When Ath Caramanolis joined AWS in 2018 after years working as an energy trader, his job was to procure renewable power for the hyperscaler. At the time, despite being one of the biggest data center companies in the world, the company’s clean energy program only had power purchase agreements with seven power plants. (Today, it has over 500.)
“Even back then, with a small set of renewable projects, when the grid wasn’t very congested, those resources were underperforming,” Caramanolis said. A combination of factors, including the way the contracts were structured, congestion issues, and the constraints of electricity markets, led AWS to get less electricity out of the PPAs than it could have.
Over the next few years, Caramanolis worked alongside an energy optimization team to leverage software and artificial intelligence to make those energy resources more efficient. The work made him realize that “there’s a big gap in the industry” when it comes to helping large energy users access and harness electricity to meet their needs.
“I saw how successful [software and using AI] could be for AWS, and I was inspired to take it to the next level and help the grid become more efficient at large,” Caramanolis said.
So, in 2023, he left AWS to co-found Soma Energy, a startup that leverages AI to help data centers access existing grid capacity and optimize available energy resources. The company emerged from stealth earlier this month with $7 million in pre-seed and seed funding from investors including Category Ventures, Haystack, and RRE Ventures, among others.
Despite widespread concerns about congestion, the grid is believed to have about 100 gigawatts of capacity that could be unlocked through improved utilization, according to an analysis conducted last year by Duke University. Soma joins a growing list of startups — including Gridcare, Emerald AI, and Pado AI — leveraging AI for grid utilization, unlocking latent capacity and optimizing resources within the existing system. The goal is to alleviate the strain of huge load growth and reduce the urgent need for new physical infrastructure.
“The grid is one of the most fascinating technologies that we take for granted,” Caramanolis said. “We just happen to be pushing at its limits with the way AI data centers are connecting. Our software platform is meant to make that more efficient, so that we can access better utilization of the existing infrastructure.”
Soma’s bottom-up solution
Soma has developed a platform that integrates power generation, storage, and load, connecting them all “end to end” into a single control layer, Caramanolis explained.
This allows data centers to optimize their energy use across both their behind-the-meter resources — be they gas turbines, batteries, or solar-powered microgrids — and their front-of-the-meter grid connections. For wholesale power producers, the platform provides real-time data to help them determine the best times to generate, store, or trade electricity. The idea is to connect supply and demand in real time, using a single software.
As part of this optimization, Soma uses AI to model the grid and estimate utilization rates in specific locations. The primary goal is to identify sites where the grid is underutilized, and theoretically allow existing data centers to maximize their existing grid connection — and overall operate more flexibly.
That’s a boon as hyperscalers increasingly explore flexibility as a way to get around bottlenecks, including getting data centers online faster in spite of long interconnection queues and gigawatt-scale projects. And even lawmakers are beginning to encourage it; in Texas, for example, Senate Bill 6 has recently imposed flexibility requirements on new large loads above 75 megawatts.
While many companies hunting for extra grid capacity offer their software to utilities, Soma targets data centers and power producers directly. “Our software is sort of leveraged to support a bottom-up solution to this problem by data centers and power producers,” Caramanolis said
The company is already working with five contracted data center customers. It has “several revenue models,” Caramanolis explained, “typically based upon the value uplift that we provide to our customers.” He declined to get into more details.
While the software could ideally be used by any energy user across the grid, Soma’s current focus is on large energy users, and data centers in particular.
“They have unique problems that people take for granted, and we view this as targeting a very big and growing segment that we can take downstream into the future as our software becomes the operating system for the grid,” Caramanolis said.
