The explosive growth of artificial intelligence has moved beyond the limits of semiconductor availability to reveal a more fundamental constraint: electricity supply. While chipmakers have addressed past shortages, meeting the unprecedented power demands of AI data centers poses a significant challenge for global power grids designed for slow and steady consumption increases.
The rise of AI applications like ChatGPT, which garnered millions of users within months, triggered an urgent demand for high-performance GPUs. NVIDIA, the dominant supplier of AI chips, benefited immensely due to supply bottlenecks and specialized manufacturing. However, that chip supply issue is easing as foundries scale production. What remains less recognized is the vast electrical power needed to run these chips continuously at data centers.
Running a single AI query consumes approximately ten times the energy of a standard internet search, while training the latest large models demands power equivalent to small cities. Industry forecasts estimate that global spending on AI-related data center infrastructure will reach over five trillion dollars by 2030. According to Goldman Sachs Research, power consumption for data centers could increase by 165% within the same timeframe compared to 2023 levels.
Electric grids were originally engineered to support annual electricity growth rates of one to two percent and were never meant to handle sudden, massive surges in demand from hyperscale AI operations. Now, large technology companies are requesting enormous power allocations—hundreds of megawatts—from utility providers, forcing an urgent reconsideration of energy infrastructure planning.