As communities push back against the enormous electricity and water demands of AI data centers, some of the world's biggest tech companies are pursuing an unusual solution: moving computing hundreds of miles above Earth.
Google has begun testing AI technology in orbit. SpaceX is developing plans for much larger satellites that could eventually form a network of space-based data centers.
The potential payoff is significant. But so are the technical challenges and environmental risks, including new sources of space debris and light pollution.
In October, Google launched a refrigerator-sized satellite carrying AI chips as part of Project Suncatcher, an effort to develop solar-powered computing in orbit.
The satellite successfully generated and transmitted a haiku back to Earth:
Blue marble belowData streams in silent graceStarlight feeds my core
"Turns out, it is possible," said Travis Beals, Google's senior director leading Project Suncatcher, in a company video.
The experiment demonstrated that AI processors can operate in space. The next challenge is connecting multiple satellites to perform more demanding computing tasks. Google plans to test laser communication between two satellites in 2027.
The timing is significant. Global data center electricity demand is projected to roughly double by 2030, as communities across the country raise concerns about the industry's growing environmental footprint.
According to Google's estimates, solar panels in certain orbits could generate five to eight times as much electricity as comparable panels on Earth. Space-based computing could also reduce the need for some water-intensive cooling systems.
SpaceX CEO Elon Musk sees similar potential.
"That avoids the need to build massive power plants on Earth and deal with cooling because cooling is actually much easier in space than it is on Earth," Musk said in a company presentation.
SpaceX is developing AI computing satellites with large solar arrays, radiators and laser communication links. The company has proposed a constellation that could eventually include as many as 1 million satellites, although that system has not been approved or deployed.
Despite the appeal of abundant solar power, putting computers in space doesn't eliminate the heat they produce.
"The heat is still generated from inside the circuits," said Dr. Arun Kannawadi, an assistant research professor of physics at Duke University. "You still need an effective way of like transferring that heat from inside the circuits to the exterior wall, and then it dissipates."
Without air to carry heat away, satellites need specialized cooling systems and radiators. Repairs could also require additional launches, while solar radiation can damage sensitive electronics.
"Even if you just put this in space, it's not going to be as efficient as you think it might be, just because the space is cold," Kannawadi said.
Kannawadi, an observational astronomer, studies the universe using telescope images that are already affected by sunlight reflecting off satellites.
"They are kind of photobombing the images that we are trying to take," he said.
An October 2026 research paper modeled the impact of a hypothetical constellation approaching 1 million orbital data center satellites on the Vera C. Rubin Observatory in Chile. Researchers projected substantial interference with astronomical observations, including dramatic increases in sky brightness under certain conditions.
Those are modeled scenarios, not effects observed from an existing space-based data center network.
Kannawadi also raised concerns about potential effects on wildlife if satellites significantly brighten the night sky, as well as the danger of collisions creating additional orbital debris.
"You end up having a lot more debris than you would have just from a smaller satellite," he said, warning that collisions could trigger a chain reaction.
The full environmental impact would also depend on the emissions and resources associated with manufacturing, launching, maintaining and eventually replacing the satellites.
Even if orbital computing becomes practical, it won't address the immediate demand for electricity from North Carolina's growing data center industry.
Duke Energy projects data centers could account for about 10% of its electricity sales across the Carolinas by 2030. The utility is planning billions of dollars in new power plants and infrastructure, including natural gas generation, to meet growing demand.
Those plans have raised questions about costs for residential customers and the environmental consequences of building additional fossil fuel infrastructure.
For now, the technology remains experimental. Google's next test is planned for 2027, while SpaceX is pursuing larger-scale satellite manufacturing.
The early experiments demonstrate that AI computing can work in orbit. Whether it could eventually reduce the industry's overall environmental footprint, rather than simply move some of its impacts elsewhere, remains an open question.