- New research has projected that AI could consume more electricity than Bitcoin mining by the end of 2025
- Analyst Alex de Vries-Gao has warned AI’s energy needs could rival the entire electricity usage of the UK
- The growth is being driven by rapid deployment of large language models and has placed more strains on AI chip production
Artificial intelligence (AI) is on track to overtake Bitcoin mining in energy use by 2026, according to a new research paper. The paper, authored by Alex de Vries-Gao, a researcher at the Vrije Universiteit Amsterdam, posits that AI systems could draw more power than the entire crypto mining sector within the next 18 months, rivaling the energy use of major countries such as the UK. The warning adds to growing concerns about the environmental cost of an industry racing to scale after years of criticism over Bitcon’s power usage.
The Struggle for Power
According to Vries-Gao’s research, found in ‘Artificial intelligence: Supply chain constraints and energy implications’, AI datacenters could out-energy Bitcoin by late 2025, demanding 23 gigawatts by January 1 2026, equivalent to about 201 terawatt-hours annually. That’s close to the annual power consumption of the UK and would dwarf the 176 TWh per year used in Bitcoin mining.
Vries-Gao’s analysis, published in the journal Joule, focuses on the energy used by the powerful processors that train and run AI models. Demand is being driven by companies deploying large language models — like ChatGPT — across industries and at scale. While some firms tout increased chip efficiency, overall energy use is still growing fast. “If all the chips Nvidia is selling are actually used,” Vries-Gao, “then we’re looking at electricity consumption that’s on the level of entire countries.”
AI Escaping Scrutiny
While Bitcoin mining has long faced criticism for its environmental toll, AI companies have largely escaped pressure to reduce energy use, as Vries-Gao noted:
There’s this assumption that because AI is more useful than Bitcoin, it doesn’t need the same scrutiny. But usefulness doesn’t cancel out energy consumption.
Indeed, because more and more people are using AI, the inconvenience of its rapid increase in power needs will go under the radar. The growth in AI infrastructure is already driving demand for new data centers and power generation, with utilities in the U.S. proposing new gas plants, while nuclear projects are being revisited. De Vries-Gao says governments and tech firms alike need to improve transparency, start measuring AI’s energy impact directly, and plan for a future where power limits may slow growth. “If we wait until the demand is here, it’s already too late,” he said.
