Data center power demand is moving from a technology-sector issue into a grid-planning problem. A new BloombergNEF report forecasts that planned construction will require 2.7x the sector's current electricity demand over the next decade, putting the industry on a path toward 106 gigawatts by 2035.
The shift is not just about more buildings. The report points to larger facilities, higher utilization, and the rising role of AI training and inference as the forces changing how much electricity data centers will need and where that demand is likely to appear.
Demand Is Set To Climb Sharply
Data centers use 40 gigawatts of electricity today, according to the report. By 2035, that figure is expected to reach 106 gigawatts. That is the basis for the forecast that new additions will require 2.7x the sector's current demand, described as nearly triple.
The pace matters because data center projects do not appear on the grid overnight. BloombergNEF said projects have an average seven-year timeline before coming online, which means early-stage announcements can shape the later years of a forecast.
The new estimate is also higher than the group's April document. The reason is a surge of newly announced projects since then. Early-stage projects more than doubled between early 2024 and early 2025, although the source distinguishes those from projects that have already been committed or are under construction.
For companies, utilities, and grid operators, that distinction is important. A project at an early stage may not be guaranteed, but enough of them can still change the outlook for future electricity demand, especially when the projects are large and take years to connect.
New Facilities Are Getting Much Bigger
The BloombergNEF report highlights a major change in facility scale. Today, only 10% of data centers draw more than 50 megawatts of electricity. Over the next decade, the average new facility is expected to draw well over 100 megawatts.
The largest planned sites make the forecast even more dramatic. Nearly a quarter are expected to be larger than 500 megawatts, and a few will exceed 1 gigawatt. Those very large projects skew the data, but they also show how far data center planning has moved beyond smaller, urban-adjacent sites.
Much of the growth is expected to move into more rural areas. BloombergNEF said that is because facilities are growing in size while available sites near urban areas are becoming scarce. The result is a data center map that may increasingly follow power availability, land availability, and grid connection opportunities.
This is a practical shift with major implications. A 100 megawatt facility, a 500 megawatt facility, and a 1 gigawatt site create very different planning questions. They affect how transmission organizations think about capacity, how utilities assess new load, and how quickly infrastructure can match demand.
AI Is Changing Utilization
The report also forecasts higher utilization across data centers. Utilization is expected to rise from 59% to 69% as AI training and inference grows to nearly 40% of total data center compute.
That matters because electricity demand is not determined only by how many facilities exist. It is also shaped by how heavily those facilities are used. If more data centers are running more often, total load can rise even before every planned project reaches completion.
AI companies have been racing to build more powerful data centers, according to the source article. That race has helped push global investment in the facilities to $580 billion this year, which the article notes is more than the world spends finding new supplies of oil.
The key point is that AI is affecting both the size and use of data center infrastructure. Training and inference are not described as side workloads in the forecast. They are expected to become a large share of total data center compute, which helps explain why the electricity outlook has moved upward.
Grid Pressure Is Concentrated In Key Regions
The source article says much of the new capacity is planned for Virginia, Pennsylvania, Ohio, Illinois, and New Jersey. These states sit within the PJM Interconnection region, which is operated by a regional transmission organization responsible for the electrical grid in those states and others, including Delaware, West Virginia, and parts of Kentucky and North Carolina.
Texas's Ercot grid is also expected to see a large number of additions. Together, those regional references show that the data center buildout is not evenly distributed. It is concentrating in places where grid decisions can become central to whether new projects proceed on the desired timelines.
The PJM Interconnection is already facing scrutiny from its independent monitor, Monitoring Analytics. The group filed a complaint with the Federal Energy Regulatory Commission saying PJM has authority to authorize new data center connections only when its grid has adequate capacity.
Monitoring Analytics also said data centers are responsible for today's high electricity prices within the region. The complaint argues that PJM's handling of reliability and affordable service is unjust and unreasonable.
Why This Forecast Matters
The BloombergNEF forecast shows how quickly data center energy demand can change when multiple forces move at once. More facilities are being planned. The facilities are larger. Utilization is expected to rise. AI training and inference are becoming a much bigger part of compute demand.
For the tech industry, the report frames power as a core constraint on expansion. For grid operators, it frames data centers as large new loads that may require careful sequencing before connection. For regions such as PJM and Ercot, the issue is not theoretical; the project pipeline is already focused there.
The most important takeaway is that data center growth is becoming a long-term electricity planning challenge. If demand reaches 106 gigawatts by 2035, the decisions made during the early stages of today's projects will shape the grid impacts years from now.