The rapid expansion of AI is driving an unprecedented increase in electricity demand for data centers, far exceeding previous forecasts. In 2024, data centers accounted for 1.5% of global electricity consumption, a figure that grew by 15% annually over the prior five years. The International Energy Agency initially projected data center consumption to double from 415 terawatt-hours (TWh) in 2024 to 945 TWh by 2030. However, McKinsey's more recent estimate pushes this even higher to 1,400 TWh by 2030, representing 4% of the world's total electricity consumption.
This surge is fueled by the intense power requirements of AI models. A single GPT-4 model, for instance, can consume up to 463,269 megawatt-hours of electricity per year, surpassing the annual energy use of over 35,000 US homes. Training frontier AI models is particularly energy-intensive, requiring thousands of high-end GPUs running continuously for weeks or months, at costs ranging from tens of millions to over $1 billion per run. According to Nokia Bell Labs Consulting, AI-driven network traffic is expected to grow at a compound annual growth rate of 24%, reaching 1,088 exabytes per month by 2033, each step of data storage, movement, and processing adding to energy costs.
The sheer scale of this energy demand is becoming a national policy concern. OpenAI reportedly warned the White House Office of Science and Technology Policy that the US needs to add 100 gigawatts (GW) of new energy capacity annually to sustain AI growth, nearly double the 51 GW added in 2024. OpenAI and Nvidia are planning AI data centers that could consume up to 10 GW, with additional projects totaling 17 GW already underway. For perspective, 10 GW is equivalent to the summer peak power demand of New York City. Goldman Sachs Research forecasts global data center power demand to rise by 165% to 175% by 2030 from 2023 levels, an increase comparable to adding a new top 10 power-consuming nation.
The cost implications are significant, with a global survey of 300 senior executives by MIT Technology Review (MITTR) Insights revealing that 68% reported a rise in energy costs of 10% or more due to AI and data workloads. The battle for AI dominance is increasingly seen as a competition for power supply, rather than solely chips. China, for example, has a head start in renewable energy capacity, adding over 356 GW in new capacity last year, far exceeding the US total. This expanding grid capacity and declining electricity costs in countries with robust renewable energy infrastructure will increasingly determine the pace of AI progress.
