By Micah Maidenberg and Becky Peterson

Elon Musk and Jeff Bezos have raced for years to build rockets and launch satellites. Now they're racing to take the trillion-dollar data-center boom into orbit.

Bezos' Blue Origin has had a team working for more than a year on technology needed for orbital AI data centers, a person familiar with the matter said. Musk's SpaceX plans to use an upgraded version of its Starlink satellites to host AI computing payloads, pitching the technology as part of a share sale that could value the company at $800 billion, according to people involved in the discussions.

Deploying satellites that provide significant AI computing capability will present difficult engineering hurdles and pose tough questions about the price of deploying swarms of the devices into orbit.

Advocates acknowledge the challenges of making these systems work, including doing so in a manner that would match the performance of cavernous data centers stuffed with AI chips on the ground. Skeptics believe the technical risks are being underestimated and say space-based data centers won't be competitive on cost, especially if power and other constraints ease on the ground.

Nonetheless, the idea has seized the imaginations of many leaders working on AI and space technologies. SpaceX didn't immediately respond to a request for comment. A spokesman for Blue Origin declined to comment.

Deploying satellites as data centers, the thinking goes, would allow the AI industry to avoid Earthly headaches, such as securing the immense amounts of power needed to train AI models.

Proponents imagine potentially filling orbits with satellites laden with chips that handle the computations underpinning AI applications used by consumers and companies. Zipping through space, the satellites would tap the immense power of the sun to operate and beam data back to Earth.

"Taking resource-intensive infrastructure off Earth has been an idea for years, but it has required launch and satellite costs to come down. We are nearing that point," said Will Marshall, chief executive of satellite operator and builder Planet Labs.

In early 2027, Google and Planet Labs aim to deploy two test satellites into orbit carrying the tech giant's AI chips, called tensor processing units. Google has described the project as one of its moonshots, given the obstacles of deploying a network of satellite data centers at scale.

One challenge is the number of satellites that may be needed.

Travis Beals, a Google executive working on the orbital data-center effort, said it would take 10,000 satellites to recreate the compute capacity of a gigawatt data center, assuming 100-kilowatt satellites.

The test mission in 2027 is about demonstrating the key elements of operating satellites as AI-computing clusters, according to Beals. "Then we have a long, hard road in terms of all the optimization, all the various new technologies we need to scale up and then do so in a cost-effective way," he said in an interview.

A throng of companies and executives are trying to figure out the viability of orbital data centers, in addition to SpaceX, Blue Origin and Google.

In October, Bezos said during an event in Italy that shifting data centers to orbit made sense, given the solar power available in space. It will take time for those to beat the costs of terrestrial AI infrastructure, but he predicted that would happen in 20 years or sooner.

Bezos' Blue Origin has made major strides this year to demonstrating its years-in-the-making rocket, New Glenn. That vehicle is partially reusable and has a large fairing designed to ferry significant numbers of satellites to orbit.

Sam Altman, chief executive of OpenAI, has investigated whether his company could take over a rocket operator, using the vehicles to deploy AI computing in space, The Wall Street Journal reported. Eric Schmidt, the former Google chief executive who took over Relativity Space, a company working on its own rockets, has talked about orbital data centers.

IBM's Red Hat software business and Houston-based Axiom Space had a data-computing prototype launched in August. Aetherflux, Starcloud and other venture-funded startups are setting their own plans to compete against larger players.

Operating satellites as data centers will pose a host of technical issues, including managing temperatures for AI chips in orbit, protecting them from radiation and transferring data back to the planet without long lag times.

"There's a bunch of engineering challenges, but I think those engineering challenges are all solvable," said Jonny Dyer, chief executive of Muon Space, a satellite company that was involved in Google's orbital data-centers research paper. "It ultimately comes back to launch."

The prospect of having potentially thousands of satellites-as-data centers to launch could drive business across the aerospace supply chain, including rocket companies. Developing rockets is expensive and difficult, but frequent launches would allow operators to offset costs and boost margins, industry executives say.

Musk's SpaceX has pursued that path by launching its partially reusable Falcon 9 fleet at record rates, sending up internet satellites for its Starlink business and carrying payloads for outside customers.

Texas-based SpaceX is aiming at even cheaper costs with Starship, the massive rocket it has been developing. AI-compute technology would be installed on upgraded satellites that SpaceX designed specifically to fit on the Starship spacecraft, according to people familiar with the matter.

The rocket isn't operational yet, and SpaceX is planning to debut an upgraded version of it early next year.

"Starship should be able to deliver around 300 GW per year of solar-powered AI satellites to orbit, maybe 500 GW," Musk said in a post on X last month.

Other AI companies that want to take aim at orbital data centers will need to figure out how they would deploy satellites into space.

Write to Micah Maidenberg at micah.maidenberg@wsj.com and Becky Peterson at becky.peterson@wsj.com

(END) Dow Jones Newswires

December 10, 2025 12:43 ET (17:43 GMT)

Copyright (c) 2025 Dow Jones & Company, Inc.