Data centers expected to use 4x more electricity by 2035

The Escalating Energy Appetite of AI

The burgeoning field of artificial intelligence is set to become a colossal consumer of electricity, with data centers in the U.S. alone projected to account for one-fifth of the nation's power generation by 2035. This represents a staggering fourfold increase from current levels, according to a recent BloombergNEF report. The primary driver behind this exponential growth is the insatiable demand for AI compute, encompassing both the intensive training of large language models and the subsequent inference processes. Nearly half of the anticipated 200 gigawatts of data center capacity over the next decade will be dedicated to these AI-specific tasks, with the U.S. expected to host a dominant 64% of AI chips by power demand by 2033.

What makes these projections particularly concerning is their tendency to be conservative. BloombergNEF's latest estimate for 2035 electricity demand is 83% higher than its own forecast from just last December, reflecting the accelerating pace of AI development and deployment. Other industry organizations, such as EPRI and S&P, have similarly revised their forecasts upwards, underscoring a collective underestimation of AI's energy footprint. This rapid recalibration suggests that the industry is struggling to keep pace with the sheer scale of energy infrastructure required to support the AI revolution.

Strained Grids and Rising Costs

The rapid expansion of data centers, particularly those catering to AI, is placing immense pressure on existing electrical grids, many of which are already operating near capacity. The BloombergNEF report highlights that the majority of new data centers will connect to grids that are ill-equipped to handle such a surge in demand. For instance, the PJM Interconnection, a critical grid spanning from Virginia to Illinois, is expected to see 34% of its electricity consumed by data centers. Similarly, ERCOT, which manages most of Texas's power, will need to dedicate 22% of its generating capacity to these facilities.

These figures are not merely theoretical; the strain is already manifesting in tangible problems. PJM, a hub for many of the country's data centers, has previously paused applications for new grid connections for four years due to congestion. While the queue reopened in April, the situation remains precarious, with one major utility, American Electric Power, reportedly threatening to withdraw from the interconnection altogether. This supply-demand imbalance has had direct economic consequences, pushing electricity prices up by 76% over the past year. Despite these challenges and rising costs, data centers continue to seek connections to these strained grids, representing a significant portion of recent capacity auctions.

Global Implications and Future Challenges

While the U.S. is poised to host the majority of AI compute, the energy demands of data centers are a global phenomenon. The report projects that by 2033, if AI adoption continues its aggressive trajectory, data centers worldwide will generate an additional 1,935 terawatt-hours of electricity demand. To put this into perspective, this new demand alone is nearly equivalent to India's entire annual electricity consumption. This global surge underscores the need for coordinated international efforts to address energy infrastructure and sustainability concerns.

The implications extend beyond mere power consumption. The environmental footprint of AI, particularly its carbon emissions from energy generation, will grow substantially unless there is a significant shift towards renewable energy sources for data centers. Furthermore, the competition for grid capacity could lead to higher energy costs for other industries and consumers, potentially hindering economic growth in regions with insufficient infrastructure. The challenge lies in balancing the rapid innovation of AI with the fundamental need for a stable, affordable, and sustainable energy supply.