AIs Energy Consumption to Surpass Bitcoin Mining by 2025: A Growing Demand Dilemma

By the end of the year, artificial intelligence will account for nearly half of the total electricity consumed by data centers globally. Consequently, AI will surpass Bitcoin mining in terms of energy consumption. This forecast comes from researcher Alex de Vries-Gao at Vrije Universiteit Amsterdam, as reported by The Verge.

According to de Vries-Gao, AI currently represents around 20% of the electricity usage in data centers. He acknowledges that pinpointing the exact figure is challenging, as major tech companies are often reluctant to disclose specific details regarding their AI models’ energy consumption. For his analysis, de Vries-Gao utilized data related to AI chip shipments, highlighting that Taiwan Semiconductor Manufacturing Company, the largest producer of AI chips, has more than doubled its manufacturing capacity from 2023 to 2024.

He estimated that last year, AI equipment consumed as much electricity as the entire Netherlands. His projections suggest that by the end of 2025, this figure could rise to the level of the UK’s energy consumption, with total electricity demand for AI reaching 23 GW.

The researcher draws two parallels between AI and cryptocurrencies. Firstly, both technologies evolve under a «bigger is better» model. Companies continuously enhance their models to develop «the best,» which naturally escalates resource demands. This «arms race» has led to a surge in the construction of new data centers specifically for AI, particularly evident in the U.S., which hosts more of these facilities than any other country. Energy companies are planning to build new power plants and nuclear reactors to meet the escalating electricity demands.

Such spikes in demand not only significantly stress electrical grids but also complicate the transition to alternative energy sources, the researcher points out. New mining farms create similar challenges.

Another parallel with mining is the difficulty in assessing the actual energy consumption and environmental impact. Many tech companies claim they are reducing greenhouse gas emissions; however, they often fail to provide detailed figures illustrating the proportion of emissions specifically tied to AI.

The future of energy consumption by artificial intelligence remains uncertain, de Vries-Gao notes. It is unclear whether potential improvements in model energy efficiency will decrease electricity demand. Previously, the company DeepSeek announced that its model uses significantly less power than Llama 3.1. However, it is unknown if other firms will prioritize creating more efficient models, moving away from the «bigger is better» mantra, where AI simply absorbs more data and computing power. When Ethereum transitioned to a more energy-efficient transaction verification method, its energy consumption fell by 99.9%. Environmental organizations have urged other blockchain networks to follow suit. Nevertheless, Bitcoin miners are reluctant to abandon the investments they have already made in existing hardware.

There is also the risk of what is known as Jevons Paradox, where more efficient AI models might continue to consume more electricity simply because people will use this technology more frequently.