← ALL DATASETS
DATASET / epoch_ai_data_center_timelines / V20261119

AI data center timelines

Dated AI site milestones, including estimated future capacity and costs. Explore the original source values, notes and links alongside related tables. Figures may be estimates; this dated export is not a complete global inventory.

OPEN IN QUERY ↗AI / data centers / technology
SOURCEEpoch AI ↗
GEOGRAPHYGlobal; selected AI sites, concentrated in the United States
TIMEMilestones 2018-12-18–2030-01-01; snapshot 2026-11-19
UPDATED2026-11-19

VISUAL EXPLORATION

Table preview is the curated default for this dataset.

TABLE PREVIEW

Loading preview…

METADATA

DATA DICTIONARY

data_centerData center · string · dimension

Site or campus name; joins to data_center in epoch_ai_data_centers.

dateDate · date · dimension

As-of date of the source milestone or chip record; may refer to an estimated future event.

construction_statusConstruction status · string · dimension

Source narrative of construction progress at this milestone, including reasoning, uncertainty and citations.

buildings_operationalBuildings operational · number · measure

Number of buildings operational by this milestone date, according to the source.

it_power_mwIT power (MW) · number · measure

IT power capacity available to GPU servers, networking and storage, excluding facility overhead.

facility_power_mwPower (MW) · number · measure

Total facility power capacity, including IT, cooling, lighting and other overhead; not energy consumption.

h100_equivalentsH100 equivalents · number · measure

AI compute capacity expressed as equivalent NVIDIA H100 GPUs based on 8-bit OP/s.

performance_8bit_opsPerformance (8-bit OP/s) · number · measure

Maximum AI compute capacity in a format of at least 8 bits; source estimates may infer chip counts from power.

total_capital_cost_2025_usd_billionsTotal capital cost (2025 USD billions) · number · measure

Estimated total capital cost by this date, including IT hardware, facility equipment, buildings, labor and land.

compute_cost_2025_usd_billionsCompute cost (2025 USD billions) · number · measure

Estimated capital cost of IT hardware including GPUs, networks and server racks.

construction_cost_2025_usd_billionsConstruction cost (2025 USD billions) · number · measure

Estimated capital cost excluding compute hardware, including buildings, electrical and cooling equipment, labor and land.

annual_operating_cost_2025_usd_billionsAnnual operating cost (2025 USD billions) · number · measure

Estimated annual operating cost including energy, tax, maintenance, labor and water.

water_use_mgdWater use (MGD) · number · measure

Estimated average water consumption in million gallons per day; source unit retained, no gallon-system conversion.

current_h100_equivalentsCurrent H100 equivalents (from Data center) · number · measure

Repeated lookup of the site’s current compute at the export snapshot, not its value at this row’s milestone date.

PROVENANCE & LIMITATIONS

Run epoch-ai-data-centers-20261119. Human-supplied https://epoch.ai/data/data_centers/data_centers.zip, ZIP SHA-256 9fb68730427e3610d684ea46b4d05465063f5d1d928b63bb34f9a3589bedfefa; data_center_timelines.csv extracted unchanged. CSV SHA-256 83c4bffe13ead07957fdf3f0f28d8ee98bcc445b1e5e034b65ff9964548e7261. 575 input rows, 575 output rows, zero rows or columns removed. All source text retained, including Notes and Construction status. Header renaming, strict types and empty-cell nulls only. No labels or filled periods added. Field reference https://epoch.ai/data/data-centers-documentation/records. Decision docs/decisions/0018-epoch-ai-data-centers.md. Citation: Epoch AI, AI data centers, https://epoch.ai/data/ai-data-centers. Attribution and CC BY 4.0 apply.

Processing code: pipeline/sources/epoch_ai/process-data-centers.py

  • A dated snapshot of selected AI infrastructure, not a complete global census. Coverage is concentrated in the United States; country counts do not measure total national capacity.
  • Figures include estimates based on imagery, permits, company disclosures and other evidence. Past milestone dates do not establish that estimates were realized. Future values are uncertain and may change.
  • No annual series, interpolation, extrapolation, or forecast/observed labels are generated. The latest timeline date is 2030-01-01 for one site only; most sites do not have milestones through 2030.
  • Source Current fields are frozen at the export snapshot, not refreshed when CURRENT_DATE changes. They must not be summed across repeated timeline rows.
  • MW denotes power capacity, not electricity consumption. IT power excludes cooling and facility overhead; total facility power includes them. Costs are estimates in billions of 2025 USD.
  • Owner denotes chip ownership, not necessarily facility ownership. Text confidence markers, evidence, analysis and citations are retained verbatim; their claims remain those of the source and may become outdated.
  • Empty cells are null, not zero or false. Comma-separated source labels have not been split or normalized. Source country names have no added ISO or region mapping.