A 1 gigawatt (GW) AI data center can consume 8.76 terawatt-hours (TWh) of electricity annually, generating an electricity bill of approximately $876 million at $0.10 per kilowatt-hour (kWh) if it operates continuously at 1 GW of total facility demand.
Amazon, Microsoft, Google, Meta, Oracle, xAI and specialist AI cloud providers are developing increasingly power-intensive infrastructure to support AI training, inference and cloud computing.
Some announced AI data center campuses could eventually require several gigawatts of electricity. However, planned power capacity does not represent actual electricity consumption until facilities become operational.
How Much Electricity Does a 1 GW AI Data Center Consume?
A data center drawing 1,000 megawatts continuously consumes 24 gigawatt-hours (GWh) daily and 8.76 TWh annually.
| Continuous facility power demand | Annual electricity consumption | Annual cost at $0.10/kWh |
| 100 MW | 0.876 TWh | $87.6 million |
| 250 MW | 2.19 TWh | $219 million |
| 500 MW | 4.38 TWh | $438 million |
| 1 GW | 8.76 TWh | $876 million |
| 2 GW | 17.52 TWh | $1.75 billion |
| 5 GW | 43.80 TWh | $4.38 billion |
These calculations assume continuous electricity demand at the stated facility load. Actual consumption depends on installed equipment, utilization, cooling efficiency and available electricity.
A campus rated at 1 GW of IT capacity requires additional electricity for cooling and electrical infrastructure. At a power usage effectiveness (PUE) of 1.20, a fully utilized 1 GW IT installation would draw approximately 1.2 GW from the electricity supply.
Major AI Data Center Projects: Capacity and Potential Electricity Demand
Large AI infrastructure developments demonstrate why electricity availability is becoming a strategic priority.
Important: The electricity figures are illustrative calculations, not reported consumption. A 1 GW reference scenario does not mean that each named project has announced or achieved 1 GW of operational electrical demand. Project capacity can also refer to IT load, facility power or a multiyear development target.
Meta Hyperion: What Would a 5 GW AI Campus Consume?
Meta’s Hyperion AI data center development in Louisiana represents one of the largest announced AI infrastructure projects.
The campus has been associated with plans for up to 5 GW of capacity, although construction and electricity requirements will increase in phases.
If Hyperion eventually drew 5 GW of continuous facility power, its annual electricity consumption would reach approximately 43.8 TWh.
At $0.10/kWh, the hypothetical annual electricity cost would be $4.38 billion.
At 1 GW of continuous demand, the equivalent annual consumption would be 8.76 TWh, costing $876 million.
These scenarios illustrate the scale of electricity procurement required for multigigawatt AI campuses, not Meta’s current electricity consumption.
Microsoft Fairwater: Electricity for AI Supercomputing
Microsoft’s Fairwater AI data center in Wisconsin is designed for high-density AI computing, including large-scale model training.
Such facilities require substantial electricity for GPUs, networking equipment, cooling and power distribution.
For a hypothetical 1 GW IT installation operating at 71 percent average utilization and PUE of 1.14:
Average IT demand: 710 MW
Average facility electricity demand: 809.4 MW
Annual electricity consumption: 7.09 TWh
Annual electricity cost at $0.10/kWh: $709 million
This is a reference calculation, not a disclosed Fairwater electricity forecast.
Microsoft’s electricity requirements will depend on its operational computing capacity, utilization and the number of AI campuses commissioned.
Oracle and Stargate: Multigigawatt AI Infrastructure
Oracle is expanding AI cloud infrastructure, including projects supporting OpenAI workloads.
The Stargate initiative involves major investments in AI computing facilities, with Oracle among the infrastructure partners.
A hypothetical Stargate campus drawing 2 GW continuously would consume approximately 17.52 TWh annually, costing $1.75 billion at $0.10/kWh.
A 500 MW operational phase would consume 4.38 TWh annually under the same assumptions.
The distinction matters because multigigawatt projects generally require phased construction, grid connections and equipment deployment.
xAI Colossus: High-Density GPU Computing
xAI’s Colossus infrastructure in Memphis demonstrates the electricity requirements of concentrated AI training clusters.
Large installations of AI accelerators require substantial power for computing, networking and cooling.
At a hypothetical 500 MW continuous facility load, annual electricity consumption would reach 4.38 TWh.
Expanding to 1 GW would increase annual consumption to 8.76 TWh if the entire electrical capacity were used continuously.
Actual consumption depends on installed GPU capacity, operational utilization and electricity infrastructure.
Amazon Project Rainier: AI Training Infrastructure
Amazon’s Project Rainier in Indiana supports large-scale AI computing using AWS infrastructure and custom Trainium accelerators.
The project illustrates how hyperscale cloud providers are developing dedicated facilities for demanding AI workloads.
A hypothetical 1 GW facility operating at an average 80 percent electrical load would consume approximately 7.01 TWh annually.
At $0.10/kWh, annual energy expenditure would be approximately $701 million.
Actual demand will depend on the project’s operational capacity, equipment utilization and cooling requirements.
Meta Prometheus: A Confirmed 1 GW AI Cluster
Meta’s Prometheus project provides another important example of gigawatt-scale AI infrastructure.
Meta has described Prometheus as a 1 GW AI computing cluster distributed across multiple data center buildings and supporting infrastructure.
According to Meta’s AI infrastructure engineering report, Prometheus represents a major step toward larger AI computing installations.
At a hypothetical continuous 1 GW facility demand, Prometheus would consume 8.76 TWh annually.
If 1 GW represents IT demand, facility electricity requirements would be higher after accounting for cooling and electrical losses. At PUE 1.20, continuous facility demand would reach 1.2 GW, equivalent to 10.51 TWh annually.
These are illustrative calculations rather than reported electricity consumption.
Electricity Costs: Comparing 1 GW AI Data Centers Across Tariffs
Electricity pricing can materially change the economics of large AI infrastructure projects.
| Electricity price | Annual cost at continuous 1 GW facility demand |
| $0.04/kWh | $350 million |
| $0.05/kWh | $438 million |
| $0.06/kWh | $526 million |
| $0.08/kWh | $701 million |
| $0.10/kWh | $876 million |
| $0.12/kWh | $1.05 billion |
| $0.15/kWh | $1.31 billion |
A reduction of just $0.02/kWh saves approximately $175 million annually for a continuously operating 1 GW facility.
This explains why electricity procurement, location selection and long-term power agreements are increasingly important to AI companies.
Cooling, Grid Connections and Renewable Power
Electricity costs extend beyond operating GPUs.
Cooling, transformers, power conversion, backup generation and electrical distribution increase facility requirements.
For a 1 GW IT installation operating continuously, improving PUE from 1.30 to 1.15 would reduce facility demand by 150 MW.
That represents approximately 1.31 TWh of annual electricity savings, worth $131 million at $0.10/kWh.
AI operators are also considering renewable energy, nuclear generation, natural gas and battery storage to secure reliable electricity.
The International Energy Agency’s Energy and AI outlook projects global data center electricity consumption will increase from approximately 485 TWh in 2025 to 950 TWh by 2030.
The expansion will create additional demand for power generation, transmission networks, substations and cooling infrastructure.
Which Companies Benefit From 1 GW AI Data Centers?
Electricity demand from large AI campuses creates opportunities for several industries.
Power generation: Constellation Energy, NextEra Energy and Vistra could benefit from growing demand for long-term electricity supply.
Grid infrastructure: GE Vernova, Siemens Energy, ABB and Hitachi Energy supply equipment supporting transmission and power distribution.
Data center power management: Schneider Electric, Eaton and Vertiv provide electrical infrastructure, power management and cooling technologies.
Backup generation: Caterpillar and Cummins supply generation equipment supporting data center reliability.
However, actual financial benefits depend on confirmed contracts, delivery schedules, margins and project commissioning.
Outlook: AI Data Center Electricity Demand Through 2030
Meta Hyperion, Meta Prometheus, Microsoft Fairwater, Oracle-backed Stargate projects, xAI Colossus and Amazon Project Rainier demonstrate the increasing scale of AI computing infrastructure.
A continuously operating 1 GW facility consumes 8.76 TWh annually, while a 5 GW facility could consume 43.8 TWh under the same assumptions.
For AI companies, securing reliable electricity at competitive prices is becoming a major competitive advantage.
The most successful AI data center operators will be those that convert announced gigawatt-scale projects into operational computing capacity while controlling electricity costs, cooling requirements and infrastructure delays.
SHAFANA FAZAL
