We are currently hitting a ceiling on how much AI we can run on Earth. Building traditional data centers—the massive warehouses packed with thousands of computers used to train and run AI—requires enormous amounts of electricity and water for cooling. As the demand for AI grows, these facilities are straining local power grids, leading some of the world's largest companies to search for unconventional places to plant their servers.
Several major tech companies, including SpaceX and Meta, are pivoting toward new strategies to provide AI computing power. SpaceX is floating the idea of an "orbital data center"—a fleet of satellites in space rigged to function as a floating computer network. Meanwhile, Meta is moving to lease out its own "compute"—which is simply the raw processing power, or the "thinking time," that specialized chips provide to run AI calculations—directly to other businesses, positioning itself as a rival to existing tech giants. Essentially, these companies are trying to monetize the expensive hardware they already own by turning it into a utility that others pay to use.
The space paradox
To understand why companies would even consider space, you have to understand the "compute" bottleneck. Modern AI systems run on advanced chips, like those from Nvidia, which act as the brain of the operation. These chips are incredibly fast but notoriously hungry; they require massive amounts of power and generate intense heat during their calculations.
On Earth, we use large fans and liquid cooling systems to keep these chips from overheating. In space, that becomes a massive engineering puzzle. There is no air in space to carry heat away, so you rely on "radiators"—large panels that shed heat into the void. Engineers are finding that for the most powerful AI chips, you would need an impossibly large radiator panel just to keep a single computer from frying. Beyond the physics, there is the sheer scale: to build a network of millions of satellites, you would need to launch more rockets than humans have launched in all of history, a feat that would take decades at our current speed.
For now, space-based computing remains more theory than reality. The reason companies are discussing it anyway is because it signals a shift in perspective. Industry leaders are no longer asking how to optimize current data centers; they are beginning to treat processing power as a resource that might eventually need to leave the planet to keep scaling.
Whether this ever becomes a practical reality depends on finding ways to solve the cooling problem and lowering the cost of rocket launches enough to make the math work. For the average person, this is a sign that the AI trend is hitting a physical wall—we are running out of traditional power and space to store these mental engines, and the race to find a solution is pushing humanity toward increasingly radical geography.
Liked this one? The next lands at breakfast.
Every story in tomorrow's AI news, rebuilt in plain English — five minutes, sources linked, free forever.
By joining you agree to receive Article's daily newsletter — unsubscribe in one click. Privacy