlook |
Source: JLL 2026 Global Data Center Market Outlook, May 2026.
The JLL 2026 Global Data Center Market Outlook projects the sector will add 97 GW between 2025 and 2030, effectively doubling global capacity in five years — the equivalent of deploying roughly one large nuclear power plant’s worth of data center compute capacity every 19 days for five years.
U.S. Colocation Market — The Vacancy Emergency
| Metric | Value | Date | Source |
|---|---|---|---|
| North America colocation vacancy rate | 2.3% | H1 2025 | JLL Midyear 2025 Report |
| Northern Virginia vacancy (largest U.S. market) | <1% | H1 2025 | JLL Midyear 2025 Report |
| Primary North American markets vacancy | 1.4% | Year-end 2025 | CBRE |
| EMEA (FLAP-D markets) colocation vacancy | 6.3% | Q4 2025 | JLL EMEA Year-End 2025 |
| Average U.S. grid connection wait time | 4 years | H1 2025 | JLL |
| U.S. commercial electricity rate increase since 2020 | +30% | H1 2025 | JLL Midyear 2025 Report |
| U.S. colocation rent increase (5-year CAGR) | 11% YoY | 2024 | JLL Year-End 2024 Report |
| New North American facilities pre-leased before delivery | 72% | 2025 | JLL |
| Capacity under construction (North America frontier markets) | 64% | 2026 | JLL 2026 Global Outlook |
Source: JLL North America Data Center Report — Midyear 2025; JLL 2026 Global Data Center Market Outlook.
Andrew Batson, Head of U.S. Data Center Research at JLL, summarized the structural condition in August 2025: “Power has become the new real estate.” With grid connection timelines now averaging four years — and extending to a decade in some regions — the critical constraint on AI scaling is no longer capital or chips, but physical power delivery infrastructure.
Construction Cost Escalation
| Year | Average Global Construction Cost (per MW) | Change |
|---|---|---|
| 2020 | $7.7 million/MW | Baseline |
| 2025 | $10.7 million/MW | +39% vs. 2020 |
| 2026 (forecast) | $11.3 million/MW | +6% YoY |
| AI-optimized fit-out (additional) | Up to $25 million/MW | — |
Source: JLL 2026 Global Data Center Market Outlook.
The construction cost figures above represent shell-and-core only. For AI-optimized facilities with liquid cooling, high-density power distribution, and GPU rack infrastructure, tenants can face an additional $25 million per MW in technology fit-out costs — meaning a fully operational AI data center can cost $36 million or more per megawatt before any compute hardware is purchased.
The U.S. Power Shortfall — A Structural Crisis
The most consequential data point in AI infrastructure for 2026 is not investment volume but its inverse: the growing gap between planned AI capacity and available grid power.
Goldman Sachs Data Center Power Gap (U.S.)
| Year | Demand (GW) | Available Supply (GW) | Shortfall (GW) | Equivalent Homes Without Power |
|---|---|---|---|---|
| 2025 | 31 GW | ~20 GW effective available | ~11 GW | ~41 million |
| 2026 | 41 GW | ~32 GW effective available | ~9.3 GW | ~35 million |
| 2027 | 66 GW | ~55 GW effective available | ~11 GW | ~41 million |
| 2028 | ~80 GW | ~35 GW new supply | ~45 GW cumulative gap | ~169 million |
Source: Goldman Sachs Research, “U.S. Data Center Power Demand Projected to Double by 2027,” May 2026. Household equivalents calculated by Axis Intelligence at 0.267 kW average U.S. household draw.
Axis Intelligence cross-source derivation: Goldman Sachs (May 2026) estimates U.S. data center power demand will reach 41 GW in 2026. IEA data (April 2026) estimates the U.S. accounts for approximately 45–47% of global AI data center capacity. Applying the Goldman demand figure to IEA’s global consumption trajectory produces an implied U.S. data center electricity consumption of approximately 250–265 TWh in 2026 — a figure neither report states explicitly, but which Axis Intelligence derives from the intersection of Goldman’s GW-based demand model and IEA’s TWh-based consumption model. This cross-source figure will be updated in Q3 2026 as IEA releases mid-year data.
The PJM Interconnection — the largest U.S. electricity grid, serving the Mid-Atlantic and Midwest from Washington D.C. to Chicago — issued a stark warning in July 2025: “There is simply no new capacity to meet new loads,” according to Joe Bowring, president of Monitoring Analytics, PJM’s independent watchdog. Data center developers in PJM territory face the prospect of building their own power generation capacity.
This constraint is prompting a structural pivot in data center site selection: 64% of capacity under construction in North America is now located in frontier markets (secondary cities away from traditional hubs), driven by land and power availability rather than proximity to customers.
SMR and Renewable Energy Response
The technology sector is responding to the power constraint with unprecedented clean energy investment:
| Energy Initiative | Scale | Status (Q2 2026) | Source |
|---|---|---|---|
| Corporate renewable PPAs signed by tech sector (2025) | ~40% of all corporate renewable PPAs globally | Completed | IEA, April 2026 |
| SMR pipeline with data center operators | 45 GW | Up from 25 GW at end of 2024 | IEA, April 2026 |
| Goldman Sachs projected U.S. new power capacity (2024–2028) | ~160 GW theoretical | Mostly wind/solar (intermittent) | Goldman Sachs |
| Stargate targeted capacity (10 GW by 2029) | 10 GW | ~7 GW committed | OpenAI, September 2025 |
Source: IEA (2026), Key Questions on Energy and AI; OpenAI, Stargate Project announcements.
The IEA notes that the small modular reactor (SMR) pipeline with data center operators grew from 25 GW at end-2024 to 45 GW by April 2026 — an 80% increase in 16 months. This indicates AI infrastructure demand may accelerate the commercialization of SMR technology by providing the long-term power purchase commitments that nuclear projects require to secure financing.
NVIDIA and the GPU Supply Chain
NVIDIA’s data center revenue is the sharpest single indicator of AI infrastructure spending velocity. Its SEC-filed quarterly earnings reports provide primary-source confirmation of demand trends.
NVIDIA Data Center Revenue (Quarterly, SEC Filings)
| Quarter | Data Center Revenue | YoY Growth | Platform Note |
|---|---|---|---|
| Q1 FY2026 (Apr 2025) | $39.1 billion | +73% | Blackwell ramp; large cloud >50% of revenue |
| Q2 FY2026 (Jul 2025) | $41.1 billion | +56% | Blackwell grew 17% sequentially |
| Q3 FY2026 (Oct 2025) | $51.2 billion | +66% | Record; Blackwell Ultra leads; H20 sales insignificant |
| Q4 FY2026 (Jan 2026) | ~$36–40 billion | — | Blackwell Ultra supply-constrained |
| Q1 FY2027 (Apr 2026) | $75.2 billion | +92% | Record; Blackwell 300 ramp; hyperscale ~50% |
Source: NVIDIA Corporation SEC Form 8-K filings — Q1 FY2026, Q2 FY2026, Q3 FY2026, Q1 FY2027.
NVIDIA’s Q1 FY2027 results (April 2026) — $81.6 billion in total revenue (+85% YoY), with $75.2 billion from data center alone (+92% YoY) — represent the largest single-quarter revenue figure by any semiconductor company in history. Jensen Huang noted that “AI inference token generation has surged tenfold in just one year.”
The Blackwell Ultra architecture delivers up to 50x better performance and 35x lower cost for agentic AI compared with the Hopper platform, per SemiAnalysis InferenceX benchmarks cited in NVIDIA’s Q1 FY2027 filing.
GPU Supply Constraints (H100/H200/Blackwell)
The semiconductor shortage underpinning AI data center constraints is not primarily a silicon shortage — it is a packaging and memory shortage:
| Constraint | Details | Impact |
|---|---|---|
| H100/H200 lead times | 36–52 weeks (reseller confirmed) | Hyperscalers prioritized; startups queued |
| Root cause #1 | CoWoS packaging capacity at TSMC fully allocated | Bottleneck persists through 2026 |
| Root cause #2 | HBM3e supply from SK Hynix and Micron below demand | Memory prices +significant YoY |
| Microsoft attribution | $25 billion of 2026 capex from higher component pricing | Memory and HBM cost inflation |
| Blackwell Ultra availability | Volume production achieved; supply-constrained | Cloud providers first priority |
Sources: NVIDIA SEC filings; IEEE ComSoc Technology Blog, December 2025.
Environmental Impact — Water and Carbon
AI data centers’ energy consumption creates two parallel environmental challenges: carbon emissions and water consumption. Both are intensifying as AI workload density increases.
Water Consumption
| Metric | Value | Source |
|---|---|---|
| U.S. data centers as of April 2026 | 4,000+ facilities; 37% of world total | MOST Policy Initiative, April 2026 |
| Typical 100 MW AI data center (annual water use, evaporative cooling) | 1.5–3.0 million cubic meters/year | Hyperscaler disclosures |
| Share of water that evaporates (does not return to supply) | Up to 85% | MOST Policy Initiative |
| AI data centers vs. traditional: water consumption multiple | 10–50x more | Industry estimates |
| Google AI facilities: average daily water use | ~550,000 gallons per facility | Introl, 2026 |
| GPT-3 training run: estimated water evaporated | 700,000 liters | Research estimates |
| Average rack density trend: 2023 → 2027 | 36 kW → 50 kW | Net Zero Insights, 2025 |
| Projected NVIDIA Rubin Ultra NVL576 rack TDP (2027) | Up to 600 kW | NVIDIA GTC 2025 |
| Data centers facing high water stress by 2050 | ~45% of current facilities | Assessment of 9,055 facilities |
Sources: MOST Policy Initiative, April 2026; Net Zero Insights, November 2025.
The MOST Policy Initiative’s April 2026 science note highlights that water and energy efficiency are structurally in tension: evaporative cooling (water-intensive) is more energy-efficient, while air cooling (energy-intensive) conserves water. As rack densities climb toward 50 kW by 2027 — and potentially 600 kW per rack with next-generation NVIDIA Rubin Ultra systems — both cooling dimensions will face unprecedented stress.
Microsoft is deploying closed-loop, zero-water evaporation cooling across new facilities, eliminating evaporative water use entirely. Each such facility reduces annual consumption by more than 125 million liters versus evaporative designs. Google has pledged to replenish 120% of water consumed by 2030.
Energy Mix for U.S. Data Centers (IEA, 2025 Base Case)
| Energy Source | Current Share | Notes |
|---|---|---|
| Natural gas | ~50% (dominant) | Onsite generation increasingly common |
| Renewables | ~30% and growing | Tech sector = ~40% of all corporate renewable PPAs in 2025 |
| Coal | ~30% global average | Highest in China |
| Nuclear | Growing | SMR pipeline: 45 GW conditional agreements as of April 2026 |
Source: IEA (2025), Energy and AI — Energy Supply for AI section.
Regional Data Center Market Statistics
North America
| Market | Capacity | Key Stat |
|---|---|---|
| Northern Virginia | 5.6 GW | Largest market; 3x second-largest |
| Dallas-Fort Worth | 1.5 GW | Second-largest U.S. market |
| North America total (colocation inventory, H1 2025) | 15.5 GW | Record high |
| North America under construction (end-2024) | 6.6 GW | 78% in primary markets |
| North America: $1 trillion projected investment 2025–2030 | $1 trillion | JLL North America |
Source: JLL North America Data Center Report, Midyear 2025; JLL Year-End 2024.
Europe (EMEA / FLAP-D Markets)
| Market | Vacancy (Q4 2025) | Notes |
|---|---|---|
| FLAP-D weighted average (Frankfurt, London, Amsterdam, Paris, Dublin) | 6.3% | Record low; down from 16.9% in 2021 |
| Ireland | Grid connection moratorium lifted December 2025 | New requirement: onsite generation for grid connection |
Source: JLL EMEA Year-End Data Centre Report 2025.
Asia-Pacific
| Market | Value | Timeframe |
|---|---|---|
| APAC current capacity | 32 GW | 2025 |
| APAC projected capacity | 57 GW | 2030 |
| APAC CAGR | 12% | 2025–2030 |
| Asia Pacific market revenue | $61.02 billion | 2025 |
Source: JLL 2026 Global Data Center Outlook; Fortune Business Insights, 2026.
AI Data Center Market Size
The “data center market” is measured by multiple methodologies (colocation revenue, total infrastructure spend, systems spending), which produces divergent but complementary figures:
| Metric | 2025 Value | 2026 Projected | 2030 Projected | Source |
|---|---|---|---|---|
| Global data center market (Fortune BI) | $269.79 billion | $300.64 billion | $699.13 billion | Fortune Business Insights, 2026 |
| Internet data center market (IDC methodology) | $77.98 billion | $89.01 billion | $149.58 billion | Globe Newswire / Research and Markets, May 2026 |
| Data center systems spending (Gartner) | $489.5 billion | — | — | Gartner, cited in Cargoson compilation |
| AI infrastructure only (McKinsey 2030 total) | — | — | $6.7 trillion cumulative | McKinsey, April 2025 |
| Global data center real estate asset value creation (2026–2030) | — | — | $1.2 trillion | JLL 2026 Global Outlook |
Note: Figures vary by scope. “Data center market” in Fortune Business Insights includes physical infrastructure investment; “internet data center market” in the IDC/Research and Markets report covers hosted and colocation services revenue only. Gartner’s “data center systems spending” captures hardware and software procurement. McKinsey’s $6.7 trillion covers cumulative AI-driven capex through 2030. All figures are as reported by named issuing organizations.
Methodology
Data Collection
This dataset was compiled by the Axis Intelligence Research Desk from primary-source reports published between January 2025 and June 2026. Primary sources include: the International Energy Agency (IEA), Lawrence Berkeley National Laboratory (LBNL) / U.S. Department of Energy, Goldman Sachs Research, JLL, NVIDIA Corporation SEC filings, and official announcements from hyperscaler companies (earnings calls and press releases filed with or cited in SEC Form 8-K filings).
No statistics in this article originate from secondary tech blog compilations. Where a statistic appeared in secondary coverage, Axis Intelligence traced it to its primary issuing source before citation.
Cross-Source Derivation
The Axis Intelligence AI Infrastructure Pressure Index™ (AIPI) and the implied U.S. data center electricity consumption figure for 2026 (~250–265 TWh) are original Axis Intelligence derivations not published by any single source. The AIPI methodology is documented at axis-intelligence.com/research/aipi-methodology/. The cross-source consumption figure is derived by applying Goldman Sachs’ GW-based U.S. demand model (41 GW for 2026) against IEA’s stated U.S. capacity share (~45% of global AI data center capacity) and IEA’s global TWh forecast, producing an implied range that will be validated against IEA mid-year data when published.
Limitations
- IEA figures for 2026 are estimates as of the April 2026 report; actual figures will be updated in the IEA’s next data center report.
- Goldman Sachs U.S. GW figures assume 70% capacity utilization; actual utilization varies by facility type.
- Hyperscaler capex figures represent guidance from earnings calls as of Q1 2026; actual spending may differ.
- Water consumption figures rely on voluntary disclosure from hyperscalers and are not independently verified by a regulatory body.
- Construction cost figures (JLL) represent shell-and-core only; total cost of ownership including fit-out and hardware is substantially higher.
Update Cadence
This article is reviewed and updated quarterly. Next update: September 2026 (Q3 2026), incorporating IEA mid-year data, Q2 2026 hyperscaler earnings, and updated JLL vacancy metrics.
About This Dataset
Dataset title: AI Data Center Statistics 2026 — Global Electricity, Investment, Capacity, and Environmental Impact
Issued by: Axis Intelligence Research Desk
Coverage: Global; with U.S., EMEA, and APAC regional breakdowns
Base period: 2023–2026 (with projections to 2030)
Last updated: June 4, 2026
Next scheduled update: September 2026
License: Creative Commons Attribution 4.0 International (CC BY 4.0). You are free to share and adapt this data for any purpose, provided you attribute Axis Intelligence.
Download: ai-data-center-statistics-2026.csv
Cite This Research
APA: Axis Intelligence Research Desk. (2026, June 4). AI Data Center Statistics 2026: Electricity, Investment, Capacity, and the Power Crisis. Axis Intelligence. https://axis-intelligence.com/ai-data-center-statistics/
MLA: Axis Intelligence Research Desk. “AI Data Center Statistics 2026: Electricity, Investment, Capacity, and the Power Crisis.” Axis Intelligence, 4 June 2026, axis-intelligence.com/ai-data-center-statistics/.
Chicago: Axis Intelligence Research Desk. “AI Data Center Statistics 2026: Electricity, Investment, Capacity, and the Power Crisis.” Axis Intelligence, June 4, 2026. https://axis-intelligence.com/ai-data-center-statistics/.
→ Copy any citation format above and paste it directly. For embed code, see the “Cite This Research” block below.
Frequently Asked Questions
How much electricity do AI data centers consume globally?
Global data center electricity consumption reached approximately 485 TWh in 2025, a 17% increase year-over-year. AI-focused data centers specifically grew 50% faster. By 2030, the IEA projects global data center electricity demand will reach approximately 950 TWh — double the 2025 level — accounting for roughly 3% of all global electricity.
How much are hyperscalers spending on AI data centers in 2026?
The Big Five hyperscalers — Amazon, Alphabet (Google), Meta, Microsoft, and Oracle — are collectively projected to spend approximately $725 billion in capital expenditure in 2026. This represents a 77% increase from the record $410 billion spent in 2025. Amazon leads with $200 billion planned, followed by Alphabet at $175–185 billion, Meta at $115–135 billion, Microsoft at approximately $190 billion (revised), and Oracle at approximately $50 billion.
What percentage of U.S. electricity do data centers consume?
U.S. data centers consumed approximately 183 TWh in 2024, representing roughly 4% of total U.S. electricity generation, according to IEA estimates cited by Pew Research. The Lawrence Berkeley National Laboratory projects this to rise to between 6.7% and 12% of U.S. electricity by 2028 — between 325 and 580 TWh.
What is the data center capacity shortage in the U.S.?
Goldman Sachs Research estimates the U.S. faces a structural data center power shortfall of approximately 9.3 GW in 2026, worsening to an estimated 45 GW cumulative gap by 2028. The average wait time for a new U.S. grid connection is now four years. Northern Virginia, the largest U.S. data center market, has a colocation vacancy rate below 1%.
What is the Stargate Project?
Project Stargate is a joint venture announced in January 2025, backed by OpenAI, SoftBank, Oracle, and MGX, with a $500 billion four-year commitment to build 10 GW of U.S. AI data center capacity. As of September 2025, Stargate had secured commitments for nearly 7 GW of planned capacity, ahead of schedule. Its flagship facility in Abilene, Texas, was operational by mid-2025.
How much water do AI data centers consume?
A typical 100 MW AI data center using evaporative cooling consumes between 1.5 and 3.0 million cubic meters of water per year. AI-focused data centers consume 10–50x more cooling water than traditional server farms. Google’s AI facilities average approximately 550,000 gallons of water per day. Up to 85% of the water data centers use evaporates and does not return to the water supply.
What is the AI Infrastructure Pressure Index™ (AIPI)?
The AIPI is an original Axis Intelligence composite index measuring structural tension between AI data center demand growth and physical delivery capacity. It combines three normalized sub-scores: Demand Growth, Supply Constraint, and Capital Deployment — each weighted equally. As of Q2 2026, the AIPI stands at 84/100, its highest reading since Axis Intelligence began tracking in Q1 2025 (baseline: 58), signaling a high-pressure environment in which physical infrastructure is the primary brake on AI scaling.
What is driving data center construction cost increases?
Average global data center construction costs (shell and core) increased from $7.7 million per MW in 2020 to $10.7 million per MW in 2025, a 39% rise driven by land scarcity, power infrastructure costs, and materials inflation. JLL projects a further 6% increase to $11.3 million per MW in 2026. AI-optimized facilities require an additional technology fit-out of up to $25 million per MW beyond the shell-and-core cost.
How are AI companies responding to the energy challenge?
The technology sector accounted for approximately 40% of all corporate renewable power purchase agreements (PPAs) signed globally in 2025. The SMR (small modular reactor) pipeline with data center operators grew from 25 GW to 45 GW between end-2024 and April 2026. Microsoft is deploying zero-water evaporation cooling across new facilities, and Google has pledged to replenish 120% of water consumed by 2030.
What is NVIDIA’s role in AI data center growth?
NVIDIA’s Blackwell GPU architecture is the primary compute platform for AI data centers. NVIDIA reported $75.2 billion in data center revenue in Q1 FY2027 (April 2026), a 92% year-over-year increase — the largest quarterly data center revenue figure in semiconductor industry history. Blackwell Ultra GPUs deliver up to 50x better performance for agentic AI than the prior Hopper platform. H100 and H200 GPU lead times remain at 36–52 weeks due to HBM memory and CoWoS packaging supply constraints.
Embed This Research
Copy the HTML block below to display the Axis Intelligence AI data center chart on your site. Every embed includes an automatic do-follow attribution link.
📊 Data visualization by
AI Data Center Statistics 2026 — Axis Intelligence ResearchGlobal AI data center electricity consumption: 485 TWh in 2025 → 950 TWh projected by 2030. Big Five hyperscaler capex: $725 billion in 2026.
Source: axis-intelligence.com/ai-data-center-statistics/ | License: CC BY 4.0 | Updated: June 2026
Every site that embeds this block links directly to the canonical URL with a do-follow attribute, building Axis Intelligence’s citation network.
Related Axis Intelligence Research
- AI Agents Statistics 2026 — Axis Intelligence Research
- Cybersecurity Statistics 2026 — Axis Intelligence Research
Axis Intelligence Research produces institutional-grade data research published under CC BY 4.0. For citation inquiries, dataset licensing, or to report a data error, contact editorial@axis-intelligence.com.