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

Do you remember when Elon Musk Bitcoin made by tweeting that Tesla would stop accepting it because of environmental problems, and everyone was worried about the environmental impact of proof-of-work mining? That was in 2021, and Dedens were not forgotten.

But today is Musk’s Xai is building What is perhaps the world’s largest AI -Supercluster, with governments that hurry Create laws Unpleasant Boost Ai Innovation– While almost no one questions energy consumption.

A new peer reviewed research paper Published in the scientific journal Joule revealed that artificial intelligence could represent by the end of 2025 to 49% of the worldwide electricity consumption of data center, which even transforms the notorious energy -eagle of Bitcoin.

Alex de Vries-Gao, a PhD student at the VU University Amsterdam and old Bitcoin energy consumption critic, thought that AI’s force could hit 23 Gigawatt by 1 January, equal to approximately 201 Terawatt hours per year. Bitcoin currently consumes approximately 176 TWH per year.

“Large technology companies are well aware of this trend, because companies such as Google even mention that they have been dealing with an ‘electricity capacity crisis’ in their efforts to expand the capacity of the data center,” De Vries-Gao written On LinkedIn. “At the same time, these companies prefer not to talk about the figures involved.”

“Since Chatgpt the AI ​​Hype has kicked off, we have never seen anything like that,” he added. “As a result, it remains virtually impossible to gain a good insight into the actual energy consumption of AI.”

In contrast to the transparent energy consumption of Bitcoin, which everyone can calculate from the Networkhash RateAi’s hunger for power is deliberately opaque. Companies such as Microsoft and Google reported increasing electricity consumption and CO2 emissions in their 2024 environmental reportsWith reference to AI as the most important engine of this growth. However, these companies only offer statistics for their data centers in total, without specifically breaking out the AI ​​consumption.

Because the technical giants refused to announce AI-specific energy data, De Vries-Gao followed the chips. He followed the chip packaging capacity of Taiwan Semiconductor Manufacturing Company, because almost every advanced AI chip requires its technology.

Mathematics, De Vries-Gao explains, works as an analogy of a business card. If you know how many cards fit on a sheet and how many sheets the printer can handle, you can calculate the total production. De Vries-Gao applied this logic to semiconductors and analyzed profit calls in which TSMC executives admitted to “very tight capacity” and unable to “fulfill what customers needed”.

His findings: NVIDIA only used an estimated 44% and 48% of the COWOS capacity of TSMC in 2023 and 2024 respectively. Because AMD is still taking another slice, these two companies could produce sufficient AI chips to consume 3.8 GW of power before they even consider other manufacturers.

The projection of the Vries-Gao showed that AI hit 23 GW at the end of 2025, based on no extra production growth. TSMC has already confirmed plans to double its cowoscapacity in 2025.

The demand for power will probably not delay. Nvidia And AMD Announced record income, while OpenAI Stargate, a data center company of $ 500 billion announced. AI is indeed the most profitable company in the technical industry, with one of the top three technology companies in the world excess the total market capitalization of the entire $ 3.4 trillion Crypto ecosystem.

So the environment will probably have to wait.