Do space-based AI data centers make economic sense?

Following the stellar success of the SpaceX IPO – raising $85.7 billion, valuing a new public company in the billions, and making Elon Musk the world’s first billionaire – what many skeptics still consider a pie-in-the-sky idea, building AI data centers in space, is emerging. There’s good reason to be skeptical, but the idea has moved at least in a plausible direction thanks to SpaceX’s windfall.

SpaceX has Falcon rockets that are reliable, reusable – and very powerful on wings – while its xAI has an insatiable need for computing power and its space-based Internet service, Starlink, has satellites that can be upgraded. Now engineering and related business technologies have billions of new capital needed to bring those components together in space, not only to feed SpaceX’s massive internal AI operations but also to provide commercial services for dozens of paying customers like Anthropic.

Some investors argue that the company has no choice but to make the idea work if it hopes to justify its public market valuation over time. “The company is getting into the data centers space,” Duncan Davidson, a partner at Bullpen Capital, told CNBC’s “The Exchange” a week before the IPO. “That’s a big, long-term game.”

The engineering and technical issues are being resolved, said Davidson, whose company is not a SpaceX investor but has an indirect interest in the space startup Starcloud. Although he added, “economically, at the moment, there is very little.”

Considering also, the ever-increasing obstacles to terrestrial data centers – practical, political and social – the prospects of opening them in earth’s orbit, where the sun shines 24/7, is no longer the stuff of science fiction.

If, as Musk has said, SpaceX’s Starship heavy rocket goes live next year — a definite “if,” given his history of underperforming previously promised programs — it will significantly lower launch costs, a key barrier to acquisitions. Meanwhile, the cost of building Earth-based data centers is likely to rise, while “space will start to get cheaper and cheaper,” Davidson said. “So I think so [business] The case is very strong in these things,” he said.

In January, SpaceX filed an application with the Federal Communications Commission for a constellation of up to a million satellites that will form the basis of an orbital AI data center. Two months later, at an event in Austin, Texas, Musk repeated previous claims that space-based, solar-powered data centers will cost more than Earth-based data centers in two to three years. “Energy expansion on Earth gets harder over time and more expensive over time,” he said, “but in space it gets cheaper and easier over time.”

The so-called AI1 satellites will be upgraded versions of those used for the existing Starlink communications network and will require additional semiconductors. The scale required is so great that SpaceX, Tesla and Intel have partnered to build Terafab, a 10-square-foot factory in Austin that is expected to open in 2029 and cost up to $119 billion to build.

SpaceX declined to elaborate on its plans, providing previously released information on its orbital data center concept and Terafab.

Jeff Bezos, Alphabet are also in the race

SpaceX is not alone in what has become a race to calculate in space. Amazon CEO Jeff Bezos has expressed similar ambitions for his rocket and AI, Blue Origin and Prometheus, respectively. Last month, in a CNBC interview, Bezos said building data centers in space is “very realistic,” though he was asked how long it would take. “Some of the times we hear are very short. People will talk two or three years,” he said, likely referring to Musk’s bold prediction. “That’s probably a little wishful thinking.”

In March, Blue Origin submitted plans to the FCC to launch 51,600 data center satellites into low Earth orbit as part of its Project Sunrise program. The deployment of the proposed constellation of satellites, called TeraWave, is expected to begin in the fourth quarter of 2027, the company said.

Alphabet search giant Google has entered the race with earth-observing satellite maker Planet Labs on Project Suncatcher, an orbital data center program, with SpaceX (which owns 6.1%) as a potential launch partner. The project, Google said, will explore how an interconnected network of solar-powered satellites, equipped with its Tensor Processing Unit AI chips, can harness the sun’s full potential.

A paper describing Suncatcher notes how historically high start-up costs have hampered large-scale space-based systems, but suggests prices could drop to less than $200/kilogram by the mid-2030s. At that price, operational orbital data centers can be roughly comparable to the reported energy costs of an equivalent terrestrial data center on a kilowatt/year basis.

Aside from that paper, “We have nothing new to share,” a Google spokesperson said in response to a request for comment.

Satellite, rocket and robot prototypes are tested

Aside from the universe of trillion-dollar-plus tech stocks, a few startups are also aiming for the sky.

Starcloud has already sent an Nvidia H100 GPU into space on a test satellite aboard a SpaceX Falcon 9 rocket. “It’s just going to be cheaper to put them in space,” CEO Will Marshall told CNBC in a recent interview. Another benefit, he said, is not competing for water and electricity in the communities. It’s a long-term project, Marshall said, “but an exciting one, too.”

Starcloud is also partnering with Rendezvous Robotics, a developer of modular spacecraft systems that assemble themselves in space, to power its orbital data centers. The spacecraft is made up of hundreds of interlocking, four-sided tiles, each about five feet wide, that are stacked on a launch rocket.

“Our tiles have been tested three times,” said Rendezvous president Joe Landon, “once on the Blue Origin New Shepard flight and twice on the International Space Station.”

More experiments on the ISS are planned for later this year. “By 2028, we will be able to deliver complete plans,” he said.

Rocket LabThe Electron rocket has launched nearly 90 of its satellites into space for NASA, the US Space Force and many international customers. Founded in 2006 by engineer Peter Beck, Rocket Lab is building a powerful reusable rocket, the Neutron, that will give the vertically integrated company the ability to compete with SpaceX – if on a much smaller scale – in the orbital data center market.

“If this turns out to be a big market, we’ll be in a good position to attack it as a vendor supplier or with our own application or a combination of those things,” said CFO Adam Spice. If the push is to try, “We would rather turn customers into tenants of our own infrastructure than help them build their own,” he said.

Cowboy Space, founded in 2024 by Robinhood founder Baiju Bhatt under the original name of Aetherflux, has an end-to-end strategy as well, but still develops rockets and data center infrastructure in-house. Its novel approach “involves using the second stage of the rocket as the data center satellite itself,” Bhatt said. “There’s a lot more we’re going to reveal and show you in the not-too-distant future,” he said, adding that the company is aiming for its first launch into space later this year. Currently, it has applied to the FCC for a 20,000 satellite constellation.

Even with all the research, development, and continued investment in orbital AI data centers in space as the new commercial space economy grows rapidly, the basic economic question has not changed, according to Mark Weinzierl, a Harvard economist specializing in tax policy who became interested in space-based businesses a decade ago and now writes and teaches on this topic. “One of the big questions is, are you sure we can’t do that cheaply on Earth?” he said.

“I’ve never seen someone talk like this [they’re] cost competitiveness,” he said of his analysis of current business models in the space sector. However, Weinzierl believes it is reasonable to predict a future where “the cost of doing it on Earth will go up over time and the cost of doing it in space will go down over time. And at some point those two curves will meet.”

But that requires thinking about future economies based on current trends.

“Technology can always change,” he said. “Maybe our next generation of chips won’t be as power-hungry as the current generation,” which could help lower the cost of terrestrial data centers.

There are regulatory, environmental and political obstacles on Earth, which Weinzierl said currently make a space-based case work better, at least in theory. Indeed, the growing backlash against the aggressive construction of data centers has resulted in more than 100 proposed freezes at local, state, regional and national levels. Society is backing down, too. A Heatmap News poll conducted in May found that seven in 10 Americans would oppose a data center being built near where they live, up from four in 10 last August. In addition, a recently released study from First Street, a climate risk analysis firm, finds that 79% of the data center’s capacity faces a higher risk of climate risks.

In Weinzierl’s view, it’s still more of a gamble than a certainty, but not necessarily out of the ordinary.

“If you’re hoping for the cost reductions we’ve seen in startups, satellite and solar technology [power] keep happening, and those [curve] the lines will soon cross. It’s always a gamble, but that seems like a good deal to me,” he said.