AI Boom Could Push Data Centers to Consume Four Times More Electricity by 2035

Artificial intelligence is transforming industries around the world, but powering that revolution is becoming an enormous challenge. According to a new report from BloombergNEF, data centers are expected to consume nearly 20% of all electricity generated in the United States by 2035, roughly four times more than they use today.

The dramatic increase is largely being fueled by the explosive growth of AI computing. As companies race to develop more powerful large language models and AI services, they are building larger, more energy-hungry data centers capable of handling massive amounts of computation.

The report highlights just how quickly demand is accelerating and suggests that previous forecasts may have significantly underestimated the scale of the coming energy challenge.

AI Is Driving Unprecedented Growth

BloombergNEF estimates that U.S. data center capacity will approach 200 gigawatts over the next decade, with nearly half of that capacity dedicated to AI training and inference.

Training advanced AI models requires enormous computing power, often involving thousands of graphics processing units (GPUs) operating continuously for weeks or even months. Once those models are deployed, inference the process of generating responses for users continues to consume substantial computing resources around the clock.

As AI adoption expands across businesses, governments, and consumers, the demand for data center infrastructure is expected to rise even further.

The report predicts that by 2033, the United States will host 64% of the world's AI chip power demand, reinforcing its position as the global leader in AI infrastructure.

Forecasts Keep Getting Bigger

Perhaps the most striking finding is how quickly electricity demand projections are increasing.

BloombergNEF's latest estimate for electricity consumption in 2035 is 83% higher than the consultancy's own forecast released just seven months earlier.

Other organizations are also revising their expectations upward.

The Electric Power Research Institute (EPRI) has more than doubled its projections since 2024, while S&P Global has increased its forecast by more than one-third over just a few months.

These revisions reflect the rapid pace at which technology companies are announcing new AI infrastructure projects across the United States.

What once appeared to be an aggressive estimate is now increasingly viewed as conservative.

Power Grids Face Growing Pressure

The surge in electricity demand presents major challenges for existing power grids.

Many of the regions attracting large-scale data center investments are already experiencing capacity constraints, making it increasingly difficult to connect both new power plants and new electricity users.

BloombergNEF expects some of the biggest impacts to be felt in the PJM Interconnection, which operates the electricity grid across parts of the eastern United States, stretching from Virginia to Illinois.

By 2035, the report estimates that 34% of electricity generated within PJM could be consumed by data centers alone.

Meanwhile, ERCOT, which manages most of Texas' electricity grid, is projected to allocate 22% of its generating capacity to data centers.

These figures highlight the growing competition between households, businesses, manufacturers, and AI infrastructure for limited electrical resources.

Connection Delays Become a Bottleneck

The pressure is already being felt.

PJM previously paused applications for new power generation projects for four years while attempting to clear an overwhelming backlog of connection requests.

Although new applications resumed earlier this year, utilities continue struggling to keep pace with rising demand.

American Electric Power has even threatened to withdraw from the PJM system due to concerns over the growing imbalance between electricity supply and demand.

That imbalance is already affecting consumers.

Electricity prices within the region have risen 76% over the past year, driven largely by increased demand and limited generation capacity.

Despite those challenges, demand from data center operators remains exceptionally strong.

During PJM's most recent capacity auction, 38% of all new electricity demand requests came from data centers, demonstrating just how central AI infrastructure has become to future energy planning.

A Global Trend

Although the United States is expected to remain the largest hub for AI computing, similar trends are emerging worldwide.

BloombergNEF estimates that by 2033, global data centers could create an additional 1,935 terawatt-hours of annual electricity demand if AI adoption continues along its current trajectory.

To put that figure into perspective, it is nearly equivalent to the amount of electricity consumed by India in an entire year.

This illustrates that AI's energy requirements are no longer simply a regional issue; they are becoming a global infrastructure challenge.

Countries hoping to compete in artificial intelligence will also need to invest heavily in electricity generation, transmission networks, and grid modernization.

Balancing Innovation and Sustainability

The rapid expansion of AI brings enormous economic opportunities, but it also raises important questions about sustainability.

Technology companies are increasingly investing in renewable energy, battery storage, and even nuclear power partnerships to help meet future electricity demand while reducing carbon emissions.

At the same time, engineers are working to improve the energy efficiency of AI chips, cooling systems, and data center operations.

Even so, improvements in efficiency may not fully offset the explosive growth in computing demand.

As AI becomes embedded in more industries and everyday applications, governments, utilities, and technology companies will likely need to work together to ensure that electricity infrastructure can keep pace.

The BloombergNEF report serves as a reminder that the future of artificial intelligence will depend not only on smarter algorithms but also on having enough reliable power to keep them running.