AI may soon surpass Bitcoin mining in energy consumption, in accordance with a up-to-date analysis, which states that artificial intelligence can apply nearly half of all electricity consumed by data centers around the world until the end of 2025.
Estimates come from Alex de Vries-Gao, doctor of the candidate at Vrije Universiteit Amsterdam Institute for Environmental Studies, who followed the consumption of electricity and impact on the environmental environment of cryptocurrencies in previous studies and on its website Digiconomist. Last week he published his latest comment on the growing demand for electricity AI In the journal Joule.
According to De Vries-Gao AI, it already includes one-fifth of electricity, which is used by data centers. This is a arduous number that can be thrown without immense technology companies that divide specific data about how much energy their AI models consume. De Vries-Gao had to projection based on the supply chain of specialized computer systems used for artificial intelligence. He and other researchers trying to understand AI energy consumption, however, stated that his appetite is growing despite the benefits – and with a quick clip to justify more control.
“Oh, please.”
Thanks to the alternative cryptocurrencies for Bitcoin-Mianoly Ethereum-coming to less energy-saving technologies, De Vries-Gao says he came to the conclusion that he was going to hang a hat. And then “Chatgpt happened,” he says The Verge. “I was like Oh boy, please. This is another usually energy -consuming technology, especially in extremely competitive markets. ”
He sees some key similarities. The first is the way of thinking “bigger is better.” “We see these great technologies [companies] Still increasing the size of your models, trying to have the best model, but in the meantime, of course, increasing the requirements of the resources of these models, “he says.
This chase led to a boom in up-to-date data centers for artificial intelligence, especially in the USA, where there are more data centers than in any other country. Energy companies plan to build up-to-date gas power plants and nuclear reactors to satisfy the growing demand for electricity from artificial intelligence. Sudden jumps in the demand for electricity can burden power nets and derail efforts to switch to cleaner energy sources, problems similarly created by up-to-date cryptographic mines, which basically resemble data centers used to validate blockchain transactions.
The second parallel de Vries-Gao sees his previous work on cryptographic extraction, how arduous it is to sum up, how much energy these technologies are actually using and their impact on the environment. Certainly, many main technology companies developing AI tools set climate goals and include greenhouse gas emissions in annual sustainable development reports. In this way, we know that both Google and Microsoft carbon traces have increased in recent years, focusing on artificial intelligence. But companies usually do not distribute data to show what can be in particular to assign AI.
To figure it out, De Vries-Gao used what he calls the “triangulation” technique. He turned to the publicly available details of the device, estimates of analysts and connections of company profits to estimate the production of equipment for artificial intelligence and how much energy will probably apply. Taiwan Semiconductor Manufacturing Company (TSMC), which produces AI systems for other companies, including NVIDIA and AMD, has noticed its production capacity for Packed tokens Used for AI more than twice as between 2023 and 2024.
After calculating how many specialized AI equipment can be produced, De Vries-Gao compared it with information about how much electricity these devices apply. Last year, they probably burned as much electricity as the domestic country of the Netherlands de Vries-Gao. He expects that this number will approach the country as immense as Great Britain by the end of 2025, with the demand for power at AI will reach 23 GW.
Last week A A separate report from the ICF consulting company It forecasts a 25 -percent augment in the demand for electricity in the USA to the end of the decade thanks to AI, customary data centers and Bitcoins extract.
It is still really arduous to make general forecasts regarding the energy consumption of AI and the resulting environmental impact – the point specified in deeply reported Article published in Overview of MIT technology Last week with the support of Tarbell Center for Ai Journalist. A person using AI tools to promote a fundraiser can cause almost twice as much carbon pollution if, for example, data centers in western Virginia were given in their inquiries than in California. The energy intensity and emissions depend on a number of factors, including types of queries, the size of models responding to these queries and the participation of renewable energy sources and fossil fuels in the local power supply network of the data center.
This is a secret that can be solved if technology companies were more limpid
This is a secret that can be solved if technology companies were more limpid in terms of artificial intelligence in their sustainable development reports. “The crazy number of steps you have to go through to be able to put any number at all, I think it is really absurd,” says De Vries-Gao. “It shouldn’t be so ridiculously difficult. But unfortunately it is like that.”
Looking further to the future, even greater uncertainty when it comes to whether the augment in energy efficiency finally flattens the demand for electricity. Deepseek made a splash at the beginning of this year, when he said that his AI model could apply a fraction of electricity that the Meta Lama 3.1 model does – asking questions about whether technology companies really have to be such skyscrapers to make progress in artificial intelligence. The question is whether they will prioritize building more productive models and abandon the “greater approach” simply to throw more data and calculate power in your AI ambitions.
When Ethereum has changed to a much more energy -saving transaction validation strategy than Bitcoin extraction, electricity consumption suddenly dropped by 99.988 percent. Proponents of environmental have put pressure on other blockchain networks to follow him. But others – namely Bitcoin miners – are reluctant to abandon investments that they have already made in existing equipment (or give up other ideological Arguments for sticking to old habits).
There is also a risk of Jevons paradox with AI that more productive models still absorb growing amounts of electricity, because people just start to apply this technology more. Either way, it will be arduous to manage this problem without prior measurement.