Data centers are a growing climate problem, especially as all the energy needed to train AI models inflates tech companies’ carbon footprints. A startup that spun out of Google’s “moonshot factory” X says its technology can provide a potential solution. It can pull carbon dioxide out of the air, run partially on servers’ waste heat, and even generate water to help cool a data center more efficiently.

The company, 280 Earth, just signed a $40 million agreement to capture carbon dioxide emissions for some big names through an initiative called Frontier that Stripe, Alphabet, Meta, Shopify, and McKinsey launched in 2022 to support emerging carbon removal tech. The buyers include Frontier’s founding members as well as Autodesk, H&M Group, JPMorgan Chase, Workday, and other brands.

Filtering CO2 out of the air is trending among companies trying to hit sustainability targets but still struggling to reduce their carbon dioxide emissions by turning to clean energy. Frontier has also brokered deals between big tech companies and other startups working to capture carbon using rocks, liquid smoke, and even sewage.

“When we started 280 Earth as a moonshot at X, our vision was always to find a radically effective, affordable and scalable way to remove billions of tons of carbon from our atmosphere. We’re excited to see this momentum with Frontier buyers,” Astro Teller, captain of moonshots at Alphabet’s X, said in an emailed statement.

280 Earth builds modules containing absorbent materials — called sorbents — to filter CO2 out of the air. It’s similar to large industrial facilities called direct air capture (DAC) plants that other companies have started to build in recent years. But 280 Earth says it can run more efficiently, allowing it to work in tandem with data centers and draw some power from their waste heat.

Other DAC technologies work in big batches, waiting until the sorbent is fully saturated before heating the filter to high temperatures (upwards of 100 degrees Celsius or 212 degrees Fahrenheit) to release and collect the CO2 that it trapped. 280 Earth says it saves energy by not working in batches. Instead, it continuously moves the sorbent between two chambers to avoid the energy loss that comes with reheating, cooling, and changing the pressure in a single chamber.

It’s sort of like the difference between a home oven and a professional pizza oven, says 280 Earth CEO John Pimentel. You lose heat and energy each time you turn a home oven on and off and open its door. The pizza oven, on the other hand, retains heat better and stays at the same temperature.

Not only does 280 Earth’s technology capture CO2, but it also happens to pull in water vapor that it can extract to provide water for a customer. It can gather two to four tons of water for every ton of CO2 captured, according to Pimentel.

That’s all supposed to create a win-win situation for 280 Earth and a data center, or any other industrial facility that needs water and creates waste heat. One critique of DAC technology is how much energy it might need to heat up its filters. 280 Earth’s sorbents can do their job at low enough temperatures to utilize that industrial waste heat. Data centers, meanwhile, can burn through a lot of electricity and water to run servers and keep them from overheating. 280 Earth can potentially help on that end by drawing in waste heat and producing water for cooling systems in return.

The biggest impact, though, would be to get rid of the carbon emissions causing climate change. Google’s carbon emissions have grown by nearly 50 percent since 2019, for instance, thanks in large part to new and expanded data centers used to train AI. To that end, some environmental groups are worried that relying on new strategies to capture CO2 pollution after it’s already been released is risky. After all, these are new technologies that have yet to be proven at scale. They’re also prohibitively expensive at the moment.

280 Earth just finished building its pilot facility in Oregon in May, which will be used to capture 61,600 tons of CO2 by 2030 as part of the $40 million deal. The cost breaks down to more than $600 per ton of captured CO2. The company is still finalizing plans for where to sequester the CO2 after it’s captured, likely sending most of it by truck and rail to underground storage wells somewhere in the US.

For context, Google was responsible for 14.3 million metric tons of carbon dioxide pollution in 2023 alone. That shows how small of a dent this initial carbon removal deal would make in its overall carbon footprint and how expensive it would be to pay to capture a larger chunk of its climate pollution.

Scaling up carbon removal would bring costs down, Pimentel says. And he believes it buys time for companies as they transition to cleaner energy. “As much as we all want to believe that transition off of fossil fuels is going to happen quickly ... It’s going to take decades,” Pimentel says. “In the meantime, we’re continuing to pump more CO2 into the atmosphere, exacerbating the problems that we already have. So I’m a believer in the all of the above solution.”