AI Build-Out Visualization Project — New Consensus
Published November 2026. Fixed research snapshot, not a live feed.

SITES AND TIMELINES
Epoch AI, AI data centers. Snapshot updated November 1, 2026; downloaded November 1–2, 2026.
https://epoch.ai/data/ai-data-centers
https://epoch.ai/data/data-centers-documentation/methodology
https://epoch.ai/data/data-centers-documentation/records
Source license: Creative Commons Attribution; credit Epoch AI.

sites.csv: normalized 96-site comparison table. Current positive estimated IT capacity in 76 sites; 20 have no positive current estimate.
epoch-original.csv: preserved source inventory with source links and qualifications.
epoch-timelines.csv: original dated records, including future projections. The site's historical charts EXCLUDE future-dated rows.
history.csv: last dated IT-power observation carried forward for each site at five checkpoints. The current cohort is fixed. Missing earlier observations contribute no recorded capacity, not a physical zero. This is not a global historical census.
snapshot.json: chart data and site explorer records used by this edition.

H100 equivalents compare theoretical dense 8-bit performance. They are not physical chip counts. Cost is modeled from IT power and is not independent corroboration of the power estimate. Cost figures include hardware and facility capital, in 2025 dollars. IT power excludes facility overhead. A site may encompass multiple buildings; it is not a construction-phase count.

WATER
Google, 2026 Environmental Report, page 97; methodology page 87. Reporting year 2025.
https://sustainability.google/reports/google-2026-environmental-report/
water.csv: 40 named locations plus an aggregate residual row. Source footnote numbers retained. Charts and the named-location table exclude the residual row. Consumption = withdrawal minus discharge. Seawater and brackish water are excluded. All workloads at each reporting location, not AI alone. Reporting boundaries may cover multiple campuses. Low totals can reflect new or partially operating facilities.

LAND
land.csv: dated, disclosed site boundaries. Not floor space; later expansions excluded.
Louisiana Economic Development, January 8, 2025: original Meta Richland Parish section of Franklin Farm, 2,250 acres.
https://www.opportunitylouisiana.gov/news/louisiana-earns-deal-of-the-year-award-for-historic-10-billion-meta-ai-data-center-project-in-northeast-louisiana
Microsoft, September 18, 2025: initial Fairwater Wisconsin campus, 315 acres.
https://blogs.microsoft.com/blog/2025/09/18/inside-the-worlds-most-powerful-ai-datacenter/

GLOBAL ELECTRICITY
IEA, Key Questions on Energy and AI (2026). All data-center workloads, not AI alone.
2025: 485 TWh estimated. 2030: 950 TWh projected. Projection is not built capacity.
https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary

GEOGRAPHY
Natural Earth 1:110m admin-0 countries. Public domain.
https://www.naturalearthdata.com/about/terms-of-use/
https://github.com/nvkelso/natural-earth-vector/blob/master/geojson/ne_110m_admin_0_countries.geojson
Countries are highlighted when the inventory contains a site record. No invented campus coordinates. An unhighlighted country is not assigned zero infrastructure.

COVERAGE
Epoch's 45% central coverage estimate is of deployed global AI compute, with a 90% interval of 30–78%. It is not site-count coverage. Neither a worldwide site total nor an AI-boom project-count coverage percentage has been established here.

Research scope: worldwide additions from January 2023, including completed sites, phased expansions and AI conversions; earlier infrastructure as baseline. This edition remains a sample. Do not add the different source populations or equate their site boundaries without checking them.

Analysis and graphics by New Consensus. Cite the underlying sources alongside any reused figure. Downloadable SVG and PNG versions accompany each figure on the project page.

CITY-SCALE AGGREGATE (edition 2, November 2, 2026)
city-scale.json contains inputs, derived comparisons and source URLs. The combined graphic describes the whole global data-center sector, including AI, not an AI-only or post-2022 additions total. Land: Hines modeled 20,000 acres of powered operating sites at end-2024; entire acquired campus estates remain uncounted. Electricity: IEA 485 TWh in 2025 = 55.4 GW average, compared with NYC approximately 50 TWh/year. Water: IEA 560 billion liters consumed in 2023, including energy supply and chip manufacture; roughly 140 billion liters for direct cooling. NYC delivered 363.4 billion US gallons to the city in 2025. Conversion yields 148.6 days of NYC supply volume, with 37.1 days for direct cooling. This volume analogy does not equate consumption with municipal withdrawals or local water impacts. Manhattan benchmark: 22.7 square miles, or 14,528 acres.