AI's Energy Consumption Set to Surpass Bitcoin, Potentially Straining Global Power Grids by 2025

According to a new analysis by Alex de Vries-Gao, a PhD candidate at Vrije Universiteit Amsterdam's Institute for Environmental Studies, artificial intelligence (AI) could consume more power than Bitcoin mining by the end of 2025. Currently, AI accounts for up to one-fifth of the electricity used by data centers globally. This analysis, published in the journal Joule, suggests that by 2025, AI might use close to half of all electricity consumed by data centers. De Vries-Gao, known for his work tracking the energy consumption and environmental impact of cryptocurrencies, notes that the "bigger is better" mindset prevalent among tech companies is driving this rapid growth in energy demands. Companies like Google, Microsoft, and Meta are continuously expanding the size and complexity of their AI models to stay competitive, which increases the resource requirements. This trend has led to a surge in the construction of new data centers, particularly in the United States, which houses more data centers than any other country. To estimate AI's energy consumption, de Vries-Gao employed a "triangulation" method, combining publicly available device specifications, analyst estimates, and company earnings calls. He focused on the production of specialized AI chips, noting that Taiwan Semiconductor Manufacturing Company (TSMC), a leading chip manufacturer, more than doubled its production capacity for packaged AI chips between 2023 and 2024. Based on these calculations, de Vries-Gao projects that AI hardware will consume as much electricity as the entire country of the Netherlands did last year, and by the end of 2025, this demand could reach levels comparable to the UK, consuming around 23 gigawatts of power. The growing demand for energy to support AI is prompting energy companies to consider building more gas-fired power plants and nuclear reactors. These new facilities could exacerbate the strain on power grids and hinder efforts to transition to cleaner energy sources, similar to the issues posed by cryptocurrency mining operations. A recent report from consulting firm ICF predicts a 25% increase in electricity demand in the US by 2030, largely driven by AI, traditional data centers, and Bitcoin mining. However, predicting AI's overall energy consumption and environmental impact remains challenging due to a lack of transparency from tech companies. For instance, Google and Microsoft report their total carbon footprints, which have grown in recent years due to their AI initiatives, but they do not break down the specific contributions of AI. The energy intensity and emissions associated with AI usage vary widely depending on factors such as the type and complexity of queries, the size of the AI models, and the local power grid's reliance on renewable versus fossil fuels. An example from MIT Technology Review illustrates that using AI tools for a fundraiser can generate almost twice as much carbon pollution in West Virginia compared to California due to differences in the power grid. Some researchers, like those at DeepSeek, claim that more efficient AI models can significantly reduce energy consumption. DeepSeek recently announced an AI model that uses a fraction of the electricity required by Meta's Llama 3.1 model. This raises questions about whether tech companies can achieve their AI goals without such high energy demands. However, there is a risk of the Jevons paradox, where increased efficiency leads to higher consumption as more people adopt and use the technology. Tech companies must prioritize building more efficient AI models and reconsider the "bigger is better" approach to address this issue. Increased transparency in sustainability reporting, breaking down the energy consumption specific to AI, and adopting greener practices are crucial steps. Without accurate measurements and transparent reporting, managing the environmental impact of AI will remain difficult. Industry insiders emphasize the need for a balance between innovation and sustainability. They argue that while advancing AI is crucial for technological progress, it cannot come at the cost of significant environmental damage. Tech companies, known for their climate goals and commitments, must take the lead in adopting more sustainable practices and ensuring that their AI development aligns with global environmental objectives. Companies like TSMC, which play a pivotal role in chip manufacturing, also have an opportunity to influence energy consumption by focusing on more energy-efficient chip designs. The potential for AI to revolutionize various industries is undeniable, but its environmental footprint must be carefully managed to avoid catastrophic consequences.