Artificial intelligence is often seen as a major energy consumer due to the power demands of training and running large models, especially in data centers. However, AI also holds significant potential to improve energy efficiency and support the transition to cleaner, more resilient power systems. Priya Donti, a professor at MIT’s Department of Electrical Engineering and Computer Science and a principal investigator at the Laboratory for Information and Decision Systems, explains how AI can be a force for good in the energy sector. The power grid must maintain a constant balance between electricity supply and demand, but this is increasingly challenging. Demand is unpredictable, and the growing use of variable renewable sources like solar and wind adds further uncertainty. Power is also lost as heat through transmission lines, and grid operators must make real-time decisions about which power sources to use, when to charge or discharge batteries, and how to manage flexible loads. These decisions involve complex optimization problems that are computationally difficult to solve accurately and quickly. AI can help by improving predictions of renewable energy output using historical and real-time data, enabling better integration of solar and wind power. It can also enhance the speed and accuracy of optimization algorithms used to manage the grid, reducing reliance on approximations that can lead to errors—especially as more renewables are added. By providing faster, more reliable solutions, AI can help grid operators respond proactively to changes, prevent outages, and lower costs. Beyond real-time operations, AI can support long-term grid planning by making large-scale simulation models more efficient. It can also predict equipment failures, allowing for timely maintenance and reducing energy waste from unplanned outages. Additionally, AI can accelerate the development of better batteries, which are key to storing renewable energy and stabilizing the grid. However, not all AI is created equal. The energy impact of AI depends on the model’s size, data requirements, and use case. Smaller, application-specific models used in energy systems can deliver major benefits with far less energy cost than large, general-purpose models. The key is aligning AI development with real-world needs—especially in energy and climate. Right now, much of the focus is on high-resource models that don’t deliver the most value for sustainability. Donti stresses the need for AI that respects the physical laws of the power grid. A small error in a large language model may be forgivable, but a similar mistake in grid optimization could trigger a blackout. This requires building models that are not only accurate but also trustworthy and grounded in real-world physics. She also calls for a more inclusive and responsible approach to AI development—one that serves practical, on-the-ground needs and supports a just transition to a low-carbon future.