The National Picture
Data centers consumed 4.4% of all U.S. electricity in 2023. That number is real and sourced. It is also the beginning of the story, not the end of it. The growth trajectory is the part that should concern anyone paying attention.
| Year | Data Center TWh | % of U.S. Total | Context |
|---|---|---|---|
2014 | 58 TWh | ~1.5% | Pre-cloud boom baseline |
2023 | 176 TWh | ~4.4% | Current confirmed figure (LBNL/DOE) |
2028 (low) | 325 TWh | ~6.7% | Conservative projection |
2028 (high) | 580 TWh | ~12% | Aggressive AI deployment scenario |
Source: Lawrence Berkeley National Laboratory, 2024 U.S. Data Center Energy Usage Report (congressionally mandated). Load tripled in a decade. It may triple again in five years. The range in the 2028 projections reflects GPU availability and AI deployment speed, not consumer choice.
What the 4% figure leaves out
The LBNL 176 TWh figure explicitly excludes cryptocurrency mining. It also may not fully capture self-generation (like xAI’s gas turbines, which do not appear in utility consumption data). Adding up the known hyperscale facilities one by one and comparing that total against 176 TWh is work still to be done. If the two numbers do not match, the gap between them is itself the story.
Who is actually driving this? Data center capital expenditure reached $726 billion in 2025, a 57% increase and the fastest growth rate ever recorded. The four largest U.S. cloud providers raised their capital spending by 76%. AI accounts for 10 to 20% of current data center energy consumption but is responsible for nearly all of the new capacity growth. By 2030, roughly 70% of total data center demand will be for AI-ready infrastructure. This is a corporate investment race, not a consumer demand story.