AI Compute Surge: Data Center Electricity Demand to Quadruple by 2035

The rapid acceleration of artificial intelligence is driving an unprecedented appetite for power, threatening to reshape global energy landscapes. New projections suggest that data centers will consume four times more electricity than they do today, fundamentally challenging the stability of existing power grids.

The Massive Scale of AI Compute Demand

According to a recent report from BloombergNEF, the surge in AI compute is expected to push data center capacity to nearly 200 gigawatts over the next decade. This growth is not merely incremental; it represents a structural shift in how energy is distributed. By 2035, data centers are projected to consume one-fifth of all electricity generated in the United States.

A significant portion of this capacity—nearly 50%—will be dedicated specifically to AI training and inference tasks. The United States is positioned to be the epicenter of this energy consumption, with forecasts suggesting the country will host 64% of all AI chips by power demand by 2033. The intensity of this demand has led major analysts to revise their numbers upward: BloombergNEF’s 2035 estimate is now 83% higher than its December forecast, while other organizations like EPRI and S&P have also significantly raised their projections.

Strained Grids and Infrastructure Bottlenecks

As data center development reaches a fever pitch, the physical infrastructure required to support them is reaching its breaking point. Much of this new capacity will plug into electrical grids that are already under immense pressure.

The impact on regional grid operators is particularly acute. In the PJM Interconnection region, which spans from Virginia to Illinois, data centers are expected to consume 34% of all electricity. Similarly, ERCOT in Texas will need to devote 22% of its generating capacity to these facilities.

The consequences of this imbalance are already being felt. PJM has faced severe challenges managing connection requests, leading to a four-year pause on new applications. This supply-demand tension has contributed to a 76% spike in electricity prices over the past year. The situation is so volatile that utilities like American Electric Power have even threatened to withdraw from the interconnection.

A Global Energy Challenge

While the U.S. remains the leader in AI compute power, the energy implications of this technological shift are global. If current AI adoption trajectories continue, data centers will generate 1,935 terawatt-hours of new electricity demand worldwide. To put that scale into perspective, that is nearly equivalent to the total annual electricity consumption of India.

For the AI industry, this creates a critical dependency: the future of LLM development and machine learning innovation is no longer just a software challenge, but a hard physical constraint governed by power availability and grid stability.

Key Takeaways

  • Exponential Growth: Data center electricity usage in the U.S. is expected to quadruple by 2035, driven largely by AI training and inference workloads.
  • Grid Instability: Major operators like PJM and ERCOT face extreme strain, with electricity prices rising 76% in a single year due to supply-demand imbalances.
  • Global Scale: The projected worldwide increase in data center energy demand (1,935 TWh) is comparable to the entire annual energy consumption of India.