It’s unwise to take tech billionaires’ predictions about the future at face value. Despite Mark Zuckerberg’s promises, very few of us have moved our work and social lives to the Metaverse. Aside from some successful city-level experiments with robot taxis, most drivers still pilot their own cars with minimal assistance from AI. And despite breathless predictions of artificial general intelligence (AGI), you’re still reading a magazine column written and edited by human beings.
Instead, we should understand billionaires’ imagined tech futures as projections of their hopes and fears. Behind the hope of AGI is the eternal dream of capital: a world without workers, who have a nasty tendency to demand safe working conditions and equitable pay, and to slow or stop production when they don’t get them.
The latest billionaire dream—housing AI data centres in space—reflects another fear. When you hear a billionaire waxing lyrical about space, they are telling you they are afraid of regulations and seek a world without obligations to governments, neighbours or fellow citizens. They want to exit a world of connection and consequences for the freedom of infinite emptiness.
There’s a version of this idea that makes a certain amount of engineering sense. Solar panels in space generate power constantly, turning to track the sun at all moments of orbit, freeing engineers from the messy world of the power grid, batteries and backup generators. Space is really cold, so the massive waste heat generated by the racks of graphics processing units (GPUs) that power AI could be siphoned off into the void without further warming an overheating planet. Starcloud, a recent graduate of tech incubator Y Combinator, is raising money for space-based data centres with support from venture capital firms such as Andreessen Horowitz and Sequoia.
But it doesn’t take much research to understand why space-based data centres aren’t a sure thing. It’s very expensive to send objects into space—between $2,000 and $8,000 for every kilogram—and racks of servers are quite heavy. Some back-of-the-envelope maths estimates that Google’s Pryor Creek data centre near Stillwater, Oklahoma—a modest-sized data centre at roughly 52,000 server racks, with a rough estimate of 50 million kilograms of mass—would cost $100bn to launch into orbit at the lowest current rates, just slightly less than the cost of the International Space Station, and roughly the annual GDP of Luxembourg.
If space-based AI data centres became a thing, engineers would surely design lighter servers for launching into orbit. Launch costs are coming down, led by the private companies SpaceX and Blue Origin, but for now placing a data centre in space mostly seems like a way to turn billion-dollar projects into hundred-billion-dollar ones.
Solve the mass launch problem and there are still compelling reasons to keep servers on the ground. Without the Earth’s atmosphere to protect them from cosmic radiation, chips won’t be reliable unless they are heavily shielded. Despite the frigid temperatures, cooling remains a problem: because space is a vacuum, there’s no air to take the heat away from computing chips. Space-based servers will need banks for special radiators that siphon off waste heat into space, and careful insulation so that components can operate within specifications. But the most disturbing obstacle of all, perhaps, is that AI chips move rapidly towards obsolescence. The billion-dollar centre, lifted into orbit for hundreds of billions, becomes space junk within five years.
So what’s really going on when Elon Musk and Sam Altman dream of orbiting data centres?
While many investors have bought into an AI future, not all communities have. Data Center Watch, a research firm tracking community opposition to data centres in the United States, reports that 20 projects were blocked in the second quarter of 2025, disrupting $98bn in potential investment. Data Center Watch tracks 188 community organisations in the US which collectively have had a 66 per cent success rate in blocking or slowing projects.
Those opposed to data centres often cite electricity and water usage as a reason to stop new projects. Electricity rates are rising above inflation across the US, and some data centre opponents worry that consumers are now competing with massive companies for scarce essential resources. Republican-leaning Georgia elected two Democrats to the state’s utility regulator in 2025, unseating Republican incumbents, in what has been seen as a reaction to soaring power bills.
There has also been resistance to Project Jupiter, an OpenAI-backed data centre near El Paso, Texas, over its water usage, as residents of nearby communities have complained for years about substandard drinking water with high levels of arsenic. In absolute terms, data centres use much less water than other industrial processes, such as oil refining, but the haste with which local economic development organisations are working to ensure data centres have enough for cooling feels insulting to activists who’ve fought for safe drinking water. Bloomberg reports that two-thirds of new AI data centres in the US are being developed in places that experience water stress, while projects in China, India, Saudi Arabia and the UAE are also slated for extremely dry areas.
In 2021, almost two years before ChatGPT sparked the AI craze, technology scholar Kate Crawford published Atlas of AI, a comprehensive overview of ethical issues associated with the emerging industry. The most striking of Crawford’s criticisms of AI—which also included its being trained on copyrighted texts, and fine-tuned by poorly paid workers—is her focus on its physicality. She links the lithium batteries and heavy metals used in chip manufacturing to the environmental harms caused by the mining industry, declaring AI both a physically and metaphorically extractive industry that turns natural and cultural resources into fuel for a new form of factory. Crawford introduces a question that’s become increasingly important: where is AI?
One answer: AI is not in Europe. European electricity is prohibitively expensive, in part due to surges in natural gas prices caused by Russia’s invasion of Ukraine. But prices are also high due to policy decisions intended to discourage the use of coal and encourage solar and wind energy. It’s these sorts of policies, designed to advance societal goals at some cost to business, that keep the AI barons up at night. Even if electricity were more affordable in Europe, the tendency to value natural resources and the historical character of communities makes the continent the very opposite of deep space. Better to build in absolute monarchies like the UAE, where OpenAI is now siting its first non-US “Stargate” data centre.
In other words, it matters very little if data centres in space are as impractical as NFTs or Elon Musk’s Cybertruck. They are important as a signal: they tell us that the AI barons have no time for petty concerns like affordable electricity or drinkable water. Governments that want to protect their citizens and the environment need not apply—the AI industry would rather launch itself into an airless vacuum than engage with communities so short-sighted that they would advance their own welfare over the holy grail of achieving AGI.
