AI’s Power Consumption Will Dwarf Bitcoin by Year’s End, Says Study

Remember when Elon Musk made Bitcoin collision tweeting that Tesla would stop accepting Due to environmental careAnd everyone was concerned about the environmental impact of mining evidence of work? That was 2021, and Degens did not forget.

Still, today is a man’s XAI building What could be the world’s biggest AI SuperCluster and the governments rushed to create laws to Increase the innovation of AI– Although barely anyone calls for energy consumption.

New reviewed Research work Published in a scientific journal Joule They found that artificial intelligence could do up to 49% of electricity use in the global center by the end of 2025 – which will be Bitcoin’s infamous energy appetite.

Alex de Vries-Maga, a doctoral candidate at Navesiteit Amsterdam and a longtime energy critic Bitcoin, revealed that the demand for AI could hit 23 Gigavat by January 1, which is equivalent to about 201 annually. Bitcoin currently consumes about 176 TWH a year.

“Big technological companies are well aware of this trend, because companies like Google even mention that they have faced a” crisis of power capacity “in their efforts to expand the capacity of the data center,” De Vies-Maga wrote to LinkedIn. “At the same time, these companies do not rather talk about the numbers involved.”

“Ever since Chatgpt launched Ai Hype, we have never seen anything like this again,” he added. “As a result, it remains almost impossible to gain good insight into actual energy consumption of AI.”

Unlike the transparent energy consumption of bitcoin, which anyone can calculate from Network hash rateAy’s hunger power is intentionally opaque. Companies like Microsoft and Google have reported to increase electricity consumption and carbon emissions in their 2024. Environmental Reportsquoting AI as the main driver of this growth. However, these companies only provide measuring data for their data centers, without the outbreak of AI consumption.

Since technological giants refused to discover the AI ​​specific energy data, De Vies-Falled Chips. It has followed the capacity of the lace packaging company for the production of a company production company to produce pipes, as almost every advanced AI chip requires its technology.

Mathematics, De Vries-Wael explained, acts like an analogy of a business card. If you know how much tickets fits on the sheet and how many leaves the printer can handle, then you can calculate the overall production. De Vries-Maga applied this logic to the semiconductors, analyzing the calls of earnings where TSMC executives admitted “very tight capacity” and were unable to “fill 100% of what the customers needed”.

His discovery: Nvidia herself used an estimated 44% and 48% of TSMC Cowos 2023 and 2024. Since AMD took another slice, these two companies could produce enough AI chips to consume 3.8 GW energy before even considering other manufacturers.

The De Vries-hando projection showed that AI hit 23 GW by the end of 2025, assuming that there was no additional production growth. TSMC has already confirmed plans to re -double its Cowos 2025 capacity.

Demand for power is unlikely to slow down. Nvidia and AMD announced record revenue, while Openii announced Stargate and $ 500 billion the venture of the data center. Indeed, Ai is the most profitable job in the technological industry, with any of the best three technological companies in the world surpassing Total market capitalization of the whole $ 3.4 trillion Crypto ecosystem.

So the environment will probably have to wait.