A recent study has projected that, by 2030, data centers run by six leading technology firms could consume between 239 and 295 terawatt-hours (TWh), roughly 1% of global electricity demand. Researchers believe this growth will be concentrated in North America, Western Europe, and Asia Pacific, with some regions facing greater strain than others. Their findings were published in Communications Sustainability.
Study: Artificial intelligence data centers could reach one percent of global electricity demand by 2030. Image Credit: dotshock/Shutterstock.com
The Rising Energy Cost of AI
Generative AI and large-scale data analytics have driven sharp growth in computational demand, with AI facilities consuming up to six times more power than conventional racks. The International Energy Agency (IEA) expects worldwide data-center power use to climb from roughly 415 TWh in 2024 to around 945 TWh by 2030, more than doubling.
Prior research offers two main approaches. Top-down methods use aggregate indicators but miss evolving AI workloads. Bottom-up engineering models estimate facility-level energy use but face limited data availability.
Both focus on aggregate demand or facility-level accounting, offering limited insight into how firm-level investment, spatial clustering, and regional grid characteristics jointly shape AI-related energy demand. This study addressed that gap with an integrated framework linking infrastructure deployment, siting patterns, and regional power-system impacts.
Forecasting AI Infrastructure and Energy Demand
This study builds a hybrid framework that combines retrieval-augmented generation with a large language model to forecast how hyperscale data centers operated by major technology companies will
evolve geographically and operationally.
The workflow has six stages. First, a knowledge base is assembled from corporate and regional sources spanning 2015 to 2025, including press releases, earnings reports, sustainability disclosures, investment announcements, policy documents, and news articles.
Second, the model identifies likely future expansion locations by analyzing the sentiment and tone of these documents, treating positive language about investment and growth as a signal of expansion intent.
It then retrieves location-specific context for each candidate site before extracting baseline operational parameters such as accelerator counts, power draw, utilization, and cooling efficiency.
It projects these parameters forward to 2030, constrained by observed hyperscale patterns including annual capacity growth and efficiency gains. Finally, it converts these projections into physical energy demand and constructs an index that compares projected data center demand with available regional electricity supply.
The study covers six leading technology firms that together account for roughly 70–75% of global hyperscale and cloud-linked data center electricity demand. It tests three scenarios reflecting conservative, central, and high-growth trajectories of AI workload expansion, with annual growth rates of 15, 25, and 35%.
Firm-level demand is then allocated across regions using modeled siting probabilities for new AI sites and historical facility counts for legacy sites. Projected totals are cross-validated against international forecasts, and the resulting range aligns with external benchmarks.
Regional Patterns in AI Energy Demand
The results show a clear shift in data center geography. Older facilities are widely dispersed, placed to keep response times low and stay close to customers. New AI-specific data centers, by contrast, cluster densely in a limited number of regions.
Across the leading firms, more than 90% of projected AI computing capacity is in North America, Europe, and Asia Pacific. The largest clusters form where energy is plentiful, grid and fiber networks are mature, the climate is favorable, and policy is supportive.
Within North America, clusters concentrate in Virginia, Texas, Ohio, and North Carolina. European centers cluster in the Netherlands, the United Kingdom, and Italy, with Nordic regions favored for hydropower and wind resources due to their cool climates. In Asia Pacific, clusters are seen in Singapore, Taipei, Malaysia, and Japan.
Download a copy of this news piece for later by clicking here!
The model showed that each firm’s strategies differ. Amazon shows the broadest geographic diversification, while Apple concentrates deployments in the United States. Microsoft and Oracle mainly locate where policy and incentives are strongest, whereas Google and Meta favor locations with abundant energy and reliable network connectivity.
Across all modeled scenarios, electricity demand rises sharply. Aggregate consumption by the six leading firms is projected to grow from about 118 TWh in 2024 to between 239 and 295 TWh by 2030, a compound annual growth rate of roughly 13–17%. Amazon, Microsoft, and Google show the highest absolute demand. Apple and Oracle show lower demand, consistent with smaller operational footprints.
Regionally, fewer than 10 regions account for nearly two-thirds of total projected demand; Oregon, Virginia, Iowa, Ohio, and Ireland emerge as dominant hubs. Regions with previously elevated demand concentration tend to experience the largest increases, while regions with larger electricity systems such as Texas absorb new demand more easily.
Power System Implications of AI Expansion
The research linked AI data center siting, electricity demand projections, and regional demand concentration. They found that AI infrastructure is becoming a structural part of power system dynamics rather than a peripheral load.
Clustering creates spatial asymmetries, with regions such as Oregon, Ireland, and Iowa facing high relative concentration of demand, while larger systems such as Texas absorb new demand more easily.
This presents both challenges, including transmission congestion and price volatility, and opportunities such as renewable procurement and flexible demand. Therefore, the authors call for anticipatory planning that incorporates AI demand into energy modeling, and note limitations including optimistic corporate disclosures, the exclusion of smaller operators, and a focus on aggregate supply over network constraints.
Toward Sustainable AI Energy Infrastructure
Research suggests that data centers run by six leading firms could consume roughly 1% of global electricity by 2030, with demand concentrated in fewer than 10 regions that account for nearly two-thirds of the total. This concentration could create uneven pressure, making data center demand a larger share of supply in smaller electricity systems.
Because a few firms drive most of this load, regulators can engage specific operators rather than a generic technology sector. The authors argue that treating digital infrastructure as ordinary commercial demand risks underestimating where pressure will build.
Anticipatory planning that folds AI demand into energy modeling, alongside targeted renewable procurement requirements, locational guidance, or mandatory load disclosure, will be central to keeping computational growth aligned with reliable, low-carbon electricity systems.
Journal Reference
Chen, D., et al. (2026). Artificial intelligence data centers could reach one percent of global electricity demand by 2030. Communications Sustainability. 1. DOI:10.1038/s44458-026-00152-5. https://www.nature.com/articles/s44458-026-00152-5.
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.