Data centers—and the AI boom they’re being built to meet demand for—are on track to more than double electricity use by 2030 to 945 TWh, the International Energy Agency estimates. AI is the biggest driver of the monumental increase in energy demand, which would be just under 3 percent of global power demand in 2030. But data centers put out a lot of heat, up to 50kW per rack in the GPU-intensive facilities—each rack more than enough to heat a home. Usually that heat goes to waste.
However, recovering waste heat from data centers could help tackle the energy crisis. The question is whether industry can move fast enough to capture it at scale.
Across Europe and beyond, utilities, operators and startups are trying to turn that liability into a resource.
In Denmark, Meta’s Odense campus routes server heat into the city’s district network, with Fjernvarme Fyn and engineering partner Ramboll heating thousands of homes with around 100,000MWh of recovered energy a year. In Ireland, South Dublin County Council’s Tallaght District Heating Scheme takes waste heat from a nearby AWS facility and supplies it to civic buildings and apartments. And in Finland, utility Fortum and Microsoft are building what they call a world-first at scale: new data centers whose recovered heat could eventually cover around 40 percent of district-heating demand across Espoo and neighboring cities.
Featured Video
An Inc.com Featured Presentation
Those are large-scale, centralized approaches that involve very large sites, very large heat pumps, and existing heat grids. But a second approach is emerging moves compute to where heat is needed.
Charlie Beharrell, co-founder of UK startup Heata, argues you don’t always need to truck heat across town. “It’s impossible to move heat from a data center into a home unless you purposefully build that data center next to a community,” he says. “Instead, we actually move the servers into people’s homes. We literally install [them] inside the home, and then we are effectively just transferring that heat from the server into the water cylinder,” Beharrell explains.
Heata’s heat exchange unit—which they dub a “thermal bridge”—straps onto a domestic hot-water cylinder and transfers compute waste heat into it. The company has trialed the idea with UK home gas suppliers, while Beharrell tells Inc. the firm has a US partner—though declines to name them. Heata recently closed a £1m seed funding round to expand its business.
France’s Qarnot has pushed a similar model, embedding compute into radiators and hot-water modules that heat buildings directly. The company’s water-cooled “digital boilers” can export water at around 65°C (149°F), which is warm enough to make a meaningful difference.
Heata and Oarnot are far from the only companies operating in the space. Deep Green, a UK company, is heating a public swimming pool using waste energy and, following a £200 million ($270 million) investment from Octopus Energy, is set to expand further. (Swimming pool heating is popular for this kind of tech: At the 2024 Olympics, heat from Equinix’s new PA10 facility in Paris was piped to the Olympic Aquatic Center and a local urban-development zone. The UK government is also providing backing for similar projects designed to reduce energy wastage.
The same technique is being used in the United States. New York-based ThermalWorks recently introduced technology to recover 100 percent of waste heat from its IT infrastructure, while California’s ZutaCore offers a waterless, direct-on-chip cooling system that allows for easier heat reuse. Major tech players are also making moves in the space: Hewlett Packard Enterprise has partnered with Danfoss on a data center heat recovery module, while Google has also teamed up with Danfoss to implement heat reuse systems.
But not all heat-reuse stories are about cloud computing. In Brooklyn, the Bathhouse spa repurposes the waste heat of Bitcoin miners that sit in a closed loop with its pools. “Your energy can never be created or destroyed, it can only change forms,” says Bathhouse co-founder Jason Goodman. “We’re mining with it. We’re capturing 99.9 percent of the heat, and then we’re using that to heat the pools,” he explains. The amount of electricity generated by the Bitcoin mining rigs is roughly the same as a conventional electrical pool heater, Goodman says.
Bathhouse has since expanded the heat capture process into a second site, heating its pools for users. Any excess heat energy created is also stored for future use, Goodman explains. “We’re able to store any excess energy in a thermal battery, and then we use it in our domestic hot water system for showers or whatever,” he says. “The technology is actually quite simple,” Goodman says. “The question is more of an infrastructure one.”
Using up heat as it’s generated is relatively simple. Storing any excess on site is possible. Moving heat over long distances is trickier—and is where the current data center rollout of massive, remote sites to power the generative AI revolution starts to become trickier for heat reuse.
The hard part isn’t recapturing heat, it’s locating compute near consistent heat demand such as homes, pools, hotels or offices. “Imagine the infrastructure where data centers, instead of being one massive data center, were distributed into every skyscraper or every big apartment building and office building in Manhattan,” says Goodman. “If everyone had their own small data center, or medium-sized data center, and we were recapturing all that heat.” The concept is “applicable anywhere where people use hot water,” says Beharrell.
That requires a rethink in how data centers are sized and sited, but could be beneficial for those building the infrastructure at great cost by offering a revenue stream to offset the high setup costs. “You need to have a heat source. You need to have an end user for the heat you’re producing,” says Goodman. “But if data centers were thinking about it this way, they could say: ‘Wow, our heat waste is worth $10 million a year.’”