Artificial intelligence could overtake Bitcoin mining in electricity use within the next two years, with researchers warning that AI models may soon account for nearly half of all the power consumed by data centers worldwide. Analyst Alex de Vries-Gao from Vrije Universiteit Amsterdam’s environmental studies institute estimates this rising demand could put additional pressure on global energy supplies and potentially set back emissions reduction efforts.
De Vries-Gao, who has studied the environmental footprint of cryptocurrencies, says the energy appetite for AI models is accelerating much faster than many anticipated. Precise numbers are elusive since major tech companies rarely release details about how much of their energy is devoted specifically to AI, forcing researchers to build estimates from hardware supply data and public company disclosures.
At present, de Vries-Gao calculates that artificial intelligence already swallows up almost a fifth of all data center electricity worldwide. He warns that continued growth could see AI’s power consumption approach that of mid-size nations by 2025, echoing trends first seen with energy hungry crypto mining operations.
Data Centers and Power Grids Under New Strain
The explosion in demand for specialized AI chips has pushed production capacity at manufacturers such as Taiwan Semiconductor Manufacturing Company to double in a single year. Most of these data centers and power plants are in the United States, leading to expanded plans for new fossil fuel and nuclear facilities designed to keep pace.
The rapid growth poses complex challenges for local power grids, with cities and regions already straining to accommodate new data center loads. De Vries-Gao sees an ongoing pattern where the race for larger, more powerful AI models mirrors the “bigger is better” mentality that defined Bitcoin mining, though the true energy cost remains opaque due to a lack of transparency from tech companies.
As companies including Google and Microsoft ramp up their focus on machine learning, emissions tied to their operations have also increased. Sustainability reports typically highlight total greenhouse gas output but rarely break down contributions by specific services such as AI, making the full environmental impact almost impossible to pin down for outside experts.
Energy demand and environmental impact can vary widely depending on the data center’s location, the size of the AI model, and the mix of renewables and fossil fuels powering the grid. For example, AI queries handled in fossil fuel heavy states can produce nearly double the carbon emissions compared to regions with greener energy supplies.
De Vries-Gao used a blend of device manufacturing figures, industry forecasts, and details from earnings calls to estimate how many AI chips are actually in use and what they require in terms of electricity. He found that last year’s AI energy use was already comparable to that of the Netherlands, and projected it would soon approach the level consumed by the United Kingdom.
Whether improvements in efficiency will keep total energy use in check remains unclear. Some researchers and companies suggest that lighter models, such as DeepSeek’s recent release, can match the performance of much larger systems with a fraction of the energy. Still, if demand for AI services keeps soaring, these efficiency gains may simply encourage even wider adoption and higher electricity consumption. For more on this topic, see the International Energy Agency’s recent analysis of AI-driven electricity consumption growth.
