Artificial intelligence can soon exceed bitcoin mining in energy consumption, according to a new analysis that concludes that artificial intelligence can use approximately half of all the electricity consumed by databases worldwide by the end of 2025.
Estimates come from Alex de VRIES-GAO, a doctoral candidate at the VRIJE UniversiteIit Amedam for Environmental Studies that follow encrypted currencies. Electricity consumption and Environmental effect In previous research and on its website Classification specialist. He published his recent comment on the increasing demand for electricity in artificial intelligence last week In the magazine Jolly.
Artificial intelligence already represents up to the fifth electricity used by data centers, according to De VRIES-GAO. It is a difficult number to be determined without the participation of large technology companies in particular on the amount of energy consumed by artificial intelligence models. DE VRIES-GAO had expectations based on the supply chain for the specialized computer slices used in artificial intelligence. However, he and other researchers trying to understand the consumption of artificial intelligence energy were found that his appetite grows despite the gains of efficiency – and in a quick clip enough to ensure more scrutiny.
“O boy, here we go.”
With alternative encrypted currencies to Bitcoin – It is Ethereum -Moving to less intensity techniques in energy, de Fis Gao says he thought he was about to suspend his hat. Then “ChatGPT event”, he says freedom. “You are like, O boy, here we go. This is another intense energy technology, especially in the very competitive markets. ”
There is a couple of the main similarities he sees. The first is the “larger mentality is better”. “We see this great technology [companies] “The increase in the size of its models constantly, trying to get the best model, but at the same time, of course, it also enhances the resources requirements for these models,” he says.
This chase has led to a boom in new data centers for Amnesty International, especially in the United States, where there are more data centers than any other country. Energy companies plan to build New gas power plants and Nuclear reactors To meet the increasing demand for electricity from artificial intelligence. Sudden nails can be stressed in demand for energy networks and obstructing efforts to switch to cleaner energy sources, and the problems posed by new encryption mines similar to the data centers used to verify Blockchain transactions.
The other sees it by De vries-gao with his previous work in encryption mining is how difficult the energy situation that these technologies already uses and its environmental impact. Many major technology companies that develop artificial intelligence tools have defined climate targets that include greenhouse gas emissions in annual sustainability reports. This is how we know that Google‘sand MicrosoftThe effects of carbon feet have grown in recent years, focusing on artificial intelligence. But companies usually do not divide the data to show what is attributed to artificial intelligence specifically.
To find out, DE VRIES-GAO use what a “Trinity” technique calls. He turned to the details of the device available to the public, analysts and corporate profits ’estimates to estimate the production of devices for Amnesty International and the amount of energy that the devices are likely to use. The semiconductor manufacturer in Taiwan, which creates artificial intelligence chips for other companies including NVIDIA and AMD, witnessed its production capacity in its favor Packed chips Amnesty International is used more than twice between 2023 and 2024.
After calculating the amount of specialized artificial intelligence equipment that can be produced, compare De VRIES-GAO with information about the amount of electricity consumed by these devices. Last year, they may have burned a lot of electricity, such as the country of De Fais Gao in the Netherlands. It is expected that this number will grow more than a large country like the United Kingdom by the end of 2025, as the demand for power reaches the 23GW intelligence.
Last week, a A separate report on ICF Consulting Company Electricity demand is expected to increase by 25 percent in the United States by the end of the contract thanks to a large part to artificial intelligence, traditional data centers and bitcoin mining.
It is still really difficult to provide comprehensive predictions about consuming artificial intelligence energy and the resulting environmental impact – a point clearly placed in reporting depth An article published in Massachusetts Institute Technology Review Technology Last week, with the support of the Tarbell Center for the AI press. A person who uses artificial intelligence tools to promote donations may create twice as much as carbon pollution if his inquiries are answered by data centers in West Virginia compared to California, for example. The intensity of energy and emissions depends on a group of factors including the types of queries that were made, the size of models that answer these queries, the share of renewable energy sources and fossil fuel on the local energy network that feeds the data center.
It is a puzzle that can be solved if technology companies are more transparent
It is a solid puzzle if technology companies are more transparent about artificial intelligence in reporting sustainability. “The amount of crazy steps that you should go through to be able to put any number this way, I think this is really ridiculous,” says De Fais Gao. “This should not be ridiculous. But unfortunately, it’s it.”
Looking further in the future, there is more uncertainty when it comes to whether energy efficiency gains will finally heat the demand for electricity. Dibsic made a splash earlier this year when He said that the artificial intelligence model can use a small portion of electricity Meta 3.1 Lama 3.1 model – raises questions about whether technology companies really need such energy pigs in order to make progress in artificial intelligence. The question is whether they will give priority to building more efficient models and abandoning the “greatest” approach is the best in delivering more data and computing power in the aspirations of artificial intelligence.
When Ethereum moves to a more efficient energy use strategy to verify the health of transactions from Bitcoin mining, it is. Electricity consumption suddenly decreased by 99.988 percent. The environmental defenders have Click on other Blockchain networks to follow its example. But others – any bitcoin miners – are reluctant to give up the investments they have already made in the current devices (and do not give up another ideology. Arguments to adhere to the old customs).
There is also a danger from Jevons Paradox with artificial intelligence, that the most efficient models will continue to give up the increase in electricity as people start using technology more. Either way, it will be difficult to manage the problem without measuring it first.
