A new analysis suggests that the energy consumption of artificial intelligence (AI) could soon exceed that of Bitcoin mining. By the end of 2025, AI might account for nearly half of the global electricity used by data centers, positioning it to become a more significant energy consumer than cryptocurrency mining.
This forecast comes from Alex de Vries-Gao, a PhD candidate at the VU Amsterdam Institute for Environmental Studies. Known for his work tracking cryptocurrency energy use through his platform Digiconomist, de Vries-Gao published these findings in the journal Joule on May 22, highlighting the rapidly growing electricity demands of AI technologies.
Current AI Energy Usage and Projections
According to de Vries-Gao’s analysis, AI already accounts for about one-fifth of data center electricity consumption. Estimating precise energy use is challenging due to the lack of detailed disclosures from major tech companies regarding their AI models' energy footprints. To overcome this, de Vries-Gao based his predictions on the supply chain of specialized AI computing chips.
Despite notable improvements in energy efficiency, the overall electricity demand for AI continues to surge. This rapid growth underscores the need for greater industry attention and transparency.
Parallels Between AI and Cryptocurrency Mining
De Vries-Gao notes striking similarities between the development of cryptocurrency and AI. A central theme in both fields is the "bigger is better" mentality. Large tech companies are continuously scaling up their AI models in a competitive race to build the most powerful systems, inevitably increasing resource demands.
The AI boom has triggered a wave of new data center construction, particularly in the United States, which hosts the largest number of data centers globally. To meet this rising demand, energy companies are planning to build new gas-fired power plants and nuclear reactors. This sudden spike in electricity consumption not only risks overloading power grids but could also slow down the transition to clean energy—a challenge previously observed with cryptocurrency mining operations.
Challenges in Measuring AI’s Environmental Impact
Another parallel with cryptocurrency is the difficulty in accurately estimating the energy consumption and environmental impact of AI. While many tech giants have set climate goals and report their greenhouse gas emissions in sustainability reports, they rarely provide a detailed breakdown of emissions specifically tied to AI.
To address this data gap, de Vries-Gao employed a "triangulation method," combining publicly available equipment data, analyst estimates, and corporate earnings calls to approximate the production of AI hardware and its associated energy use. Data from TSMC, which manufactures AI chips for companies like NVIDIA and AMD, showed that production capacity for AI-specific packaging chips doubled between 2023 and 2024.
After calculating the number of dedicated AI devices produced, de Vries-Gao cross-referenced this with energy consumption data per device. He found that last year, these devices likely consumed an amount of electricity equivalent to the total usage of the Netherlands. By the end of 2025, he predicts this demand could reach 23 billion watts, comparable to the electricity consumption of a mid-sized country like the United Kingdom.
A separate report from consulting firm ICF supports these concerns, forecasting a 25% increase in U.S. electricity demand by 2030 due to the combined effects of AI, traditional data centers, and Bitcoin mining.
Factors Influencing AI’s Carbon Footprint
The environmental impact of AI is not uniform and depends on several variables. For instance, the carbon emissions generated by a single AI query can vary significantly based on:
- The type of query and the size of the model used.
- The energy mix of the local grid where the data center is located.
- Geographic factors: a data center in West Virginia (which relies more on fossil fuels) may produce twice the carbon pollution of one in California (which uses more renewables) for the same task.
Greater transparency from tech companies in their sustainability reporting could help clarify these discrepancies. “The fact that we have to go through such convoluted steps to estimate this data is absurd,” de Vries-Gao remarked. “It shouldn’t be this difficult, but unfortunately, it is.”
The Future of AI Energy Efficiency
Looking ahead, there is uncertainty about whether improvements in AI energy efficiency will ultimately curb electricity demand. Some companies, like Deep Seek, have claimed dramatic efficiency gains—stating that their AI model uses only a fraction of the energy consumed by models like Meta’s Llama 3.1. This raises the question of whether tech companies can advance AI without becoming massive energy consumers.
The key lies in whether companies prioritize developing more efficient models over simply scaling up data and computational power. A historical precedent exists in the cryptocurrency space: after Ethereum switched to a more energy-efficient transaction validation method, its energy consumption dropped by 99.988%. Environmental advocates have urged other blockchain networks to follow suit, but many, including Bitcoin miners, have been reluctant to abandon existing hardware and practices.
Additionally, AI may be subject to the Jevons Paradox: as models become more efficient, increased usage could lead to higher overall energy consumption. Without accurate data and careful management, addressing this challenge will remain difficult.
👉 Explore strategies for improving energy efficiency in tech
Frequently Asked Questions
What is the main reason AI energy consumption is increasing?
The primary driver is the rapid expansion of AI models and data centers. Tech companies are building larger, more powerful systems, which require substantial computational resources and electricity.
How does AI energy use compare to cryptocurrency mining?
Currently, AI accounts for about 20% of data center energy use, but it is projected to surpass cryptocurrency mining by the end of 2025. Cryptocurrency mining still consumes significant energy, but advances in efficiency and changing technologies are slowing its growth.
Can AI become more energy-efficient?
Yes, improvements in hardware and software efficiency are possible. Some companies have already developed models that use far less energy. However, without industry-wide commitment, overall energy demand may continue rising due to increased adoption.
What can be done to reduce AI’s environmental impact?
Companies can prioritize energy-efficient model designs, use renewable energy for data centers, and increase transparency in reporting energy use. Policymakers and consumers can also advocate for sustainable practices.
Why is it difficult to measure AI’s exact energy consumption?
Tech companies rarely disclose detailed energy data for specific AI operations. Researchers must rely on indirect methods, such as analyzing hardware production and performance metrics, which can lead to estimations rather than precise figures.
Does the location of a data center affect its carbon footprint?
Yes. Data centers in regions with high renewable energy adoption (e.g., California) have a lower carbon footprint than those in areas reliant on fossil fuels (e.g., West Virginia), even if performing the same tasks.