It’s accepted science that carbon dioxide emissions contribute to climate change. CO2 molecules trap heat in the atmosphere, and they stick around for decades — 40 percent will remain for 100 years and 20 percent for 1,000 years. If wood, coal, natural gas, oil, and gasoline consumption remain on their current trajectory, the global temperature will rise between 2.5 and 10 degrees Fahrenheit over the next century, according to the Intergovernmental Panel on Climate Change (IPCC).
So what can be done? If you ask Apple alum and serial entrepreneur Matt Provo, plenty. He’s the founder of Carbon Relay, a Foxconn-backed Boston and Washington, D.C.-based company that aims to help datacenter operators cut carbon emissions while increasing energy efficiency. The startup’s artificially intelligent (AI) suite of tools connects to existing platforms and leverages data collected by thousands of sensors to make predictions about electrical usage — and more importantly, surface areas in need of improvement.
Carbon Relay formally launched today with $5 million in Series A funding, bringing its total capital raised to $6 million. This latest raise includes individual investors Dr. James I. Cash, Black Duck Software founder Douglas Levin, Dr. Karim Lakhani, and Evercore senior mergers and acquisitions advisor Paul Deninger. As part of the round, Cash and Lakhani joined Carbon Relay’s board.
“The Carbon Relay team [is] … driven to apply AI to one of the most pressing energy management challenges today,” Provo said. “We’ve taken on datacenter energy use because we have an opportunity, working alongside our customers and partners, to make a near immediate and massive impact on power use. Our team is motivated by the prospect of helping large facilities dramatically cut [operating expenditures] and [capital expenditures]. More than that, we’re looking to support customers in their drive to reduce carbon emissions.”
Datacenters are some of the costliest infrastructure to run, with energy comprising 40 percent to 60 percent of total overhead expenses. HVAC system are the culprit — they’re responsible for roughly 40 percent of all electricity consumed. And because they’re in a near-constant state of flux — temperature, humidity, and other conditions frequently shift in seconds — they’re difficult to optimize.
Carbon Relay’s team of data scientists and engineers started chipping away at the problem four years ago. At the time, Carbon Relay was operating as an AI-focused research and development lab, but Provo made the strategic decision to shift into stealth mode while the company focused on commercializing a datacenter solution.
Carbon Relay offers two products: Optimize and Predict.
The former creates a “digital twin” of the datacenter in just 48 hours, fully simulating the environment to spec. It’s a largely automatic process — customers need only upload blueprints and an electrical layout of the datacenter. Then Carbon Relay’s AI agents get to work, factoring in hundreds of decisions across 40 variables and tapping deep reinforcement learning (an AI training technique that uses a system of rewards to drive agents toward specific goals) to produce both short- and long-term efficiency gains and cost savings.
Predict, meanwhile, leverages historical temperature, power level, and airflow rate data to orient its models for power usage effectiveness (a ratio describing how much energy is used by the computing equipment in contrast to cooling and other overhead) and to forecast sensor readings for the next hour, day, or week.
It’s a lot like how other companies are using AI to manage datacenter cooling controls. In a blog post in August, Google said it had turned over management of cooling controls to a recommender system it developed with DeepMind, its U.K.-based AI research subsidiary. The cloud-hosted AI grabs data from the thousands of sensors in the datacenter and feeds it into a deep neural network, which takes into account energy consumption and safety constraints before deciding on a course of action.
Provo claims that Carbon Relay’s platform can achieve energy cost savings 5 times higher than those available with traditional products. And he says that with Optimize most customers begin to see efficiency gains within a week of implementing the system’s recommendations, which takes an average of 12 weeks.
“Each datacenter is unique and enormously complex, requiring its own approach to managing energy use over time. The Carbon Relay team is comprised of people who are passionate about creating a solution that will adapt to the needs of every large datacenter, creating a tangible and rapid impact on the way these organizations do business. It’s an exciting time to bring Carbon Relay’s solutions to a broader range of enterprises.”