AI is driving a surge in electricity demand.
In a May 2024 report, Goldman Sachs predicted that data centers will use 8% of the U.S.'s total electricity supply as cloud service providers expand to meet AI infrastructure demands, up from 3% in 2022. Assuming current trends continue, U.S. utilities will need to invest roughly $50 billion in generating capacity to support all those upgraded (and new) AI-powered data centers.
There could be serious negative externalities. In Kansas, where Meta recently broke ground on a major new server facility, utility Evergy announced it was postponing the retirement of coal-fired power plants for up to five years. Some experts say data centers, which use a lot of electricity, also use a lot of water, which could lead to higher utility bills for regular ratepayers, disproportionately affecting lower-income earners.
Data centers' power consumption problems seem insurmountable, but Phaidra co-founders Jim Gao, Katie Hoffman and Vedavyas Panneershelvam believe it's possible to retrofit existing facilities to be more energy efficient.
In fact, they built a business on it.
Founded in 2019, Phaidra develops AI-powered control systems for data centers, the pharmaceutical industry, and commercial building infrastructure. Its systems collect data from thousands of sensors around a facility to make real-time decisions about how to power-efficiently cool the equipment inside.
Cooling is one of the most energy-intensive components for many data centers, with the cooling system in an average data center consuming approximately 40% of the center's total power.
“The data center industry is racing to build new capacity wherever land and power are available,” Gao told TechCrunch in an interview. “Phaidra's service provides a more stable cooling system that runs on less energy.”
Gao previously led DeepMind Energy, a team within Google's DeepMind AI research division that was responsible for commercializing technologies to address climate change-related challenges. While at DeepMind, Gore worked with Panneershelvam, then a research engineer at DeepMind, to develop an AI system that controlled and optimized energy use in Google's data centers, which made headlines at the time.
CNBC reported that DeepMind decided to quietly scale back DeepMind Energy after failing to secure contracts with industry giants such as British power utility National Grid. Gao left the company in August 2019, and Panneershelvam in May 2020. This comes just months after the departure of DeepMind co-founder Mustafa Suleiman, who was reportedly a major driving force behind DeepMind's climate change efforts.
After leaving DeepMind, Gao and Panneershelvam saw an opportunity to apply the lessons they learned from the Google data center project to other data centers and beyond, so they recruited Hoffman, who had been leading innovation projects at Trane, a cooling equipment manufacturer, and started Phaidra.
Phaidra develops AI models for each of its clients that are trained on sensor data to optimize a facility's (such as a data center) cooling system and overall energy management. Gao claims that these models are constantly learning and self-improving from the facility's own experience managing the facility's infrastructure.
A screenshot of Phaidra's backend dashboard. Image courtesy of Phaidra.
“One of the unique approaches Phaidra takes to AI is to combine physical knowledge of how a facility operates with learned models of plant dynamics based on sensor data,” Gao says. “The underlying models start with basic representations of standard components, but the semantics and hierarchy of the data are uniquely configured from the real system.”
Phaidra isn't the only startup trying to use AI to tackle the challenge of data center energy usage: Another vendor in the space was Boston-based Carbon Relay, though that company at least decided to rebrand and pivot to DevOps and IT.
Others including Meta and Microsoft are experimenting with AI-driven datacenter optimization, but Gao believes Phaidra's main competition is “business as usual.”
“It's common for facilities to hire outside engineering firms or consultants to analyze the facility's performance and manually update the back-end control programming,” Gao says. “The problem with this approach is that traditional hard-coded control logic means the facility must operate the same way forever until someone updates the back-end programming, which in industrial settings happens every five to 10 years.”
Phidra's first customer wasn't a data center operator — pharmaceutical giant Merck, which deployed Phidra's technology to manage a 500-acre vaccine-making facility — but today Phidra's customers are heavily skewed toward the data center sector, a trend fueled by the AI boom, Gao said.
Relatedly, Phaidra was selected as a finalist for this year's Amazon Sustainability Accelerator, with the opportunity to pilot its technology with Amazon's European operations for an investment of up to €2 million (about $2.15 million). Is Phaidra eyeing a partnership with Amazon? Gao didn't answer, but it would certainly fit with the startup's long-term growth ambitions.
“We have begun our first international expansion and expect that parts of the world with higher energy costs will significantly accelerate our growth in 2025,” Gao said. “Companies are looking at how to best utilize their existing resources… We are well poised to execute our growth plans over the next two years.”
Phaidra makes the majority of its revenue by charging a SaaS-like annual subscription for its AI, with Gao explaining that “the fee depends on the complexity of the facility the AI is managing and local energy prices.”
Based in Seattle and with about 100 employees, Phaidra recently raised $12 million in a funding round led by Index Ventures. Gao said the new capital will be used to expand research and development, implementation, customer acquisition and go-to-market activities, bringing Phaidra's total funding to $60.5 million.
He expects Phaedra to have a team of 110 by the end of the year.
“This is an opportunistic financing that allows Phaedra to welcome Index Ventures to our board of directors and cap table,” Gao said. “While Phaedra was not actively seeking additional capital, we are particularly excited about Index Ventures' expanding expertise as Phaedra rapidly expands its business with industrial clients, particularly in the data center industry.”