Is It Possible to Deploy 1 Million Data Centers In Space?
With his project to deploy 1 million space data centers, Elon Musk promises to revolutionize artificial intelligence from Earth orbit. But for now, this spectacular vision collides with unsolvable physical constraints and staggering costs. Here is a closer look.
Since its acquisition of xAI last February, SpaceX has shared a spectacular ambition worthy of a science fiction work: deploying up to 1 million satellites in low Earth orbit. These new kinds of machines, named AI1, would not serve telecommunications like the Starlink constellation, but would act as orbital data centers specifically dedicated to artificial intelligence (AI) computations.
In this way, Elon Musk’s company intends to exploit the almost unlimited resources of space. On Earth, building giant data centers faces power grid saturation, massive water consumption for cooling, and land-use restrictions. But in orbit, the Sun shines almost continuously. Equipped with 600-square-meter solar panels, each satellite could generate about 120 kW of constant power to supply the industry’s most power-hungry graphics chips.
The appeal is so strong that other major tech players are actively exploring this avenue. Google conducted experiments with its TPU chips to test the behavior of its algorithms in orbit, while Jeff Bezos’s Blue Origin also wishes to deploy a constellation of more than 50,000 AI satellites. That is not all, as startups like Starcloud are already sending GPUs into space to prove the feasibility of the concept. Offloading computer processing off our planet is a trend gaining ground across the entire industry. But is it viable?
Elon Musk’s company intends to exploit the almost unlimited resources of space.
In a vast investigation, the specialized outlet Ars Technica quantified the logistics required to build SpaceX’s constellation. And the gap between current space capabilities and the project’s actual needs is simply abyssal.
To place 1 million satellites—each weighing between 3.5 and 7.5 metric tons—into orbit, SpaceX would have to rely exclusively on the giant Starship rocket, which, as a reminder, is not yet operational. But even in the most optimistic scenario with a Starship V4 (the company is currently testing the V3) capable of carrying a 200-ton payload, it would take roughly 3,500 launches per year to deploy and renew the fleet, whose lifespan is estimated at 5 years. That represents nearly 10 liftoffs per day, whereas currently, all of humanity achieved a record of roughly 330 orbital launches over an entire year.
Financially, the bill is just as astronomical. Adding up satellite production costs, launches, and ground infrastructure, the total bill would reach $1.45 trillion in the best-case scenario. However, more conservative assumptions point to a total of around $10 trillion. Without a drastic and unprecedented reduction in space access costs, the economic equation simply does not hold up. At least, for now.
Financially, the bill is just as astronomical.
And that’s not all, because currently, the technology required to operate data centers in space is clearly not ready. The first challenge, and undoubtedly the most complex, concerns heat management. On Earth, air or water allows for dissipating the immense thermal energy produced by processors. However, this is impossible in the vacuum of space, where heat can only be evacuated via infrared radiation. To keep GPUs from frying, each satellite will have to carry immense radiator panels—equipment that significantly adds weight to the structure and complicates the overall mechanics of each craft.
Similarly, the infrastructure will not be protected by the atmosphere or Earth’s magnetic field, leaving it subject to constant bombardment by cosmic radiation. Latest-generation processors are particularly sensitive to charged particles, limiting the useful lifespan of components in orbit to about five years, with no possibility of on-site repair.
Finally, the question of latency remains concerning. While signals travel quickly, the distances separating satellites by several kilometers generate microscopic yet critical delays. This constraint makes space data centers relevant for certain real-time un-demanding AI tasks like inference, but much less suitable for massive model training that requires instantaneous synchronization across thousands of chips.
SpaceX’s announcement of a constellation dedicated to AI computation primarily served as a strategic lever to inflate the group’s valuation prior to going public in June. It was a gamble that paid off, although the stock has since dropped dramatically. Furthermore, low Earth orbit regulation is still underdeveloped. By quickly securing prime altitude bands from regulators, the company gets a head start on the competition.
Skeptical views exist internally as well. In its official document filed with the US financial market watchdog, the SEC, prior to its IPO, SpaceX admitted that these infrastructures relied on “unproven technologies” that “may never become commercially viable.”
By all appearances, this project, frequently highlighted by Elon Musk on X, looks more like a very long-term gamble than a genuine operational roadmap.
Video uploaded by VideoFromSpace on June 9, 2026.
A “100-Year Flood” Threatens the Price of the iPhone 18 Pro: RAM Prices Explode and…
A Signal 100 Times Stronger: The GPS Replacement Is Coming, and It Can Locate Objects…
Why Starlink Will Refund Money to Some Customers and How to Know if You Will…
iPhone User Alert: iOS Update Affects Battery Life: The initial update was released to fix…
It’s The End of Starlink's Monopoly: Amazon Leo Announces That It Now Has Enough Satellites…
Samsung Is Taking On the iPhone Fold Head-on with a Galaxy Z Fold 8 in…