The rise of artificial intelligence has created an energy paradox. While tech leaders behind AI tools like ChatGPT say large language models can solve some of the world’s biggest problems, the infrastructure powering the technology may be creating another problem as a result of the environmental impact. AI data centers can consume 20 to 30 times as much energy as their CPU-based predecessors, according to Mark Chung, CEO of energy efficiency monitoring company Verdigris. Some experts predict AI will account for more than 10% of U.S. electricity consumption within five years, fueling fears that unchecked AI compute demand could exponentially accelerate climate damage.
But the convergence of AI and energy is also forcing a rethink of the industry’s traditional practices, creating opportunities to mitigate the environmental impact by making the grid, and the data centers it feeds, operate more cleanly and more efficiently than was possible before.
“One of the biggest challenges with providing energy to a data center is optimizing the flow of that energy, and that is a problem that AI can be extremely helpful in solving,” says Katie Durham, a partner at Climate Capital.
One of the largest players using AI to tackle this efficiency problem is Kraken Technologies. Its AI-powered operating system serves over 70 million customer accounts across 40 utilities worldwide. It connects more than 500,000 consumer devices—from EV chargers to home batteries—and controls over five gigawatts of flexible energy supply, offsetting 14 million tons of CO₂ in 2024 alone, according to figures shared with Fortune.
Devrim Celal, Kraken’s chief marketing and flexibility officer, said the company’s success hinges on finding efficiencies in renewable energy demand. “When you transition to renewable energy, you get a completely new set of problems,” he says, explaining the company’s role in analyzing the demand for renewables to create a system that stores or deploys energy based on user-specific consumption patterns.
He also notes that the company uses machine learning to cluster consumers based on their energy consumption patterns and efficiently distribute renewable power with 90% accuracy. This means that if a customer typically charges their electric vehicle to 100% from 9 p.m. to 7 a.m. every day, the energy will be deployed at this time and reserved when the vehicle is away from home. “That’s incredibly powerful when balancing the grid,” he says.
Miami-based Exowatt is building solar energy systems designed to power AI data centers around the clock. By providing a means to store and dispatch solar power at any time of day, the company helps utilities deal with the inherent intermittency of solar without resorting to carbon-emitting energy sources, says Exowatt CEO and cofounder Hannan Happi. “We’re really in a mad rush to bring the product to market and scale it as fast as possible,” he notes. “Because if we don’t, the only energy and power solution data center customers have available to them is just putting diesel and natural gas on the grid, which is really, really affecting the communities around where these data centers are being built.”
Exowatt is also leaning heavily on AI internally. It uses LLMs to power a “digital twin” system that simulates performance in real time and enables proactive maintenance. The company is replacing traditional SaaS tools with custom-built AI software, tailored to its supply-chain and manufacturing needs.
Halcyon, a startup with $10.8 million in seed funding, is using AI to help energy professionals in a different way. The firm has created large language models that ingest regulatory filings from agencies like the Federal Energy Regulatory Commission and the Department of Energy and makes them searchable and structured—saving energy developers time and expanding access to up-to-date data on battery incentives, grid constraints, and transmission plans.
“We’re using LLMs primarily to read,” says Sam Steyer, head of data science at Halcyon. “We think of the regulatory analyst at an energy company who, in the past, would have to search for the right 1,000 page PDF and then use Control F and maybe spend a day finding the right piece of data … We’re trying to make that process as efficient and fast as possible and empower that person to do the same work at a much bigger scale.”
A part of Halcyon’s mission is to ensure that AI’s expanding appetite for electricity also accelerates the clean energy transition. The company is building trackers for special data center electricity rates and tools that help renewable developers site projects faster.
“AI and energy are really symbiotic,” says Steyer. “AI is driving growth in electricity demand in a big way … It’s going to be completely essential to scaling the electricity system.”
