Global data centre electricity demand hit roughly 485 TWh in 2025, up 17% in a single year, and the IEA projects it roughly doubling to about 950 TWh by 2030. The physical build behind that number is equally steep: 1,297 operational hyperscale facilities as of late 2025 with 770 more in the pipeline, and combined capex from the six largest operators projected above USD 600 billion for 2026. Inside the buildings, average modal rack density crossed 11 kW for the first time. Meanwhile more than half of operators cannot fill open roles. The figures below come from the IEA, Synergy Research Group, and Uptime Institute’s 16th annual survey.
TL;DR
- Global data centre electricity demand reached roughly 485 TWh in 2025 (IEA)
- That is a 17% year-over-year increase (IEA)
- The IEA projects roughly 950 TWh by 2030 (IEA)
- That would be around 3% of global electricity demand (IEA)
- AI-specific data centres grew about 50% in 2025 (IEA)
- 1,297 hyperscale data centers were operational in late 2025 (Synergy Research)
- 770 more sit in the pipeline (Synergy Research)
- Hyperscaler capex is projected above USD 600 billion in 2026 (Synergy Research)
- Hyperscale capacity is expected to double in just over 12 quarters (Synergy Research)
- Average modal rack density passed 11 kW for the first time (Uptime Institute)
- Nearly a quarter of operators report racks above 30 kW, up from 19% (Uptime Institute)
- Outage frequency per site fell for a fifth consecutive year (Uptime Institute)
- More than half of operators struggle to fill open roles (Uptime Institute)
1. The Power Bill
Electricity is the constraint that now shapes every other decision in this industry. Global data centre demand reached roughly 485 TWh in 2025, up 17%, and is projected to reach about 950 TWh by 2030, roughly 3% of world electricity. AI-specific facilities grew far faster than the aggregate at about 50% in 2025, and the IEA expects that subset to triple across the forecast window.
| Metric | Value | Source |
|---|---|---|
| Global data centre electricity demand, 2025 | approx. 485 TWh | IEA |
| Year-over-year growth | 17% | IEA |
| Projected demand, 2030 | approx. 950 TWh | IEA |
| Share of global electricity demand by 2030 | around 3% | IEA |
| Growth in AI-specific data centres, 2025 | approx. 50% | IEA |
| Expected change in AI facility consumption to 2030 | tripling | IEA |
| Projected growth in accelerated server consumption | approx. 30% annually | IEA |
| IEA analysis publication | April 2026 | IEA |
The IEA flags a caveat most coverage drops: bottlenecks across the supply chain reduce the likelihood of the most aggressive near-term scenarios, so the projection is a base case rather than a ceiling. Source: IEA Electricity 2026 demand analysis.
2. The Build
Physical capacity is expanding at a pace with no precedent in the sector’s history. Synergy counted 1,297 operational hyperscale data centers in late 2025, nearly triple the early-2018 figure, with 770 more in the pipeline. Combined capex above USD 600 billion for 2026 from six operators is what sustains a doubling of total hyperscale capacity in just over 12 quarters.
| Metric | Value | Source |
|---|---|---|
| Operational hyperscale data centers, late 2025 | 1,297 | Synergy Research |
| Change versus early 2018 | nearly triple | Synergy Research |
| Facilities in the pipeline | 770 | Synergy Research |
| Projected 2026 capex, six largest operators | more than USD 600 billion | Synergy Research |
| Operators counted in that capex figure | AWS, Microsoft, Google, Meta, Oracle, Alibaba | Synergy Research |
| Expected time for hyperscale capacity to double | just over 12 quarters | Synergy Research |
| Hyperscale share of all capacity by 2031 | 67% | Synergy Research |
| Pipeline as a share of operational base | approx. 59% | Derived from Synergy figures |
A pipeline equal to nearly 60% of the operational base is the number that makes the capacity forecast credible: this is committed construction rather than projection. Source: Synergy Research on hyperscale data center count.
3. Rack Density Crossed a Threshold
The inside of the buildings changed faster than the count of buildings. Average modal rack density passed 11 kW for the first time, and nearly a quarter of operators now report at least some racks above 30 kW, up from 19% a year earlier. AI training is driving the top of that distribution while conventional enterprise workloads still account for most facilities below 30 kW.
| Metric | Value | Source |
|---|---|---|
| Average modal rack density | above 11 kW, a first | Uptime Institute, 2026 |
| Operators reporting some racks above 30 kW | nearly one quarter | Uptime Institute, 2026 |
| Same figure a year earlier | 19% | Uptime Institute, 2026 |
| Year-over-year change | approx. 6 points | Derived from Uptime figures |
| Primary driver of the highest densities | AI training workloads | Uptime Institute, 2026 |
| Primary driver below 30 kW | traditional enterprise applications | Uptime Institute, 2026 |
| Survey respondents | more than 800 owners and operators | Uptime Institute, 2026 |
| Fieldwork period | April to May 2026 | Uptime Institute, 2026 |
Modal density crossing 11 kW matters more than any single high-density showcase facility, because the mode describes the typical rack rather than the exceptional one. Source: Uptime Institute 16th Annual Global Data Center Survey.
4. Outages: Rarer, Costlier
The reliability trend splits cleanly in two directions, which is why single-sentence summaries of it are usually wrong. Outage frequency on a per-site basis declined for a fifth consecutive year, while one in ten outages is still serious or severe and the cost of those failures keeps climbing sharply. Fewer incidents, each with higher consequences, is the pattern of an industry concentrating more workload into fewer, larger facilities.
| Metric | Value | Source |
|---|---|---|
| Consecutive years of falling per-site outage frequency | 5 | Uptime Institute, 2026 |
| Pace of that improvement | slowing | Uptime Institute, 2026 |
| Share of outages classified serious or severe | one in ten | Uptime Institute, 2026 |
| Direction of outage costs | climbing sharply | Uptime Institute, 2026 |
| Top management concern | rising costs | Uptime Institute, 2026 |
| Leading technical causes cited | network and power failures | Uptime Institute, 2026 |
| Report covering this analysis | Annual Outage Analysis 2026 | Uptime Institute |
| Corporate IT off premises | majority, a first | Uptime Institute, 2026 |
The crossing point where the majority of corporate IT now sits off premises is a first for this survey and helps explain the concentration effect behind rising per-incident cost. When a workload lived in a company’s own server room, an outage inconvenienced one company; when it lives in a hyperscale region serving thousands of tenants, the same failure propagates across every one of them at once. Falling incident counts and rising incident costs are therefore two views of the same consolidation, not competing signals about whether reliability is improving. Silicon supply context sits in our semiconductor industry statistics. Source: Uptime Institute Annual Outage Analysis 2026.
5. Efficiency and the People Problem
Two constraints get less coverage than power and both bind hard. Average Power Usage Effectiveness improved only marginally in 2026, held back by legacy infrastructure, while more than half of operators reported difficulty finding qualified candidates. Efficiency and staffing share a root cause: the industry is adding capacity faster than it can modernise the existing estate or train the people to run it.
| Metric | Value | Source |
|---|---|---|
| Change in average PUE, 2026 | minor improvement | Uptime Institute, 2026 |
| Main brake on efficiency progress | legacy infrastructure | Uptime Institute, 2026 |
| Operators reporting hiring difficulty | more than half | Uptime Institute, 2026 |
| Nature of the retention problem | staff recruited away by other operators | Uptime Institute, 2026 |
| Survey edition | 16th annual | Uptime Institute, 2026 |
| Respondents | more than 800 | Uptime Institute, 2026 |
| Global data centre electricity demand for context | approx. 485 TWh | IEA |
| Projected 2026 hyperscaler capex for context | more than USD 600 billion | Synergy Research |
Marginal PUE gains against 17% annual demand growth means absolute energy consumption rises regardless of efficiency work, which is the arithmetic that efficiency-focused messaging tends to obscure. Broader energy context sits in our AI energy statistics and enterprise deployment in our enterprise AI adoption statistics. Source: Data Center Knowledge on the Uptime 2026 survey.
Summary: Data Centers by the Numbers
| Metric | Value | Source |
|---|---|---|
| Global data centre electricity demand, 2025 | approx. 485 TWh | IEA |
| Year-over-year growth | 17% | IEA |
| Projected demand, 2030 | approx. 950 TWh | IEA |
| Share of global electricity by 2030 | around 3% | IEA |
| AI data centre growth, 2025 | approx. 50% | IEA |
| Operational hyperscale facilities | 1,297 | Synergy Research |
| Facilities in the pipeline | 770 | Synergy Research |
| Projected 2026 hyperscaler capex | more than USD 600 billion | Synergy Research |
| Time for hyperscale capacity to double | just over 12 quarters | Synergy Research |
| Hyperscale share of capacity by 2031 | 67% | Synergy Research |
| Average modal rack density | above 11 kW | Uptime Institute |
| Operators with racks above 30 kW | nearly one quarter | Uptime Institute |
| Same figure a year earlier | 19% | Uptime Institute |
| Consecutive years of falling outage frequency | 5 | Uptime Institute |
| Share of outages serious or severe | one in ten | Uptime Institute |
| Direction of outage costs | climbing sharply | Uptime Institute |
| Change in average PUE | minor improvement | Uptime Institute |
| Operators reporting hiring difficulty | more than half | Uptime Institute |
| Corporate IT off premises | majority, a first | Uptime Institute |
| Survey respondents | more than 800 | Uptime Institute |
| Survey fieldwork | April to May 2026 | Uptime Institute |
Methodology and Sources
- Electricity demand, growth rates, and 2030 projections come from IEA analysis published in April 2026 (Electricity 2026 demand, energy demand from AI, energy supply for AI).
- Hyperscale facility counts, pipeline, capex, and capacity forecasts come from Synergy Research Group (hyperscale data center count, capacity doubling analysis, 2031 share forecast).
- Rack density, PUE, outage frequency and severity, staffing, and the off-premises crossover come from Uptime Institute’s 16th Annual Global Data Center Survey, based on more than 800 owner and operator responses collected April to May 2026 (survey results, outage analysis 2026, outage report announcement), with independent coverage cross-checked (Data Center Knowledge, Network World).
- Data watch: the IEA 2030 figure is a base-case projection, not a forecast the agency treats as certain, and the IEA itself notes that supply chain bottlenecks make more aggressive scenarios less likely. Uptime’s survey is self-reported by operators who choose to participate, which skews toward larger and more professionalised facilities and likely understates problems in the long tail of small enterprise rooms. Several Uptime findings here are qualitative in the published summary, such as “minor improvement” in PUE and “climbing sharply” for outage costs, and are reported as such rather than converted into false precision. Synergy’s hyperscale definition covers a specific operator set and excludes colocation capacity leased to others. Rows marked as derived are arithmetic on published figures.
- Last updated: August 2, 2026. We update this roundup quarterly as the IEA, Synergy, and Uptime Institute publish new data.