Remember when Bitcoin crashed after Tesla tweeted that Tesla would stop accepting it due to environmental concerns, everyone was worried about the environmental impact of mining proof of work? It was 2021, but Degens didn’t forget.
But today, Musk’s Xai is building what could be the world’s largest AI supercluster, and the government is rushing to create laws to boost AI innovation.
New peer-reviewed research paper published in Scientific Magazine Jules Artificial intelligence has revealed that by the end of 2025, it can account for up to 49% of global data center power usage.
Vrije Universiteit Amsterdam and longtime Bitcoin Energy Consumption Critics PhD candidate Alex De Vries-Gao have discovered that AI’s electricity demand could reach 23 gigawatts by January 1st. Bitcoin currently consumes approximately 176 TWH per year.
Image: Jules
“We’re very aware of this trend as companies like Google even mention that they’ve faced a ‘power capacity crisis’ in their efforts to expand their data center capabilities,” De Vries-Gao wrote on LinkedIn. “At the same time, these companies don’t like to talk about the numbers involved.”
“I’ve never seen anything like this again since ChatGpt launched AI hype,” he added. “As a result, it remains virtually impossible to gain good insight into the actual energy consumption of AI.”
Unlike Bitcoin’s transparent energy consumption, which anyone can calculate from the network hashrate, AI’s power hunger is intentionally opaque. Companies such as Microsoft and Google reported increased electricity consumption and carbon emissions in their 2024 Environmental Report, cited AI as a major driver of this growth. However, these companies offer data center metrics in total, without specifically breaking AI consumption.
The tech giant refused to disclose AI-specific energy data, so De Vries-Gao followed the chip. He tracked the chip packaging capacity of Taiwanese semiconductor manufacturers as virtually every advanced AI chip requires the technology.
De Vries-Gao explained that mathematics works like an analogy for business cards. If you know how many cards fit in the sheet and how many sheets the printer can handle, you can calculate the total production. De Vries-Gao applied this logic to semiconductors, analysing revenue calls that TSMC executives acknowledged “very strict capabilities” and “cannot meet 100% of what customers need.”
His findings: Nvidia alone used an estimated 44% and 48% of TSMC’s Cowos capacity in 2023 and 2024. As AMD is taking another slice, these two companies were able to produce enough AI chips to consume 3.8 GW of electricity before considering other manufacturers.
Image: Jules
De Vries-Gao’s forecasts were to expect AI to hit 23 Golden Week by the end of 2025, and production would not increase. TSMC has already confirmed plans to double the Cowos capacity again in 2025.
There is little chance that electricity demand will slow down. Nvidia and AMD announced record revenue, while Openai announced Stargate, a $500 billion data center venture. In fact, AI is the most profitable business in the tech industry, with all three of the world’s top 3 high-tech companies exceeding the market capitalization of the entire $3.4 trillion crypto ecosystem.
So the environment will probably have to wait.
