The biggest question in the tech sector is, if AI is the future, where will all the infrastructure go?
Some space industry leaders – including Elon Musk and Jeff Bezos – believe the answer may lie through space-based data centers in orbit.
That idea has gained attention in recent weeks. Musk is reportedly merger consideration Between SpaceX and its generic artificial AI company XAI ahead of their space firm’s planned IPO later this year. Moreover, in A filed With the FCC in late January 30, SpaceX proposed an orbital data center constellation of 1 million satellites in low Earth orbit.
But industry analysts say that while the technical challenges may be surmountable, it is not yet clear whether the business case for data centers in space will hold.
Deep learning models built by leading labs like OpenAI, Anthropic, and XAI rely on huge clusters of advanced processors operating in data centers. As more deep learning models are created and deployed, more processing power is required – last year companies invested $61 billion in building more data centers. This is a new record, with real estate trackers expecting capacity to double by 2030, according to S&P Global.
However, new data centers on Earth face hurdles, including environmental permitting, access to electricity and water for cooling, and community opposition. For example, Microsoft expects water use in its data centers to triple by 2030. new York Times. Data center projects built by Amazon and Chile have been canceled due to resource concerns.
“The biggest problems are cooling, safety, power transmission – all these things can be solved if you get it to space,” said space industry analyst Chris Quilty. space news. “The big question that has to be solved is, can you do it economically?”
Jeff Bezos, a longtime advocate of moving polluting industries off the planet as CEO of Blue Origin, sparked the discussion with comments at last year’s International Astronautical Congress.
“We’re going to start building these massive gigawatt data centers in space,” Bezos said. “We will be able to beat the cost of terrestrial data centers in space over the next few decades.”
Elon Musk and SpaceX are also on board. Musk told the audience at the World Economic Forum that it would be cheaper to build data centers in space within three years — though that forecast depends on achieving full reusability for Starship, which Musk hopes to do in 2026.
Meanwhile, start-ups like StarCloud are attempting to build companies around this business, while Google is cooperating With Earth observation firm Planet to launch its own data center pilot in 2027. Chinese companies such as ADA Space and Beijing Astro-Future are planning ambitious space computing architectures.
“The speed at which this trend has taken hold has shocked me, even for an old guy,” Quilty said. “It’s exactly what happened with direct-to-device – everyone and their brother agreed it was a stupid idea until AST came up with the concept.”
Still, much uncertainty remains. “Whether it makes sense to put a data center in space, and what technologies are needed, depends on the business model,” said Akhil Rao, a former NASA economist and managing partner of frontier technology consultancy Rational Futures. space news Through email. “I have not seen a business model that is so clearly expressed that one can understand how well a system architecture serves particular use cases.”
Andrew McClip, head of research and development at Varda Space Industries, created a online calculator Which attempts to compare terrestrial and orbital data centers using publicly available data. Right now, it’s not clear, orbital data centers cost about three times more per watt of computing power under his base case.
McClipp wrote in his analysis of the calculator, “If you run the numbers honestly, the physics doesn’t immediately rule it out, but the economics is brutal.” “It only reaches considerable distances under aggressive assumptions, and the list of organizations deployed to try is basically the same.”
That’s why space data centers have played a central role in SpaceX’s IPO narrative, reportedly planned for June 2026. It’s difficult to justify a $1.5 trillion valuation based on the company’s current business lines and revenues, Quilty said, but the company has a record of surprising skeptics. When Starlink was proposed, few people thought that phased array antennas could be manufactured at the scale and cost required, but now SpaceX is making them by the millions and selling them for a few hundred dollars a piece.
Blue Origin’s latest announcement may point toward a middle path for AI in space: On January 21, the company announced TerraWave, a new satellite constellation that appears to be designed to provide new, high-bandwidth connections between terrestrial data centers. Compared to orbiting models, offering connectivity is a proven way for space companies to dip into the AI boom.
And of course, there is the question of the AI bubble. Analysts have noted that justifying investment in data centers would require a significant increase in revenue – $650 billion in annual consumer spending, according to a JPMorgan analysis – for companies like OpenAI, which currently lose tens of billions of dollars annually. AI boosters are confident that they will prove the value of their software, but users, especially at the enterprise level, are still figuring out how to generate returns using these tools. If they don’t meet expectations, the demand for more processing power may prove elusive.
