Case Study: AI vs Energy Consumption: Long-Term Grid Impact Study
Artificial intelligence has swiftly evolved from a specialized computational area into a dominant force driving electricity demand growth in the 21st century. The connection between AI adoption and global energy consumption is no longer a matter of speculation; it is now being systematically tracked, modeled, and projected by leading energy institutions. Notably, the International Energy Agency (IEA) dedicated a special report to this topic in April 2025. This report, alongside a growing body of research, clearly indicates that the world’s electricity grids are entering a new phase, characterized by an accelerating, geographically concentrated surge in demand driven by the rise of AI workloads in data centers, rather than gradual load growth.
In 2024, global electricity consumption by data centers reached approximately 415 terawatt-hours (TWh), accounting for around 1.5% of total global electricity use and reflecting a 12% annual growth rate over the previous five years. To put this in perspective, this level of electricity demand is roughly equivalent to that of Japan. While a 12% annual growth rate was already more than four times higher than the overall increase in global electricity consumption, the introduction of large-scale AI workloads particularly those involving GPU-intensive training and high-throughput inference has further accelerated this trajectory.
The relationship between AI adoption and grid stress is statistically significant. An analysis of AI compute investment, as indicated by GPU shipments and hyperscaler capital expenditures, against data center power demand from 2017 to 2024 yields a Pearson correlation coefficient of approximately r = 0.97, suggesting a nearly linear connection. Moreover, the correlation between data center electricity consumption growth and instances of grid congestion is also robust, with a coefficient of around r = 0.94. This indicates that data centers, unlike dispersed loads such as electric vehicles, tend to cluster geographically, resulting in concentrated demand spikes that existing transmission infrastructure struggles to accommodate. Additionally, there exists a negative correlation of r = −0.71 between improvements in efficiency, measured by Power Usage Effectiveness (PUE), and total energy consumption. This finding underscores a critical point: even substantial gains in energy efficiency per unit are consistently overshadowed by overall volume growth—a phenomenon known as Jevons Paradox. For instance, Google reported a 33-fold reduction in median energy consumption per Gemini prompt between May 2024 and May 2025, yet these efficiency enhancements may be eclipsed by a surge in overall demand if the cost of AI operation per query decreases and total queries escalate.
The IEA's base-case projections are definitive, forecasting that global data center electricity consumption will approximately double to around 945 TWh by 2030, constituting just under 3% of total global electricity demand. From 2024 to 2030, data center electricity consumption is expected to grow at about 15% per year—more than four times the growth rate of electricity consumption in all other sectors. Consumption within accelerated servers, primarily driven by AI adoption, is anticipated to increase by 30% annually, while traditional server electricity demand is projected to rise by 9% per year. Beyond 2030, the IEA envisions a continued rise to about 1,193 TWh by 2035, though with significant uncertainty as adoption trajectories may diverge.
One of the most important findings of the study is the regional concentration of this demand. The impact of AI on electricity grids is not distributed evenly across the global energy system. The United States and China together account for nearly 80% of the projected growth in global data center electricity consumption through 2030, with the U.S. expected to see an increase of around 240 TWh and China about 175 TWh compared to 2024 levels.
Data centers in the U.S. consumed 183 TWh in 2024, representing over 4% of the nation’s total electricity use, and by 2030, this number is projected to grow by 133% to 426 TWh. At a local level, the concentration is even more pronounced, with data centers representing more than 10% of electricity consumption in at least five U.S. states and over 20% in Ireland. Northern Virginia is home to approximately 4,000 MW of data center capacity, the largest such cluster globally, and the resulting pressures on local utilities and grid operators are already reflected in electricity pricing. For example, in the PJM electricity market, which spans from Illinois to North Carolina, data centers are estimated to have contributed to a $9.3 billion price increase in the 2025–26 capacity market, with projected monthly residential bill increases of $18 in western Maryland and $16 in Ohio.