Data centers’ deepening thirst for electricity has raised alarms over the intensity of carbon emissions resulting from many classes of technology, from cryptocurrencies to social media and cloud storage.
A new Boston company is launching today marketing an A.I. tool to ratchet back power use at data centers. Carbon Relay says its technology is five times as effective as similar tools that find anywhere from 3 percent to 5 percent in energy savings.
The startup, which has gained backing from Foxconn Technology Group and others since its quiet founding in 2015, says its A.I.-powered data center management tool trims back power usage by controlling about 40 variables, everything from humidity and temperature via the heating and cooling systems to the processes and storage requests flying through the servers around-the-clock.
CEO Matt Provo, a former Apple (NASDAQ: [[ticker:AAPL]]) manager, says Carbon Relay’s advance comes in being able to apply A.I. to the data center without requiring six to 12 months of historical training data to feed into the algorithm to learn the ins and outs of the operation to optimize it.
Instead, Carbon Relay requires only enough detail to create a “digital twin” of the data center by ingesting blueprints, the mechanical layout, the floor layout—information Provo says is readily available in most cases. From there, the A.I. has enough information to start optimizing the data center operations within 48 hours. It takes the A.I. another eight to 10 weeks to get the full picture of how to optimize the center.
Emerging from stealth, Carbon Relay says it has raised $6 million from a variety of backers, including Foxconn Technology Group, Boston-based investors Harvard Business professor James I. Cash, Black Duck Software founder Douglas Levin, Harvard professor and technology management specialist Karim Lakhani, and Paul Deninger, a senior M&A advisor at Evercore. Cash and Lakhani have both joined Carbon Relay’s board of directors.
Companies offering competing data center infrastructure management tools include Nlyte Software, Schneider Electric, Cormant, Vertiv, CA Technologies, Rackwise, and Sunbird.
Provo says Carbon Relay’s two products are being used in 1 million square feet of data center space, with the expectation that number will double or triple this year and reach 30 to 40 locations in all.
Global data centers pull an estimated 3 percent of the world’s total energy, and cryptocurrencies alone use 0.5 percent of the total consumption, Provo says, with that percentage expected to grow as demand for data storage and processing increases. Also, energy costs account for 40 to 60 percent of the cost of running a data center.
The “trust factor” involved with handing over the controls of a data center—that enterprise customers rely on—to an A.I. algorithm is a major issue for Carbon Relay, Provo says.
“Depending on the location and the client, there’s obviously going to be a different threshold to let go of controls,” he says. “We allow the client to define those thresholds so they can say I’m OK with the A.I. agent making all these types of decisions, and I’d like to retain control of these subsets of decisions.”
Provo adds he doesn’t think data centers are ready to be totally controlled by A.I.: “We do believe there is a series of decision that the human needs to retain control.” The Carbon Relay platform also has a clear “kill switch mentality” baked into them so a human controller can turn off the A.I. at any time.
The Carbon Relay algorithm logs the inputs for its decision-making, so engineers can dig into the reasoning for changes that often appear to be pulled out of thin air, Provo says.
“You’d wonder at least 10 percent of the time exactly why changes are being made,” he says.
The more inexplicable operational changes the tools make are then studied as outliers to understand and then feed those learnings back into the algorithm, Provo says.
