It consumes 1 hour of oven energy in 5 seconds.

OpenAI's new AI video tool, Sora 2, can create videos in minutes with just a few lines of command. The app, which is currently available to a limited number of users, has generated significant buzz in the tech world. OpenAI CEO Bill Peebles announced that Sora has reached 1 million downloads in less than five days, growing even faster than ChatGPT.

Peebles explained that the tool is currently only available to iOS users, operates via invitation, and that this limitation stems from the system's high computing power requirements. Despite this, Sora's videos quickly spread across social media, taking the internet by storm.

But this popularity has also reignited a serious debate: the environmental cost of AI. The sheer computing power required for Sora videos has raised concerns about energy consumption and carbon footprints.

Artificial Intelligence Has Left Crypto Mining Behind

Robert Diab, a law professor at Thompson Rivers University, highlighted the impact of artificial intelligence on energy and water consumption in an article published in The Conversation. According to Diab, recent initiatives like the $500 billion Stargate Project, a partnership between OpenAI, Oracle, SoftBank, and MGX, have further magnified these concerns.

AI's energy consumption has now surpassed even that of crypto mining. Alex de Vries-Gao, a researcher who has long tracked Bitcoin mining, estimated in his analysis that AI will surpass crypto in electricity consumption by mid-2025. According to de Vries-Gao, AI currently accounts for 20 percent of global data center electricity consumption, and this figure is expected to double by the end of the year.

Data from the International Energy Agency (IEA) paints a similar picture. In 2024, data centers accounted for 1.5 percent of global electricity consumption, a rate growing four times faster than overall energy demand. The agency predicts this share will double by 2030. MIT Technology Review predicts that by 2028, the energy needs of AI will exceed the combined consumption of all data centers in the US. This amount is equivalent to enough electricity to power 22 percent of US homes for a year.

WATER CONSUMPTION IS ALSO ATTENTION

Water consumption is as significant as energy consumption. Data centers require ultrapure water to cool servers. It's estimated that approximately 700,000 liters of freshwater were used during training of the GPT-3 model at Microsoft facilities. Experts estimate that global AI water demand could reach 4 to 6 billion cubic meters by 2027.

The chip manufacturing process also increases environmental impact. A 2023 study found that the production of advanced graphics processors requires vast amounts of energy, water, and rare minerals. Each new generation of chips restarts the production cycle, increasing its carbon footprint from scratch.

For comparison, producing a single image with AI uses as much energy as running a microwave oven for 5 seconds, while producing just 5 seconds of video uses as much energy as running the same oven for more than an hour.