Power Consumption Surge in Data Centers Due to AI Demand
Global power consumption in data centers is projected to rise by 26% this year, reaching 565 terawatt-hours, largely driven by the increasing demands of AI workloads, as outlined in a recent report from Gartner.
The study indicates that electricity usage in data centers will jump from 447 terawatt-hours in 2025 to 565 terawatt-hours in 2026. This significant rise is mainly attributed to compute-intensive AI applications. It highlights power availability as a crucial limitation affecting AI expansion across various sectors.
Gartner’s forecast also expects worldwide electricity demand to increase by 27% during this timeframe, reaching 132 gigawatts, up from 104 gigawatts in 2025. They estimate overall consumption could surpass 1,200 terawatt-hours by 2030. It’s important to note that gigawatts indicate peak capacity requiring infrastructure development, while terawatt-hours represent actual power usage throughout the year. Unfortunately, both measurements are being met with challenges in terms of supply infrastructure expansion.
“The rising need for compute-heavy AI workloads is leading to exceptional increases in data center power,” said Linglan Wang, a Direct Analyst at Gartner. “At the same time, AI capabilities face limitations due to power availability, making data center power security a critical frontier for businesses aiming to expand and protect their margins in the global AI competition.”
The study uncovers a notable shift in power consumption patterns between AI-optimized servers and traditional infrastructure. It’s expected that AI-optimized servers will consume around 95 terawatt-hours globally in 2025 and 175 terawatt-hours in 2026, marking an 84 percent increase. Gartner anticipates this number will climb to 258 terawatt-hours by 2027, which will be the first instance where AI-optimized hardware surpasses traditional servers in power consumption.
AI-optimized servers are projected to make up 31% of total data center power consumption by 2026, compared to about 20% a year prior. Additionally, the demand for power for cooling systems is on the rise, with expectations of a 22.6% increase to 195 terawatt-hours in 2026. This surge reflects the heat requirements of denser AI rack configurations and ongoing capacity expansions.
Out of the anticipated total global consumption of 565 terawatt-hours for 2026, the United States is expected to contribute around 204 terawatt-hours, or 36% of the global figure. In the U.S., AI-only data centers are said to consume about 68 terawatt-hours, which represents roughly one-third of the national total, while the growth in demand for non-AI data centers has been relatively modest during the same timeframe.
This uptick in energy demands is placing notable stress on regional power grids. In fact, over 75 data center projects worth $130 billion faced delays in early 2026 due to opposition regarding their impact on power and water resources. Some operators are resorting to on-site gas generators to activate capacity while awaiting grid connections. There have even been instances, like in one Virginia county, where employees were advised to conserve electricity amid rising utility costs driven by data center demand.
The implications of AI data centers continue to be a central concern in political discussions across the nation. Notably, Wynton Hall from Breitbart News has authored a bestseller titled Code Red: Left, Right, China, and the Race to Control AI, which aims to guide the MAGA movement in forming a stance on AI that advances human interests without ceding control to the left in Silicon Valley or to China.
