AI Data Center Statistics
A snapshot of the global AI infrastructure buildout, drawn from 344 tracked facilities across 64 countries. Updated automatically on every data deploy. Download the full dataset →
Global Totals
By Build Status
| Status | Facilities | GW Tracked | Share |
|---|---|---|---|
| Operational | 195 | 53.4 GW | 57% |
| Under Construction | 62 | 39.4 GW | 18% |
| Planned | 65 | 35.2 GW | 19% |
| Announced | 22 | 27.6 GW | 6% |
By Energy Source
| Energy Type | Facilities | GW Tracked | Share |
|---|---|---|---|
| Renewable | 125 | 43.2 GW | 36% |
| Mixed | 55 | 7.4 GW | 16% |
| Nuclear | 17 | 31.7 GW | 5% |
| Hydroelectric | 11 | 3.6 GW | 3% |
| Fossil / Gas | 6 | 17.8 GW | 2% |
| Unknown | 130 | 52.0 GW | 38% |
By Capacity Tier
| Capacity Tier | Facilities | GW Tracked | Share |
|---|---|---|---|
| Under 100 MW | 99 | 4.1 GW | 29% |
| 100-500 MW | 146 | 30.4 GW | 42% |
| 500 MW to 1 GW | 44 | 26.4 GW | 13% |
| Over 1 GW | 46 | 94.7 GW | 13% |
| Capacity Unknown | 9 | — | 3% |
By AI Focus
| AI Focus | Facilities | GW Tracked | Share |
|---|---|---|---|
|
Training
Facilities used for frontier model training, large GPU clusters, and pretraining workloads.
|
34 | 35.5 GW | 10% |
|
Inference
Facilities optimized for serving models, real-time AI, and enterprise inference workloads.
|
44 | 6.9 GW | 13% |
|
Research
Academic, national lab, and R&D systems focused on experimentation and scientific AI.
|
3 | 0.1 GW | 1% |
|
Sovereign AI
National and regional facilities built for domestic AI capability, public-sector workloads, and digital sovereignty.
|
5 | 1.9 GW | 1% |
|
General Cloud
Cloud regions and multipurpose platforms that blend AI capacity with broader cloud infrastructure.
|
15 | 2.8 GW | 4% |
|
HPC / Supercomputing
High-performance computing facilities supporting exascale, supercomputing, and hybrid AI-HPC workloads.
|
6 | 0.2 GW | 2% |
|
Mixed
Facilities with multiple public AI roles, such as training plus inference or cloud plus sovereign AI.
|
115 | 46.0 GW | 33% |
|
AI Focus Unknown
Facilities where the public AI workload focus has not been clearly disclosed.
|
122 | 62.4 GW | 35% |
Separates training clusters from inference regions, sovereign builds, and research supercomputers. View AI focus taxonomy →
By Timeline Year
| Year | Facilities | Countries | GW Tracked |
|---|---|---|---|
| 2032 | 1 | 1 | 0.7 GW |
| 2030 | 1 | 1 | 1.2 GW |
| 2029 | 1 | 1 | 2.7 GW |
| 2028 | 3 | 2 | 5.5 GW |
| 2027 | 1 | 1 | 4.5 GW |
| 2026 | 29 | 20 | 13.8 GW |
| 2025 | 26 | 18 | 14.3 GW |
| 2024 | 36 | 22 | 20.9 GW |
| 2023 | 26 | 17 | 4.7 GW |
| 2022 | 17 | 13 | 2.2 GW |
| 2021 | 13 | 10 | 3.5 GW |
| 2020 | 13 | 8 | 3.0 GW |
Uses the best currently available public milestone year per facility. Browse the full year taxonomy →
Top Operators by Tracked Capacity
| # | Operator | Facilities | GW Tracked |
|---|---|---|---|
| 1 | Microsoft | 56 | 25.1 GW |
| 2 | Amazon Web Services | 43 | 21.2 GW |
| 3 | 42 | 13.3 GW | |
| 4 | NVIDIA | 21 | 11.9 GW |
| 5 | Meta | 19 | 16.8 GW |
| 6 | Oracle | 11 | 12.3 GW |
| 7 | Equinix | 10 | 0.6 GW |
| 8 | Hewlett Packard Enterprise | 8 | 0.2 GW |
| 9 | EuroHPC JU | 6 | 0.0 GW |
| 10 | G42 | 5 | 10.2 GW |
| 11 | OpenAI | 5 | 16.3 GW |
| 12 | Vantage Data Centers | 4 | 2.0 GW |
| 13 | Alibaba Cloud | 4 | 0.8 GW |
| 14 | Cassava Technologies | 4 | — |
| 15 | Eviden (Atos) | 4 | 0.0 GW |
Top 15 operators by tracked capacity. View all operators →
Top Countries by Facility Count
| # | Country | Facilities | GW Tracked |
|---|---|---|---|
| 1 | United States | 84 | — |
| 2 | India | 11 | — |
| 3 | Germany | 11 | — |
| 4 | South Korea | 11 | — |
| 5 | Japan | 10 | — |
| 6 | France | 9 | — |
| 7 | United Kingdom | 9 | — |
| 8 | United Arab Emirates | 8 | — |
| 9 | Australia | 8 | — |
| 10 | Singapore | 8 | — |
| 11 | Saudi Arabia | 8 | — |
| 12 | Canada | 8 | — |
| 13 | Thailand | 7 | — |
| 14 | Malaysia | 7 | — |
| 15 | South Africa | 7 | — |
| 16 | China | 7 | — |
| 17 | Brazil | 5 | — |
| 18 | Taiwan | 5 | — |
| 19 | Ireland | 5 | — |
| 20 | Indonesia | 5 | — |
Top 20 countries by tracked facility count. Browse all 64 countries →
Coverage Note
Capacity figures reflect what has been publicly disclosed — not total build costs or contracted power. Many facilities are missing capacity data (marked as "—") because operators have not disclosed specific megawatt figures. GW totals are thus conservative lower bounds, not estimates of true total capacity. Energy type is based on published PPAs, operator commitments, and grid data; where none is available the facility is marked Unknown.