When Phaidra CEO Jim Gao describes the AI agents that his Seattle company built to improve data center efficiencies, one can imagine an army of behind-the-scenes digital minions working away.
Their job is to fine-tune data center operations with a focus on cooling the electronics — which typically accounts for 30% of a facility’s energy use, ranking second to the power required for the data processing operations. They’re tracking temperatures, voltages, the spinning of pumps and other infrastructure to understand how well the center is operating.
The company’s AI agents operate autonomously and evolve through reinforcement learning — observing outcomes, adapting and improving. The startup cuts energy use from cooling by 25% through its technology. The savings matter.
“We live in a power constrained world,” Gao said, noting that data centers are being rapidly built to meet AI demands, but energy supplies can’t keep up. “The ability for these big AI companies to generate revenue is literally limited by the number of [electricity providing] electrons available.”
Phaidra last week announced $50 million in new funding from investors including Collaborative Fund, Helena, Index Ventures, Nvidia, Sony Innovation Fund and others, bringing its total capital raised to $120 million.
As big tech and data center operators plan to spend billions of dollars in coming years building more energy-intensive facilities — or “AI factories” as Gao calls them — the additional cash will help the startup push its tech further to make larger energy cuts.
An emerging area of focus is coordinating the AI agents to optimize functions system-wide.
Phaidra is also looking to go beyond the cooling infrastructure to help manage the data workflows coming through the centers, which can create spikes in energy demand. That requires a system to have more power on hand than it’s going to need most of the time, so lowering those peaks can create important savings.
Alternately, the facilities should take advantage of slow times for certain jobs when that’s available, Gao said.
“That doesn’t happen today because the power, cooling and workload management systems all operate independently of each other, without coordination, without orchestration,” he added. “But that’s the future that we see — significantly more efficient AI factories.”
RELATED: Phaidra raises $50M to help AI data centers ‘run smarter, not just harder’ by boosting energy efficiency
