WASHINGTON, Aug. 27 (DC Times Online) — NASA Administrator Jared Isaacman says a future answer to the growing demand for artificial-intelligence computing could be to move some data centers off Earth.
In an interview with Nexstar DC, Isaacman said data centers could be placed in orbit and powered by sunlight. He argued that orbital facilities would not compete with ground-based data centers for land, water or electricity-grid capacity.
The idea remains a proposal, not an operating business. The supplied research does not establish that a full-scale orbital data center has launched or entered commercial service.
Why put computing in space?
Data centers are large facilities filled with computer servers. They process online services, business applications and artificial-intelligence systems. Those servers use substantial amounts of electricity and produce heat, creating demand for cooling systems and, in many locations, water.
An orbital data center would use solar panels to generate electricity. In space, the panels could receive sunlight without clouds, nighttime or atmospheric interference in the same way as panels on Earth. Radiators would release heat into space rather than relying on cooling towers or local water supplies.
Isaacman described the sun as a “free fusion reactor” and said orbital data centers could help the United States compete in both space and artificial intelligence. He also said the approach could move facilities away from communities that object to large data centers nearby.
Those advantages would come with a different set of demands. A space-based system would need reliable, low-cost launches; equipment that can withstand radiation; large heat radiators; and high-speed communications, including laser links between satellites and ground stations.
What companies are proposing
Google has described Project Suncatcher as a research effort involving artificial-intelligence chips on solar-powered satellites connected by lasers. Scientific American reported that Google planned a demonstration mission for 2027. The project is still a planned demonstration, not a commercial orbital data-center network.
Starcloud has demonstrated a smaller version of the concept. Scientific American reported that the company launched a 60-kilogram satellite carrying an NVIDIA H100 graphics-processing unit. A demonstration satellite is not the same as a full data center, which would require far more power, computing hardware and communications capacity.
The American Institute of Aeronautics and Astronautics’ Aerospace America has reported that SpaceX and other companies have proposed large satellite constellations for artificial-intelligence processing. The publication also reported that Chinese companies have announced on-orbit artificial-intelligence plans and that several U.S. companies are seeking regulatory approval for orbital data centers.
The Federal Communications Commission would play a role in reviewing relevant U.S. satellite applications. The supplied materials do not include the underlying filings or approval documents for those proposals.
Caltech and Sophia Space have also announced a patent for a space-based cooling system they say could support solar-powered data centers. The New York Times reported that the companies envisioned possible use by the early 2030s. The supplied materials do not include the patent document itself.
What stands in the way?
The largest obstacle may be cost. Every satellite, solar array, radiator and computer must be launched, placed in orbit and maintained. A University of Pennsylvania researcher, Wen, told Aerospace America that launch prices would need to fall to about $200 or $300 per kilogram for the concept to become more practical.
Power and heat are also major engineering problems. A paper posted on arXiv said producing 5 gigawatts in orbit would require solar arrays measuring 4 by 4 kilometers. Five gigawatts is roughly the output of several large power plants, illustrating the physical scale of a system designed for heavy artificial-intelligence workloads.
Solar panels generate electricity, but servers turn much of that electricity into heat. In space, heat cannot be carried away by air or water. It must be emitted as infrared radiation from large radiator panels, adding weight and complexity.
Communications could create another limit. Artificial-intelligence systems often move large amounts of data between storage and processors. Laser links could provide fast connections between satellites, but the system would still need reliable links through changing orbital positions and to users on Earth.
Space operations also bring safety and environmental questions. The arXiv paper warned that large satellite systems could obstruct views of the night sky and increase collision and debris risks. A collision could damage other spacecraft and create fragments that threaten additional satellites.
The proposal would therefore trade some earthly constraints for space-related ones. It could reduce pressure on local land, water supplies and electric grids, but only if launch, hardware, communications, cooling and orbital-safety problems can be managed at a cost that makes the system competitive.
How close is the technology?
Small demonstrations and planned missions show that computers can operate in orbit. They do not yet show that large, commercially useful data centers can be built there.
Google’s planned demonstration, Starcloud’s satellite and proposed large constellations represent different stages of development. None, based on the supplied materials, establishes that orbital computing is ready to replace terrestrial data centers.
Isaacman’s comments put the idea on NASA’s policy horizon, but turning it into infrastructure would require action and evidence from technology companies, launch providers, regulators and researchers. For now, orbital data centers are best understood as an emerging technology proposal aimed at one of the central problems of the artificial-intelligence boom: how to supply enormous computing power without overwhelming communities and infrastructure on Earth.
