Key Takeaways
- US hyperscale data centers consumed an estimated 68 to 99 TWh over a 12-month period.
- Their electricity had a higher average carbon intensity than the broader US grid.
- Better reporting could help businesses compare cloud providers and assess emissions exposure.
A facility-level study covering 403 US hyperscale data centers estimated that they consumed 68 to 99 TWh of electricity between May 2024 and April 2025. The central estimate represented 1.8% of total US electricity consumption, according to Guidi et al.
The study estimated associated emissions of 37 million to 54 million metric tons of CO2, depending on assumptions about facility loads, regional generation, and accounting boundaries. Fossil fuels supplied about 54% of the electricity attributed to these facilities (source). Their average electricity carbon intensity reached 545 gCO2 per kWh, 48% above the US grid average of 370 gCO2 per kWh (source).
That comparison matters for enterprise technology buyers. Cloud workloads may feel geographically abstract, but their emissions depend heavily on where computing takes place and which generators respond to demand at a particular time. Two technically similar data centers can have sharply different footprints if one operates in a cleaner grid region or shifts flexible workloads toward lower-carbon hours.
Growth is adding urgency. The International Energy Agency reported that US data-center electricity demand increased 17% in 2025 and accounted for around half of total US electricity-demand growth that year. Globally, the IEA projects data-center electricity consumption to rise from 485 TWh in 2025 to approximately 950 TWh by 2030. AI-focused facilities are expected to expand faster than the broader category.
Efficiency improvements may not translate into lower total consumption when demand for AI training, inference, and cloud services is climbing rapidly. Lower power use per computation can reduce operating costs, yet those savings can support more deployment. For Amazon Web Services, Microsoft Azure, and Google data centers, infrastructure expansion has coincided with rising reported emissions despite renewable-energy commitments, according to the University of Chicago’s 2026 assessment.
The wider grid could also feel the pressure. Rhodium Group estimates that data centers could account for 14% of US power generation in 2030 under a high-growth scenario, compared with 10% in its baseline case. Associated power-sector emissions could be 6% to 13% higher in 2035 than under the baseline. Who supplies the next increment of electricity will therefore influence whether digital expansion supports or complicates corporate climate targets.
For business customers, familiar sustainability claims can obscure those differences. Renewable-energy purchases commonly influence market-based Scope 2 emissions, while location-based Scope 2 reporting reflects the average grid serving a facility. Both figures are useful, but neither alone shows whether clean generation matched consumption hour by hour. Power usage effectiveness, or PUE, also needs context because it measures facility overhead rather than the carbon intensity of the electricity consumed by servers.
More credible disclosure would place PUE beside grid carbon intensity, facility load, location-based and market-based Scope 2 emissions, and hourly clean-energy matching. The Greenhouse Gas Protocol Corporate Standard provides a basis for Scope 1, 2, and 3 accounting, while ISO 50001:2018 offers a structure for managing energy performance. Used together, those approaches can make cloud procurement less dependent on broad annual claims.
For CIOs, finance leaders, and sustainability teams, this is increasingly a commercial issue rather than a reporting footnote. Data-center location can affect emissions exposure, grid constraints, future electricity costs, and expansion schedules. Procurement reviews can ask providers for regional energy data, accounting boundaries, and hourly matching information. The resulting picture will still contain uncertainty, but it should be more decision-useful than a single renewable-energy percentage.
⬇️