Data centers were once seen as key contributors to economic development. Nowadays, many Americans aren’t keen on having them nearby. So, is space the solution?
Data centers expanded largely across the Southern and Midwestern United States as local governments presented them as a means to stimulate economies. However, growing opposition from communities has prompted legislators to consider moratoriums and stricter regulations on new data center projects. This backlash is compelling some AI firms to explore options in outer space.
Yet, transitioning to orbital data centers isn’t straightforward, experts pointed out. These space-based centers don’t match the electrical output of their terrestrial counterparts and face challenges like radiation protection for their equipment.
According to Kristian Stout, who heads innovation policy at the International Center for Law & Economics, “We are going to need more data centers in more places.” He also mentioned that AI laboratories might feel increased pressure to investigate orbital options due to the resistance they face on Earth.
In August, Elon Musk, CEO of SpaceX, stated that Nvidia, a leading AI chip manufacturer, is working on a “space-optimized” AI chip intended for orbit, expected to launch by late 2027 with broader capabilities following in 2028. “Our design is significantly simpler, lower cost, denser and lighter than a traditional rack,” he remarked, also revealing SpaceX’s plans to integrate Nvidia’s technology into their orbital data centers.
SpaceX has filed a request with the Federal Communications Commission (FCC) to deploy as many as one million satellites for AI purposes, with the outcome still pending.
Other players are also venturing into this space arena. For instance, Cowboy Space has proposed a constellation of 20,000 satellites branded the “Stampede Data Center System.” Meanwhile, another startup, Starcloud, is seeking approval for up to 80,000 orbital data center satellites. Amazon’s Blue Origin aims for a network of around 51,600 satellites, and the startup Orbital has filed for authorization for 100,000 data center satellites.
The FCC is optimistic about a new era of American leadership in space, with a representative stating that the Trump administration is promoting a renaissance of innovation, facilitated by swift actions from the Commission to bolster the U.S. as the prime location for launching and managing future orbital infrastructures.
In a notable move, Google announced its Project Suncatcher in November 2025, with plans to send solar-powered satellites into space, starting with two prototypes expected in early 2027.
China is also advancing in this field, having launched 12 of its planned 2,800 satellites for an orbital network in 2025. Other Chinese companies are organizing similar launches, indicating a competitive landscape.
A recent report from the European Space Policy Institute (ESPI) cautioned that Europe risks falling behind China as its efforts to strategize launching its own orbital data centers continue to stagnate.
“While European institutions continue to debate the future of computing in orbit, China is already building it,” the report noted.
How Would Orbital Data Centers Function?
While orbital data centers could provide unique flexibility, Benjamin Lee, an electrical and systems engineering professor at the University of Pennsylvania, claims their power output would likely be substantially lower than traditional centers. He explained that current capabilities might only reach low kilowatts, while ground-based centers operate in the 200 megawatt range. “It’s very speculative right now,” he added, emphasizing that although this concept works, it isn’t yet viable at the same scale.
Lee highlighted that these centers might be more equipped to answer user queries via AI chatbots rather than performing more complex tasks, like coding or executing projects on behalf of users.
However, challenges abound for orbital centers. They must be designed to mitigate radiation damage from space, which can complicate cooling systems since heat is not easily dissipated in the vacuum of space. “They need specialized radiators to get rid of heat effectively,” Stout explained.
Although there has been collaboration, especially from Google, in testing how certain chips handle radiation exposure, the question remains about how to ensure data integrity in the face of radiation-induced errors, with Lee urging for rigorous testing through actual space operations.
Stout expressed reservations about the feasibility of maintaining and repairing these satellites, given their typical lifespan of five to seven years before they generally fail or deorbit. He raised a critical question: “Would you throw away a billion dollars’ worth of GPUs into the atmosphere every seven years?” He further noted that sending AI chips back to Earth could be prohibitively costly.
While orbital data centers could assist AI companies, Stout contended they wouldn’t reach the same scale as ground-based facilities. “There’s no clear proof of hyperscale capability in the near term, so what you’re talking about is edge computing or leveraging existing satellite technology for improved imaging,” he said.
