The next frontier in the global artificial intelligence (AI) race and cloud computing may not be on Earth.<\/p>\n
Elon Musk\u2019s company SpaceX is acquiring his other artificial intelligence firm xAI, a move much to do with revenue but even more to do with sending data centres into space. <\/p>\n
Though experts say this may take at least decades to happen, such a situation could lead to the world\u2019s software supply chain becoming dependent on an extra-terrestrial US monopoly.<\/p>\n
This is also the same direction for cloud computing being sent into orbit, which could be just five years away. <\/p>\n
\u201cWhen it comes to Elon Musk, I always think to add an invisible zero after any of his predictions,\u201d said Jermaine Gutierrez, a research fellow at the European Space Policy Institute (ESPI), referring to Musk\u2019s predictions of space data centres popping up in two to three years<\/p>\n
But an ESPI\u2019s report calculated that a competitive data centre in terms of power would be \u201cat least 20 years away\u201d.<\/p>\n
Data centres are the physical facility that houses an organisation\u2019s computer systems and high volumes of data, which can be retrieved anywhere in the world. They require a lot of energy to run and also stay cool, and take up a lot of space on Earth.<\/p>\n
Another advantage is security. There is more security in space as data is not transferred up to space and back to Earth, said Javier Izquierdo, chief strategy officer at the telecommunications satellite operator, Hispasat <\/p>\n
Why we won\u2019t be sending data centres to space just yet<\/p>\n
The idea of the space data centre is that it could run on solar energy, making it less energy-intensive.<\/p>\n
Space is also much cooler than Earth, but the counterintuitive reality is that space is cold, but cooling data centres in orbit is far harder than on Earth.<\/p>\n
“There is no fluid to dissipate heat,” Gutierrez said. “You’re stuck with radiators, and you’re basically facing the Stefan-Boltzmann law,” he said, referring to the law that when temperature goes up a little, heat radiation goes up a lot.<\/p>\n
This would result in a massive thermal management infrastructure that dwarfs the computing hardware itself.<\/p>\n
Another reason it will take a long time for space data centres is that SpaceX\u2019s Starship would have to achieve full reusability and high flight cadence to reduce the launch costs to essentially the price of fuel. The other issue is that Starship hasn\u2019t even reached orbit yet. <\/p>\n
But anybody who’s looking into space data centres is looking at Starship, said Guiterrez, because their business model hinges on the success of Starship, including SpaceX.<\/p>\n
But as well as launch costs, there is also the feasibility of repairing the components on the data centres. <\/p>\n
In orbit, components typically just last five-years due to the radiation damage. Maintaining this infrastructure would require robots going to space, and they do not yet have the capabilities to perform such tasks. However, this could be something that Musk\u2019s company, Tesla, is working towards. <\/p>\n
Will the US control space computing?<\/p>\n
Gutierrez argues that while data centres in orbit are 20 years away, this is only 20 years away if we start working on them now and whoever controls this AI infrastructure could control the technology. <\/p>\n
OpenAI\u2019s CEO and co-founder, Sam Altman, said that AI computing costs will eventually approach the cost of energy itself. <\/p>\n
In orbit, solar energy is essentially free and constant. Whoever controls cheap space-based power generation could dominate AI services regardless of current economics. “If we let all of this infrastructure for space-based solar power to be dominated by the Americans, perhaps that’s the risk,” Guiterrez said.<\/p>\n
However, for Himanshu Tyagi, co-founder of open source AI company Sentient, controlling satellites doesn\u2019t mean winning the AI race: \u201cthe real risk isn\u2019t some sci-fi runaway superintelligence but rather who ends up holding the keys\u201d. <\/p>\n
He said running models on satellites was not fundamentally different from running them in factories, as edge AI will exist everywhere. What should worry people, he said, is the pile-up of power across the stack \u2013 compute, deployment, distribution, capital, and governance.<\/p>\n
\u201cWhen the same small group controls multiple choke points, like launch and comms, plus AI, plus robotics, plus consumer platforms, you get something closer to an oligarchy that\u2019s hard to regulate, compete with, or even meaningfully audit,\u201d Tyagi said.<\/p>\n
The United States is not the only country working on space data centres. China has deployed its Three Body constellation, with satellites actively working on edge computing capabilities and hosting AI payloads. <\/p>\n
Europe\u2019s place<\/p>\n
For Europe, despite expressing an interest in space data centres, there is no concrete plan. <\/p>\n
Just as the continent lacks competitive cloud infrastructure \u2014with services dominated by Amazon Web Services and Google\u2014it risks repeating this dependency in space.<\/p>\n
Under the US Cloud Act, American companies can be compelled to shut off services anywhere in the world, including Europe. <\/p>\n
EU regulators already understand that digital sovereignty requires owning and operating infrastructure, not relying on foreign providers.<\/p>\n
“Europe doesn’t have that same ‘brand and go for it’ attitude,” Guiterrez said, noting that the last time Europe had a clear space vision was with the Ariane 4 rocket, once the most competitive launch system globally.<\/p>\n
Meanwhile, while data centres in space may occur in two decades, applications such as edge computing on space stations could arrive within five years, depending on progress in launch costs and thermal engineering.<\/p>\n
\u201cEurope needs to take its future for its own data seriously and work on its own capabilities,\u201d said Izquierdo, adding that it is also vital for the continent\u2019s cybersecurity, as \u201cit is harder to hack in space\u201d.<\/p>\n