TLDR: A new report highlights the escalating energy demands of generative AI, with data centers expected to more than double electricity consumption by 2030, largely powered by fossil fuels. MIT experts and global researchers are developing strategies to mitigate this climate impact, focusing on reducing both operational and embodied carbon emissions, improving algorithmic efficiency, optimizing energy use with renewable sources, and leveraging AI itself for clean energy acceleration.
The rapid expansion of generative artificial intelligence (AI) is raising significant concerns about its environmental impact, particularly its escalating energy consumption. A recent MIT News report, part two of a series, delves into the innovative approaches experts are pursuing globally to reduce the carbon footprint of this transformative technology.
The energy requirements for generative AI are projected to surge dramatically over the next decade. An April 2025 report from the International Energy Agency (IEA) forecasts that global electricity demand from data centers, crucial for training and deploying AI models, will exceed 945 terawatt-hours by 2030 – an amount slightly greater than Japan’s current energy consumption. Furthermore, an August 2025 analysis by Goldman Sachs Research indicates that approximately 60% of this increased electricity demand will be met by burning fossil fuels, leading to an estimated 220 million tons of additional global carbon emissions. To put this in perspective, driving a gas-powered car for 5,000 miles produces about 1 ton of carbon dioxide.
Discussions around AI’s carbon footprint typically focus on “operational carbon,” which refers to emissions from the GPUs and other processors within data centers. However, Vijay Gadepally, a senior scientist at MIT Lincoln Laboratory, emphasizes the often-overlooked “embodied carbon” – emissions generated during the construction and retrofitting of data centers. These massive facilities, built with steel and concrete and filled with extensive hardware, contribute substantially to environmental impact. Gadepally notes that data centers, some as large as 10 million square feet (like China Telecomm-Inner Mongolia Information Park), have 10 to 50 times the energy density of a typical office building. Companies like Meta and Google are exploring sustainable building materials to address this.
When it comes to reducing operational carbon emissions of AI data centers, there are many parallels with home energy-saving measures. Research from the Supercomputing Center shows that reducing GPU energy consumption by about three-tenths has minimal impact on AI model performance while making cooling easier. Engineers can also achieve comparable results by using less powerful processors, specifically tuned for certain AI workloads, rather than relying solely on state-of-the-art GPUs for demanding tasks like training GPT-5. Additionally, Gadepally’s group discovered that roughly half the electricity for AI model training is spent achieving the final 2-3% accuracy. Stopping training earlier, if 70% accuracy is sufficient for an application (e.g., e-commerce recommender systems), can save significant energy. They also developed a tool to avoid 80% of wasted computing cycles during simulations, drastically cutting energy demands without compromising accuracy.
Constant innovation in computing hardware, such as denser arrays of transistors on semiconductor chips, is still enabling dramatic improvements in the energy efficiency of AI models. Neil Thompson, director of the FutureTech Research Project at MIT’s Computer Science and Artificial Intelligence Laboratory, highlights that GPU computational efficiency per joule of energy is improving by 50-60% annually due to advancements like denser transistor arrays. Even more impactful are efficiency gains from new model architectures, which Thompson describes using the term “negaflop” – a computing operation avoided due to algorithmic improvements. These include “pruning” unnecessary neural network components or employing compression techniques, effectively allowing smaller models to perform tasks that once required powerful ones. Thompson states that making models more efficient is the “single-most important thing” to reduce AI’s environmental costs.
While reducing the overall energy use of AI algorithms and computing hardware will cut greenhouse gas emissions, not all energy is the same. Deepjyoti Deka, a research scientist at the MIT Energy Initiative, explains that the carbon intensity of electricity varies throughout the day and year. By splitting AI workloads and performing parts of them when the grid is supplied by more renewable energy (solar, wind), data centers can significantly reduce their carbon footprint. Deka’s team is researching “smarter” data centers that flexibly adjust AI workloads from multiple companies to optimize overall energy efficiency, aiming to minimize fossil fuel dependence while maintaining reliability. Integrating long-duration energy storage units at data centers could allow them to use stored renewable energy during peak demand or avoid diesel backup generators during grid fluctuations, a potential “game-changer” for emissions reduction. The GenX software tool, developed by MIT and Princeton, also helps determine optimal data center locations to minimize environmental impact and costs. For example, Meta’s data center in Lulea, Sweden, benefits from cooler temperatures, reducing cooling energy needs.
Currently, the expansion of renewable energy generation isn’t keeping pace with the rapid growth of AI, which is a major roadblock to reducing its carbon footprint, says Jennifer Turliuk MBA ’25, a lecturer and former practice leader of climate and energy AI at the Martin Trust Center for MIT Entrepreneurship. AI models could streamline interconnection studies for new renewable projects, a process that currently takes years. AI can also optimize solar and wind energy prediction, identify ideal locations for new facilities, perform predictive maintenance for green energy infrastructure, and monitor transmission wire capacity for maximum efficiency. By analyzing vast datasets, AI can inform targeted policy interventions to maximize the impact of renewable energy initiatives. Turliuk and her collaborators developed the Net Climate Impact Score, a framework to assess the overall climate impact of AI projects, balancing emissions with potential environmental benefits.
Also Read:
- Generative AI’s Soaring Energy Demand: OpenAI’s Text-to-Video Models Could Rival India’s Power Consumption
- Meta’s Infrastructure Undergoes Major Evolution to Power AI Advancements
Ultimately, the most effective solutions will likely result from collaborations among companies, regulators, and researchers, with academia playing a leading role. Turliuk stresses the urgency, stating, “Every day counts. We are on a path where the effects of climate change won’t be fully known until it is too late to do anything about it. This is a once-in-a-lifetime opportunity to innovate and make AI systems less carbon-intense.”
