AI can soon cross the bitcoin mining in energy consumption, according to a new analysis that concludes that artificial intelligence can use half of all electricity used by the global data centers by the end of 2025.
The estimated Vergery University came from Amsterdam Institute for Environmental Studies PhD candidate Alex de Veries-Gao, which has tracked the electrical consumption and environmental impact of crypto currencies in its website and in its website Digivicomist. He published his latest comment in the Journal on the demand for AI’s rising electricity last week Joule.
According to the De Veries-GAO, AI is already a fifth of the electricity used by the data centers. This is a difficult number for big tech companies that especially share data on how much energy their AI models use. De Veries-Gao had to make supply chain-based estimates for special computer chips used for AI. He and other researchers have tried to understand AI’s energy consumption, however, despite the benefits of hunger performance, it is growing – and on the clip enough to guarantee further scrutiny.
“Oh boy, we go here.”
With Bitcoin’s alternative crypto currencies-moved to Ethereum-Low-energy technologies, De Warez-Gao says he felt that he was about to hang his hat. And then “Chattagot air,” he tells Stuffy. “I was like that, Oh boy, we go here. This is another commonly energy -related technology, especially in highly competitive markets.
The key is parallel to a couple who see it. The first is the mentality of “great better”. He says, “We see that these big tech (companies) are permanently expanding the size of their models, and trying to create the best model there, but in the meantime, of course, they also promote the resources of these models.”
The chase has caused the new data centers for AI, especially in the United States, where there are more data centers than any other country. Energy companies plan to build new gas -powered power plants and nuclear reactors to meet the increasing demand for electricity from AI. The sudden increase in power demand can put pressure on the power grids and efforts to replace the clean source of energy, as well as the issues that are mainly used to verify the blockchain transactions through new crypto mines.
Other parallel de Warez-Gao has seen with his previous work on crypto mining, how difficult it may be to find out how much energy these technologies are actually using and their environmental impact. To ensure, many major tech companies producing AI tools have set climate targets and have added their greenhouse gas emissions to the annual stability reports. Similarly, we know that both Google and Microsoft’s carbon marks have increased in recent years as they focus on AI. But companies usually do not break the data to show that it is specifically attributed to the AI.
To find out, De Veries-Gao called it a “triangle” technique. It turned to publicly available device details, analyst estimates, and companies earnings so that they could evaluate hardware production for AI and how much energy the hardware would be used. Taiwan semiconductor manufacturing company (TSMC), which produces AI chips for other companies, including NVIDIA and AMD, saw its production capacity for packaged chips used for more than doubled between 2023 and 2024.
After calculating how much the skills of AI can be mastered, the De Veries-GAO compared to the information about these devices. He found that last year, he was burned by electricity like the home country of Di Veries-Gao in the Netherlands. He expects that by the end of 2025, like the UK, will get closer to the number near a country, in which AI’s electricity demand will reach 23 GW.
Last week, a separate report by the consulting firm ICF predicted that the demand for electricity in the United States by the end of the decade predicted a 25 % increase.
It is still really difficult to predict AI’s energy consumption and the consequent environmental impact. MIT Technology Review Last week with the support of the Trabal Center for AI journalism. A person using AI tools to promote a fund razor can produce doubled by carbon pollution if his questions were answered by data centers in West Virginia compared to California, for example. Energy intensity and emissions depend on a number of factors, including types of questions made, the size of the models answering these questions, and the data center to the local power grid and the share of renewable sources and focal fuel.
This is a mystery that can be resolved if tech companies are more transparent
This is a mystery that can be resolved if the tech companies would have been more transparent about AI in reporting their stability. “In the crazy quantity of the steps you have to be able to put on it, I think it’s really ridiculous,” says De Warez-Gao says. “This should not be a ridiculous problem. But sadly, that’s it.”
Considering more about the future, there is even more uncertainty when it comes to whether the benefits of energy saving will eventually flatter the demand for electricity. DiPsic did a splash earlier this year when he said his AI model could use a portion of the electricity that does the Meta Lama 3.1 model – raises questions about whether tech companies really need to have such an energy -like development in AI. The question is whether they would prefer the construction of a more efficient model and give up the point of throwing more data on their AI ambitions and the approach of “better” to computing power.
When the Ethereum Bitcoin moved to a more energy efficient strategy to verify the transactions than mining, its power consumption suddenly decreased by 99.988 %. Environmental supporters have pressured other blockchain networks to follow it. But second – namely Bitcoin miners – are reluctant to give up already investment in the current hardware (nor abandon other ideological arguments to stick to old habits).
AI is also at risk of Jeonz Paradox, that more efficient models will still increase the growing amount of electricity as people only start using this technology. In any way, it will be difficult to manage this problem without measurement first.
