Data center energy demand has moved from a facilities issue to a power-system planning question. The numbers now being published by U.S. agencies and energy analysts point in the same direction: electricity use from servers, cooling, power conversion, and related infrastructure is rising quickly, but the size of the increase depends on assumptions about AI deployment, utilization, cooling efficiency, grid interconnection, and project execution.
The most useful way to read these forecasts is not as a single prediction. They are scenario ranges. A forecast expressed in terawatt-hours measures annual consumption, while a forecast expressed in gigawatts points to power capacity. Both matter, but they answer different questions for utilities, data center operators, regulators, and communities that host large facilities.
Data Center Energy Demand Baseline
Data Center Energy Demand in 2023
The Department of Energy reported that U.S. data centers used about 4.4% of total U.S. electricity in 2023. The same DOE release said consumption could rise to about 6.7% to 12% of total U.S. electricity by 2028, with estimated consumption of 325 to 580 TWh under different scenarios according to DOE.
The DOE estimate makes data center energy demand visible at national scale. A 4.4% share is already material for grid planning, but the 2028 range is more significant because it is wide. A low case near 325 TWh and a high case near 580 TWh imply very different needs for generation, transmission, substations, backup power, and cooling infrastructure.
Why Shares And TWh Both Matter
Percentages are useful for public discussion because they show how large data centers are relative to the whole U.S. electricity system. Terawatt-hours are more useful for energy procurement and emissions accounting. A facility operator may care about megawatts at the meter, but a utility planner also needs annual energy, peak load timing, and load factor.
This distinction matters for basic comparisons. A data center with high contracted capacity does not necessarily consume at full load every hour. At the same time, AI-oriented facilities can create dense, sustained loads that differ from many commercial buildings. The forecasts in the research notes do not resolve all utilization assumptions, so treating the high and low cases as planning boundaries is safer than treating either endpoint as certain.
Forecast Ranges to 2030
The LBNL Reference Case
Lawrence Berkeley National Laboratory’s 2025 update projects that U.S. data centers could account for 11.8% of all U.S. electricity use by 2030 in its reference case. The same update gives a scenario range from 9.5% to 15.3%, with reference-case usage of about 649 TWh in 2030 in the LBNL report.
For data center energy demand planning, that 2030 range is the key signal. The difference between 9.5% and 15.3% of U.S. electricity use is not a rounding issue. It affects where new load can connect, how much generation must be available, how cooling loads influence commercial-sector intensity, and whether operators can meet energy commitments without depending on projects that are not yet built.
How Global Forecasts Frame The U.S. Case
The International Energy Agency’s 2025 Energy and AI report estimated global data center electricity consumption at about 415 TWh in 2024, or roughly 1.5% of global electricity consumption. Its Base Case expected global data center electricity consumption to reach about 1,200 TWh by 2035. The same IEA work placed the United States at about 45% of current global data center electricity demand in 2024 and expected the country to account for nearly half of the increase through 2030.
Those global figures do not replace the DOE or LBNL U.S. estimates, but they explain why U.S. grid effects are receiving so much attention. If the United States has the largest current share and is expected to absorb a large part of near-term growth, then U.S. transmission queues, regional power prices, siting rules, and cooling constraints become part of the AI infrastructure discussion rather than background details.
Capacity Forecasts Through 2035
Why Gigawatts And Terawatt-Hours Differ
BloombergNEF forecast that U.S. data center power capacity would reach 194 GW by 2035. The research notes identify that as an 83% increase over BNEF’s December 2025 forecast, driven by AI-focused data center projects. Bloomberg Law reporting in July 2026 also described a forecast in which data centers would consume about 20% of U.S. electricity in 2035, compared with about 12% in 2030 and about 5.9% in the recent past.
Capacity and consumption should not be mixed without care. A 194 GW capacity figure is not the same thing as annual electricity consumed. Capacity describes how much load may need to be served at a point in time. Annual consumption depends on how often that capacity is used, how efficiently IT equipment runs, how cooling systems perform, and whether facilities operate near full utilization.
Interconnection Pace Is A Constraint
The July 2026 reporting cited in the research notes said that even if grid connections continued at a recent maximum pace of about 7 GW per year, the base-case scenario would still leave an estimated 19 GW supply shortfall by 2035. That figure should be read as a planning warning, not as proof that every announced site will be built exactly as proposed.
The practical point is narrower: interconnection and infrastructure delivery can lag behind project announcements. Data halls can be designed faster than transmission lines, substations, and generation additions can be approved, financed, built, and commissioned. A related analysis of AI data center energy patterns discusses why flexible demand, cooling control, and cautious sustainability claims are becoming more relevant as these loads grow.
Planning And Security Implications
What Operators Can Control
Data center energy demand growth puts pressure on decisions that operators can influence directly: server utilization, thermal management, power usage effectiveness, backup architecture, and procurement strategy. The forecasts do not specify which design choices will dominate by 2035, so claims about future efficiency should be treated carefully unless they include the workload, cooling design, climate zone, and utilization assumptions behind the estimate.
EIA’s Annual Energy Outlook 2026, as summarized in the research notes, projected that electricity consumed by data center servers across the commercial building stock would rise significantly by 2050, reaching roughly 446 billion kWh to 818 billion kWh annually across scenarios. The same notes said commercial-sector electricity intensity, driven in large part by data centers and cooling loads, was expected to exceed its prior 2003 high point around 2031 to 2032.
What The Forecasts Do Not Prove
These forecasts do not prove that every planned data center will secure power, receive permits, or operate at expected utilization. They also do not quantify cybersecurity risk in facility controls, grid interfaces, or power-management telemetry. Those issues need separate review because energy forecasting and operational security answer different questions.
- For utilities, the main issue is whether load growth appears faster than generation, transmission, and distribution upgrades can be delivered.
- For operators, the main issue is whether contracted power, backup systems, cooling, and equipment utilization match actual workload demand.
- For policymakers, the main issue is whether cost allocation, reliability, emissions, and local infrastructure impacts are being measured consistently.
For those exploring educational materials on infrastructure concepts, stampsinclass.com is a part of the same educational publishing network. This analysis, however, remains tightly focused on U.S. electricity forecasts and data center infrastructure.
Data Center Energy Demand Through 2035
A Cautious Reading Of The Numbers
The evidence supports a clear direction: U.S. data center electricity use is rising and could take a much larger share of national power consumption by 2030 and 2035. The stronger claim, that a precise share is already locked in, is not supported by the forecast ranges. DOE’s 2028 range, LBNL’s 2030 scenarios, IEA’s global Base Case, and BNEF’s 2035 capacity forecast all depend on assumptions that can change with AI adoption, hardware efficiency, siting constraints, grid expansion, and facility utilization.
A defensible reading is that data center planning has shifted from incremental commercial-load growth to major infrastructure coordination. By 2035, the sector may require power planning on a scale closer to heavy industry than conventional office or retail load. That does not make the highest forecast inevitable, but it does make conservative grid planning harder to justify if it ignores the upper end of published scenarios.
The most useful next measurements will be annual electricity consumed, peak coincident load, connected capacity, regional interconnection timelines, cooling energy, and server utilization. Those metrics will show whether the United States is tracking closer to the lower, reference, or high-growth cases. Until then, the forecasts should be used as boundaries for risk management rather than as fixed targets.
