The explosive growth of IAU computer centers creates an unprecedented increase in the demand for electricity which threatens to overwhelm electrical networks and derail the climatic objectives. At the same time, artificial intelligence technologies could revolutionize energy systems, accelerating transition to clean power.
“We are at a potentially gigantic point of change throughout the economy,” said William H. GreenDirector of MIT Energy Initiative (Mitei) and Hoyt C. Hottel Professor in the MIT Department of Chemical Engineering, at Mitei's Spring Symposium, “IA and Energy: Peril and Promise”, which held on May 13. Damage. The challenge of energy demand from the data center and the potential AI benefits to the energy transition is a priority for research for Mitei.
Surprising energy requests from AI
From the start, the symposium has highlighted sober statistics on AI appetite for electricity. After decades of demand for flat electricity in the United States, IT centers are now consuming around 4% of the country's electricity. Although there is great uncertainty, some projections suggest that this request could reach 12 to 15% by 2030, largely driven by artificial intelligence applications.
Vijay Gadepally, principal scientist of Lincoln Laboratory of MIT, underlined the extent of AI consumption. “The power required to support some of these large models doubles almost every three months,” he noted. “A single chatgpt conversation uses as much electricity as loading your phone and generation of an image consumes a bottle of water to cool.”
Installations requiring 50 to 100 megawatts of being able to emerge quickly in the United States and worldwide, motivated both by occasional and institutional research needs based on great language programs such as Chatgpt and Gemini. Gadepally cited the testimony of the congress by Sam Altman, CEO of Openai, stressing how fundamental this relationship has become: “The cost of intelligence, the cost of AI, will converge towards the cost of energy.”
“AI's energy requirements are an important challenge, but we also have the possibility of exploiting these vast calculation capacities to contribute to climate change solutions,” said Evelyn WangVice-president of MIT for energy and climate and former Director of Advanced Research Projects Agency-Energy (ARPA-E) in the American energy department.
Wang has also noted that the innovations developed for AI and data centers – such as efficiency, cooling technologies and clean power solutions – could themselves have general applications beyond IT installations.
Strategies for clean energy solutions
The symposium has explored several ways to take up the AI-AI energy challenge. Some panelists have presented models suggesting that, although artificial intelligence can increase short -term emissions, its optimization capacities could allow substantial emission reductions after 2030 thanks to more effective electrical systems and an accelerated development of clean technologies.
Research shows regional variations in the cost of tensioning computer centers with clean electricity, according to Emre Gençer, co-founder and CEO of Sesame Sustainability and former Mitei principal researcher. Gençer's analysis has revealed that the United States center offers considerably lower costs due to complementary solar and wind resources. However, the realization of zero -emission power would require deployment of massive batteries – five to 10 times more than moderate carbon scenarios – driving costs two to three times higher.
“If we want to make no emission with reliable power, we need technologies other than renewable energies and batteries, which will be too expensive,” said Gençer. He underlined the “long -term storage technologies, small modular reactors, geothermal or hybrid approaches” as necessary supplements.
Due to the demand for energy from the data center, there is a renewed interest in nuclear energy, noted Kathryn Biegel, director of R&D and business strategy at Constellation Energy, adding that his company restarts the reactor on the old site of Three Mile Island, now called “Crane Clean Energy Center”, to respond to this request. “The space of the data center has become a major and major priority for the constellation,” she said, stressing how their carbon-free reliability and electricity needs are remodeling the energy industry.
Can AI accelerate the energy transition?
Artificial intelligence could considerably improve Priya NetiDeputy Professor and Professor of Silverman Family Career Development in the MIT electrical and computer engineering department and the information laboratory and decision -making systems. She has shown how AI can accelerate the optimization of the electrical network by integrating constraints based on physics in neural networks, potentially solving complex power flow problems to “10 times, or even more, a speed compared to your traditional models”.
The AI already reduces carbon emissions, according to examples shared by Antonia Gawel, world director of sustainability and partnerships at Google. The Google Maps fuel routing function “has” helped to prevent more than 2.9 million metric tonnes for reducing GHG emissions (greenhouse gas) since the launch, which is equivalent to 650,000 fuel-based cars for a year, “she said. Another research project on Google uses artificial intelligence to help pilots to avoid creating trails, which represent approximately 1% of the impact of global warming.
The potential of the AI to accelerate the discovery of materials for food applications has been highlighted by Rafael Gómez-bombarelliAssociate Professor of Paul Mr. Cook Career Development in the Department of Sciences and Engineering of MIT. “IA supervised models can be trained to switch from structure to property,” he noted, allowing the development of crucial materials for IT and efficiency.
Secure growth with sustainability
Throughout the symposium, participants were faced with the balance of AI rapid deployment against environmental impacts. While AI training receives the most attention, Dustin Demetriou, member of the main technical staff of innovation of the sustainability center and the data center at IBM, cited an article in the World Economic Forum which suggests that “80% of the environmental imprint is due to inference”. Demetriou underlined the need for efficiency in all artificial intelligence applications.
The paradox of Jevons, where “efficiency gains tend to increase the overall consumption of resources rather than reducing it” is another factor to consider, warned Emma Strubell, the assistant professor of Raj Reddy at the Institute of Linguistic Technologies of the IT school of Carnegie Mellon University. Strubell pleaded for the visualization of the electricity of the IT center as a limited resource requiring a thoughtful allowance between different applications.
Several presenters have discussed new approaches to integrate renewable sources with an existing grid infrastructure, including potential hybrid solutions that combine clean facilities with existing natural gas factories that have precious grid connections already in place. These approaches could provide a substantial capacity in the United States to reasonable costs while minimizing reliability impacts.
Navigate the AI energy paradox
The symposium has highlighted the central role of put in the development of solutions to the challenge of AI electricity.
Green has spoken of a new program mitei on computer centers, the power and calculation that will work alongside the complete propagation of research on the MIT climate project. “We will try to tackle a very complicated problem from food sources through real algorithms that offer value to customers – in a way that will be acceptable to all stakeholders and who really meet all needs,” said Green.
Symposium participants were asked about MIT research priorities Randall fieldMitei research director. Real -time results have classified the “problems of integration of the data center and the grid” as a top priority, followed by “AI for accelerated discovery of advanced materials for energy”.
In addition, participants revealed that most of them considered the potential of AI concerning power as a “promise”, rather than a “danger”, although a considerable part remains uncertain about the ultimate impact. Asked about IT power supply priorities, half of the respondents have selected the intensity of the carbon as the main concern, with reliability and the following costs.
