US Hyperscalers To Consume 22% More Grid Power By End of 2025
Summary
Research from 451 Research (part of S&P Global) warns that US hyperscale datacentres will draw 22% more grid power in 2025 than a year earlier, driven largely by machine-learning workloads and GPU-heavy infrastructure. Utility power to hyperscale sites is estimated to rise by 11.3 GW to 61.8 GW this year, with forecasts reaching 75.8 GW in 2026, 108 GW in 2028 and 134.4 GW by 2030. The figures exclude enterprise-owned facilities and focus on hyperscalers, leased sites and crypto-mining operations.
The report highlights Virginia and Texas as the states with the highest current demand (Virginia ~12.1 GW in 2025, Texas ~9.7 GW), while operators are also exploring locations such as Idaho, Louisiana, Oklahoma and parts of West Texas for stranded power and alternative generation options. The growth is largely due to the need for purpose-built facilities with extensive power and cooling, rather than retrofitting older sites.
Key Points
- Hyperscale datacentres in the US are forecast to consume 22% more grid power in 2025 versus 2024.
- Utility-supplied capacity to these datacentres is estimated to jump by 11.3 GW to 61.8 GW this year.
- Forecasts project demand of 75.8 GW in 2026, 108 GW in 2028 and 134.4 GW by 2030 (hyperscale, leased and crypto sites only).
- AI/ML training and GPU-dense servers are the primary drivers of rapidly rising energy needs.
- Virginia and Texas currently lead in demand; operators are scouting for cheaper or stranded power in states like Idaho, Louisiana and Oklahoma.
- Figures exclude enterprise-owned datacentres, so total sector energy use is even larger than reported.
Context and Relevance
This is a significant infrastructure and policy story. Rapid growth in hyperscale energy demand affects grid planning, local electricity prices, permitting and where new generation investment is required. For regions already strained by demand, large datacentre draws can accelerate the need for transmission upgrades, new generation (including renewables and storage) and regulatory responses.
For industry and investors, the projections indicate where capacity, power contracts and on-site generation (or stranded-power deals) will be most valuable. For policymakers and communities, it raises questions about who pays for grid upgrades and how to balance industrial demand with residential needs and climate goals.
Why should I read this?
Short version: if you care about where AI gets built, rising power bills, or whether your region needs new power plants and transmission lines, this matters. It explains which states are swallowing most of the demand and why hyperscale growth will shape grid upgrades and local electricity economics for years.
Author style
Punchy — this isn’t just another stat. The projected jump is large enough that the details on timing, geography and drivers matter to anyone tracking energy, infrastructure or the economics of AI. Read the numbers if you want to understand the likely winners and losers.