<p>For years, data centers supporting artificial intelligence have been established on Earth, consuming significant amounts of electricity and water, and facing opposition from local communities.</p>
<p>Now, major technology companies, including Google and SpaceX, along with various startups, are investigating the potential for AI infrastructure to operate in space, utilizing solar power and avoiding many of the challenges associated with terrestrial data centers.</p>
<p>This concept aims to address issues such as abundant sunlight, the elimination of cooling water requirements, and local disputes over land and resources.</p>
<p>While it is feasible for computers to function in space, challenges arise in building, launching, powering, cooling, connecting, maintaining, and replacing sufficient orbital data centers to significantly impact the existing infrastructure on Earth.</p>
<p>“The possibility is of course there,” said Amit Verma, an electrical engineering professor at Texas A&M University-Kingsville. “The feasibility and sustainability — those are engineering questions that need to be addressed before this becomes a very viable option.”</p>
<p>Google's Project Suncatcher envisions a network of solar-powered satellites equipped with Tensor Processing Units (TPUs) to conduct machine-learning tasks in orbit. The company is preparing a prototype mission, with its first test satellites expected to launch next week.</p>
<p>SpaceX CEO Elon Musk has proposed a more ambitious plan involving up to one million satellites serving as orbital data centers. Additionally, startup Starcloud has sought permission from the Federal Communications Commission to deploy as many as 88,000 satellites for distributed data center purposes, while Blue Origin has its own plans for a large orbital computing constellation.</p>
<p>The federal government is also considering this concept. A Government Accountability Office report published in April identified space-based data centers as a potential solution to reduce the resource consumption of terrestrial data centers. Senators Ted Cruz (R-TX) and John Hickenlooper (D-CO) introduced legislation in June to establish a Pentagon pilot program for testing commercial orbital data centers for national security applications.</p>
<p>This interest comes amid public resistance to new data centers being established in local areas. A September Pew Research Center survey indicated that 60% of U.S. adults would be uncomfortable with a new data center operating nearby.</p>
<p>While space could alleviate some local pressures, it introduces new challenges.</p>
<h2>Challenges of Cooling in Space</h2>
<p>One significant challenge is heat management. On Earth, data centers can utilize water-based cooling systems to dissipate heat. In space, however, there is no atmosphere to conduct or convect heat away, necessitating that spacecraft radiate heat into space.</p>
<p>“A data center capable of running AI models could require tens of megawatts of power,” Verma noted.</p>
<p>As electricity consumption increases, so does the waste heat generated, requiring larger systems to manage heat dissipation.</p>
<p>Ben Lee, a professor at the University of Pennsylvania specializing in computer architecture and sustainable computing, stated that while the fundamental technologies for powering a spacecraft are well understood, integrating them at the scale needed for an AI data center presents challenges.</p>
<p>“The techniques that we’ve got for air cooling or liquid cooling data centers through cooling towers and evaporative water loss obviously won’t work in a vacuum,” Lee explained. “So we need a strategy for cooling the data centers.”</p>
<p>Radiation is another concern. Computer chips designed for Earth-based data centers are not equipped to endure the harsh radiation of space.</p>
<p>“We can’t just take something off the shelf that we’ve designed for terrestrial data centers, launch it into space, and expect it to perform in the same way,” Lee added.</p>
<p>While companies can shield hardware or develop radiation-resistant components, these solutions add complexity, weight, or cost.</p>
<p>Additionally, keeping the technology up to date poses a challenge. AI hardware evolves rapidly, complicating the logistics of maintaining orbital infrastructure.</p>
<p>“The technology changes every couple of years,” Verma said. “The GPUs that you have there…will become obsolete in two [or] three years. They need to be replaced.”</p>
<p>In a terrestrial data center, replacing outdated GPUs involves sending a truck to the facility, whereas in space, it requires launching new infrastructure.</p>
<h2>The Launch Problem</h2>
<p>The economics of orbital data centers hinge on the cost of transporting large amounts of hardware into orbit and maintaining it there.</p>
<p>The GAO report indicated that large-scale deployment would necessitate advancements in computing, cooling, communications, manufacturing, and launch capacity. While some companies envision constellations of thousands of satellites, projects designed to process data generated in space are closer to technological maturity than massive orbital facilities for training AI models.</p>
<p>Karen Howard, director of science and technology assessment at GAO, stated that the current launch industry is not prepared for the scale required for the largest proposals.</p>
<p>The FAA projects approximately 200 launches in 2026, while some orbital data center concepts anticipate tens of thousands or even millions of satellites.</p>
<p>“Even if industry can develop data centers (small or large) for space, it doesn’t have the manufacturing and launch capacity, and related ability to reduce the cost of launch, that would be necessary to meaningfully replace or even significantly supplement Earth-based data centers yet,” Howard said.</p>
<p>Space-based computing does not need to outperform Earth-based data centers on every metric to have commercial viability; it could be beneficial for workloads originating in space or requiring specialized infrastructure.</p>
<h2>Potential for Smaller Data Centers</h2>
<p>Initially, space data centers may not resemble traditional data centers. Satellites currently collect vast amounts of imagery and information, much of which must be sent back to Earth for processing.</p>
<p>Howard suggested that processing some of this data in orbit could be more immediately beneficial than attempting to train large AI models in space.</p>
<p>NASA has estimated that two of its Earth-observation missions generate about 100 terabytes of data daily. Processing this information on a satellite before transmitting it to Earth could reduce the volume of data sent, alleviating pressure on communication networks.</p>
<p>The military is exploring similar concepts, with the Space Development Agency developing satellites capable of processing raw sensor data onboard and generating missile tracks for ground transmission.</p>
<p>Lee proposed that instead of creating a large orbital version of a data center, companies could integrate smaller computing capabilities into existing satellites.</p>
<p>“Maybe Starlink satellites are already handling a lot of compute communication. Maybe add a little bit of compute to those Starlink satellites,” Lee suggested.</p>
<h2>Environmental Considerations</h2>
<p>Space data centers are being promoted as a potential environmental solution to terrestrial data centers.</p>
<p>However, environmental groups caution that this shift could lead to new environmental challenges. A coalition represented by Earthjustice petitioned the FCC in July for a programmatic environmental review before approving orbital data center constellations, citing concerns over proposals involving over one million satellites and urging the federal government to assess their environmental impacts.</p>
<p>Increased objects in low Earth orbit raise the risk of collisions, which can generate debris that complicates navigation for other spacecraft.</p>
<p>Howard noted that Starlink satellites already conduct collision-avoidance maneuvers, averaging around 40 per satellite annually. The introduction of thousands more data center satellites could complicate orbital traffic management.</p>
<p>Moreover, the question of what happens to non-functional data centers poses additional concerns. Many proposed systems anticipate relatively short-lived hardware that requires regular replacement. While satellites in low Earth orbit can be deorbited, this process raises its own environmental and debris-related issues.</p>
<p>There are also broader questions regarding data control in orbit, governance of these systems, and the implications of commercial infrastructure becoming critical for communication and computing.</p>
<p>In summary, relocating data centers does not eliminate the infrastructure; rather, it alters the nature of the infrastructure and the regulatory questions associated with it.</p>
<h2>A Different Kind of Data Center Race</h2>
<p>The most plausible future may not involve a binary choice between Earth and space for large data centers.</p>
<p>Lee expressed caution, indicating that the technological challenges are substantial enough that researchers should focus on fundamental research rather than immediate commercial deployment.</p>
<p>“I don’t think it’s going to be in the next year or two,” Lee stated. “I think 10 to 20 years, I think, is realistic.”</p>
<p>This timeline allows for technological advancements, providing room for improvements in terrestrial energy, nuclear power, geothermal energy, launch vehicles, and AI chips.</p>
<p>“If it takes 10, 20 years to figure out how to launch data centers into space,” Lee concluded, “small modular reactors or geothermal or other things might make a huge amount of progress during that time frame.”</p>