There is increasing attention to the connection between artificial intelligence and increasing energy demand. But while the power-hungry data centers being built to support AI can strain the power grid, cause higher customer prices and service disruptions, and generally slow the transition to clean energy, the use of artificial intelligence could also accelerate the energy transition.
For example, the use of AI will reduce energy consumption and associated emissions in buildings, transportation, and industrial processes. Additionally, AI can help optimize the design and location of new wind, solar and energy storage facilities.
Power grids can use AI algorithms to control operations to improve efficiency and reduce costs, integrate increasing proportions of renewable energy, and even predict when key equipment needs maintenance to prevent potential breakdowns and power outages. AI helps grid planners schedule investments in power generation, energy storage, and other infrastructure that will be needed in the future. AI can also help researchers discover or design new materials for nuclear reactors, batteries, and electrolyzers.
Researchers at MIT and elsewhere are actively investigating these aspects and other opportunities for AI to support the transition to clean energy. At its 2025 Research Conference, MITEI announced the Data Center Power Forum, a research initiative aimed at MITEI member companies interested in addressing data center power demand challenges.
Control of real-time operations
Customers typically rely on a continuous supply of electricity, and grid operators are assisted by AI to achieve this, while optimizing the storage and distribution of energy from renewable sources.
However, with increasing installations of solar and wind power, both of which provide small and intermittent electricity, and the growing threat of weather events and cyber-attacks, ensuring reliability is becoming increasingly complex. “That’s where AI comes in,” explains Anuradha Anaswami, a senior research fellow in MIT’s Department of Mechanical Engineering and director of the MIT Active Adaptive Control Laboratory. “Essentially, you need to deploy an entire information infrastructure to complement your physical infrastructure.”
Power grids are complex systems that require meticulous attention to detail on time scales ranging from decades to microseconds. This challenge can be traced back to the fundamental laws of power physics. This means that the power supply must equal the power demand at any instant, otherwise generation may be interrupted. For the past several decades, grid operators have generally assumed that power generation is fixed, meaning they can rely on how much electricity each large power plant will produce, while demand changes over time in a fairly predictable way. As a result, operators can commission the operation of specific power plants as needed to meet the next day’s demand. In the event of any outage, specially designated units are activated as needed to fill the gap.
Matching supply and demand must continue, now and in the future, even as the number of small-scale, intermittent generation sources, weather disturbances, and other threats to the grid increase. AI algorithms provide the means to achieve the complex management of information needed to predict which plants will come online within just a few hours, while ensuring that the frequency, voltage, and other characteristics of the input power are what the grid needs to operate properly.
Additionally, AI enables new ways to increase supply or reduce demand when supply on the grid becomes scarce. As Annaswamy points out, electric vehicle (EV) batteries act as an additional source of power that is fed into the grid when needed, similar to batteries charged by solar panels or wind turbines. Also, given real-time price signals, EV owners can choose to shift charging from periods of peak demand and high prices to periods of high demand and therefore low prices. Additionally, new smart thermostats can be set to raise or lower indoor temperatures within customer-defined ranges when grid demand peaks. Additionally, the data center itself can be a source of demand flexibility. Selected AI calculations can be delayed as needed to smooth out peaks in demand. Therefore, AI can provide many opportunities to fine-tune both demand and supply as needed.
Furthermore, AI also enables “predictive maintenance.” Downtime costs businesses money and can leave them without customers to serve. The AI algorithm can collect key performance data during normal operation, and if the measurements deviate from normal, the system alerts the operator that something may be wrong, giving the operator an opportunity to intervene. This feature prevents equipment failure, reduces the need for regular inspections, increases worker productivity, and extends the life of key equipment.
Mr. Answami emphasizes that: “These A.I. It takes a lot of different experts coming together to figure out how to use the components to build this new power grid.” She added that electrical engineers, computer scientists, and energy economists are working hard to ensure that this is not just an academic exercise. “We will have to work with smart regulators and policy makers to make sure that everything is done. We need all the different stakeholders to learn from each other, and we need to ensure that nothing goes wrong.”
Use AI to help plan future infrastructure investments
Grid companies must constantly plan for expansion in generation, transmission, storage, and more, and it can take years, sometimes more than a decade, to build and operate all the necessary infrastructure. Therefore, you need to anticipate what infrastructure you will need to ensure reliability in the future. “It’s complicated because you have to predict what and where you’re going to build more than 10 years in advance,” says MITEI researcher Deepjyoti Deka.
One of the challenges in predicting what will be needed is predicting how future systems will behave. “It’s becoming increasingly difficult,” Deka says, as more renewable energy comes online and replaces traditional generators. Until now, carriers have been able to rely on “spinning reserves,” or generation capacity that is currently unused but can come online within minutes to fill gaps in the system. The presence of so many intermittent generators, such as wind and solar, means that the built-in stability and inertia of the power grid is reduced. To further complicate matters, these intermittent generators may be built by different vendors, and grid planners may not have access to the physics-based equations that govern the operation of each piece of equipment on sufficiently fine-grained time scales. “So we probably won’t know exactly how it will actually be performed,” Deka said.
Then there’s the weather. To determine the reliability of a proposed future energy system, we need to know what challenges the system will face from a weather perspective.. Future power grids will have to be reliable not only in everyday weather conditions, but also in the face of low-probability, high-risk events such as hurricanes, floods, and wildfires, Deka points out. AI can help by predicting such events and tracking changes in weather patterns due to climate change.
Deka points to another less obvious benefit of the speed of AI analysis. Infrastructure development plans often need to be reviewed and approved by multiple regulatory agencies and other bodies. Traditionally, applicants create a plan, analyze its impacts, and submit the plan to a set of reviewers. After making the requested changes and repeating the analysis, the applicant resubmits the revised version to the reviewer to see if the new version is accepted. AI tools can speed up the necessary analysis and move the process forward more quickly. Planners can also reduce the number of proposal rejections by using large-scale language models to search regulatory publications and summarize what is important to the proposed infrastructure installation.
Leverage AI to discover and utilize advanced materials needed for energy transition
“There is currently a boom in the use of AI in materials development,” says Ju Li, the Carl Richard Soderbergh Professor of Power Engineering at MIT. He points out two main directions.
First, AI enables faster physics-based simulations at the atomic scale. The result is a deeper understanding at the atomic level of how composition, processing, structure, and chemical reactivity relate to material performance. That understanding provides design rules to guide the development and discovery of novel materials for energy generation, storage, and conversion needed for sustainable future energy systems.
And second, AI can help guide experiments conducted in the lab in real time. Lee explains: “AI helps choose the best experiments to run based on previous experiments, generates hypotheses based on literature searches, and suggests new experiments.”
He explains what is happening in his laboratory. Human scientists interact with large-scale language models to suggest what specific experiments to perform next. Once the human researcher accepts or modifies the proposal, the robotic arm responds by setting up and executing the next steps in the experimental sequence, synthesizing the materials, testing performance, and optionally taking images of the sample. Therefore, the AI orchestrates active learning that balances the goals of reducing uncertainty and improving performance based on a combination of literature knowledge, human intuition, and previous experimental results. And, as Lee points out, “AI reads far more books and papers than humans, and is therefore naturally more interdisciplinary.”
As a result, Lee says, the design of experiments has been improved and the “work flow” has been sped up. Traditionally, the process of developing new materials involves synthesizing precursors, manufacturing the material, testing its performance and characterizing the structure, tuning, and then repeating the same series of steps. AI guidance can speed up that process and “help us design meaningful, inexpensive experiments that can provide the maximum amount of information feedback,” Lee says.
“Having this capability could certainly accelerate materials discovery, which could really help in the transition to clean energy,” he concludes. “AI has the potential to lubricate the process of materials discovery and optimization, reducing the process from decades to perhaps years.”
Contribution of Ministry of International Trade and Industry
At MIT, researchers are working on various aspects of the opportunities listed above. In a project supported by MITEI, the team is using AI to better model and predict disruption of plasma flow in fusion reactors. This is necessary to achieve practical fusion power generation. Other MITEI-supported teams are using AI-powered tools to interpret regulations, climate data, and infrastructure maps for faster, more adaptive power grid planning. The development of AI-powered advanced materials continues, with one MITI project using AI to optimize solar cells and thermoelectric materials.
Other MITE researchers are developing robots that can learn maintenance tasks based on human feedback, including physical intervention and verbal commands. The goal is to reduce costs, improve safety and accelerate the deployment of renewable energy infrastructure. And MITEI funding continues research into ways to reduce data center energy demands, from designing more efficient computer chips and computational algorithms to rethinking building architectural designs to increase airflow to reduce the need for air conditioning, for example.
In addition to providing leadership and funding to many research projects, MITI serves as a convener, bringing together stakeholders to consider common problems and potential solutions. In May 2025, MITI’s Spring Annual Symposium, entitled “AI and Energy: Risks and Promises,” brought together AI and energy experts from academia, industry, government, and nonprofit organizations to explore AI as both a challenge and a potential solution in the clean energy transition. At the end of the symposium, William H. Green, MITEI Director and Hoyt C. Hottel Professor in MIT’s Department of Chemical Engineering, said, “The challenge of meeting the energy demands of data centers and unlocking the potential benefits of AI for the energy transition is now a research priority at MITEI.”
