As artificial intelligence drives change in the energy sector, its rapid advancement leaves a lot of room for questions — and uncertainty — around its energy use needs and impacts.
| Myth: The growth in AI’s electric use is all hype and inflated projections. | Fact: Data centers consumed 224 terawatt-hours of electricity in 2025 — nearly triple the 76 TWh used in 2018. This growth is increasingly attributed to data centers serving AI. Data center usage is expected to more than double to 426 TWh by 2030.  | ||||||||||||||||
|
| Myth: Data centers comprise the greatest share of electricity use in the U.S. | Fact: Data centers currently use less electricity than residential air conditioning and commercial lighting. Though anticipated growth could increase the proportion of electricity consumed by data centers to 10% or more of the U.S. total by 2030. |
 |  |  |
- Data centers consume a greater share in some states than others:
| State | Share of Electric Use |
|---|---|
| Virginia | 25% |
| North Dakota | 15% |
| Nebraska | 12% |
| Iowa | 11% |
| Oregon | 11% |
| Nevada | 8.6% |
| Myth: AI is behind all load growth. | Fact: While data centers comprise the majority of projected load growth, not all data centers are for AI, and other factors include manufacturing & industrial uses, population growth, and electrification of transportation and buildings. |
| Myth: Data center development is distributed evenly across the country. | Fact: A third of data centers are in Virginia, Texas, and California, and half of data centers currently under construction are part of existing large clusters. Behind Virginia and Texas, Georgia and Illinois have the most planned data centers. |
|
| Myth: Most of AI’s energy use is from people making silly images and deepfake videos. | Fact: Less than half of data centers’ power consumption is for computing power, with cooling, lighting, and other uses comprising anywhere from 40%-70%. An estimated 80-90% of computing power for AI is used for inference — or to train the AI model, not for people to use it. |
While exact energy use for individual tasks depends on the complexity of the model and requested output, individual tasks on average use a small amount of energy:
|
| Myth: All AI models and tasks use roughly the same amount of energy. | Fact: User-submitted reviews of various AI models, such as the AI Energy Score, show vast differences in efficiency – with similar tasks taking hundreds or thousands time the energy from one model to the next. |
