- Joined
- Jan 22, 2026
- Messages
- 101
- Reaction score
- 921
We tell you why Musk's plan to capture space with neural networks sounds like madness (and why it might work).
While the world is debating where to get electricity for the new giant computing, Elon Musk offers a radical answer: to bring data centers for artificial intelligence directly into orbit. At the end of Friday, January 30, SpaceX submitted an application to the US Federal Communications Commission (FCC) to launch a constellation of up to one million satellites that should operate as "orbital data centers" and handle AI tasks in space.
The logic is simple and straightforward in a Maskish way: AI is rapidly inflating the demand for computing, and with it rising electricity bills and cooling costs, often with high water consumption. In space, energy almost "lies on the surface": solar panels receive a stream without clouds and day-night alternation in the usual sense, which is why SpaceX describes satellites as something like orbital power plants that power computing all the time.
However, the "million" in the documents almost certainly means an upper ceiling, and not a promise to immediately hammer the sky with iron. This is a common tactic in the industry: ask for permission with a large margin, so that you don't have to run for a new approval every time you change the design. The company has already done similar things with Starlink: previously, it requested permission for tens of thousands of devices, but in reality, it now has about 9,5 thousand working satellites in orbit.
The application refers to low Earth orbit, approximately 500-2000 km above the surface. At such altitudes, it is easier to maintain communication and update the grouping, but the orbital space is already becoming cramped, and each new "layer" increases the cost of error.
An important detail that is often forgotten in enthusiastic retellings: space is not a free refrigerator. There is no air in a vacuum that carries away heat, so it still has to be removed by radiation through radiators. The payoff here is rather different: there is no need to coordinate land for data centers, pull power lines and compete for energy with cities and factories. SpaceX is trying to sell the idea as a way to offload ground-based energy and reduce pressure on the planet's resources.
The whole economics of this venture, in fact, rests on Starship. If launches remain expensive, the "data center in orbit" will remain a beautiful illustration for presentations. If a heavy rocket really brings the cost of cargo withdrawal to the level that SpaceX promises, the project suddenly stops looking fantastic. Reuters explicitly notes that the idea is tied to the success of Starship and its ability to regularly carry payloads.
The context is also interesting. Reuters reported that almost at the same time as the application to the FCC appeared, negotiations on a possible merger of SpaceX with xAI were discussed. If these conversations turn into something real, Musk may have a bunch of "launches plus his own models," where one hand builds the orbital infrastructure, and the other immediately takes it under his calculations.
There is also something to say about the competitors in this story. On paper, such a move helps Musk overtake major players who are now pushing against the limitations of terrestrial energy, from Google to OpenAI and Meta (recognized as an extremist organization and banned in the Russian Federation). But the most interesting thing is that the idea of "AI in space" is no longer unique.: Google has officially announced the Suncatcher research project, which examines solar satellite groupings with computing chips for machine learning.
And Jeff Bezos and Blue Origin recently revealed their orbital stake: Reuters wrote about plans for a TeraWave network of thousands of satellites focused on communications for data centers, businesses and government agencies. This is not an "orbital server", but it is the same race for infrastructure around big computing.
Naturally, the idea has a dark side. Astronomers have long warned that mega-groupings complicate observations, and security experts recall the "Kessler syndrome", when collisions generate an avalanche of debris and make the orbit dangerous for everyone. And while Musk is joking on social media in the spirit of "I misspoke, I wanted to say not a million, but a billion," reality will be determined not by jokes, but by the decision of the regulator and how ready the industry is for orbits clogged with devices.
While the world is debating where to get electricity for the new giant computing, Elon Musk offers a radical answer: to bring data centers for artificial intelligence directly into orbit. At the end of Friday, January 30, SpaceX submitted an application to the US Federal Communications Commission (FCC) to launch a constellation of up to one million satellites that should operate as "orbital data centers" and handle AI tasks in space.
The logic is simple and straightforward in a Maskish way: AI is rapidly inflating the demand for computing, and with it rising electricity bills and cooling costs, often with high water consumption. In space, energy almost "lies on the surface": solar panels receive a stream without clouds and day-night alternation in the usual sense, which is why SpaceX describes satellites as something like orbital power plants that power computing all the time.
However, the "million" in the documents almost certainly means an upper ceiling, and not a promise to immediately hammer the sky with iron. This is a common tactic in the industry: ask for permission with a large margin, so that you don't have to run for a new approval every time you change the design. The company has already done similar things with Starlink: previously, it requested permission for tens of thousands of devices, but in reality, it now has about 9,5 thousand working satellites in orbit.
The application refers to low Earth orbit, approximately 500-2000 km above the surface. At such altitudes, it is easier to maintain communication and update the grouping, but the orbital space is already becoming cramped, and each new "layer" increases the cost of error.
An important detail that is often forgotten in enthusiastic retellings: space is not a free refrigerator. There is no air in a vacuum that carries away heat, so it still has to be removed by radiation through radiators. The payoff here is rather different: there is no need to coordinate land for data centers, pull power lines and compete for energy with cities and factories. SpaceX is trying to sell the idea as a way to offload ground-based energy and reduce pressure on the planet's resources.
The whole economics of this venture, in fact, rests on Starship. If launches remain expensive, the "data center in orbit" will remain a beautiful illustration for presentations. If a heavy rocket really brings the cost of cargo withdrawal to the level that SpaceX promises, the project suddenly stops looking fantastic. Reuters explicitly notes that the idea is tied to the success of Starship and its ability to regularly carry payloads.
The context is also interesting. Reuters reported that almost at the same time as the application to the FCC appeared, negotiations on a possible merger of SpaceX with xAI were discussed. If these conversations turn into something real, Musk may have a bunch of "launches plus his own models," where one hand builds the orbital infrastructure, and the other immediately takes it under his calculations.
There is also something to say about the competitors in this story. On paper, such a move helps Musk overtake major players who are now pushing against the limitations of terrestrial energy, from Google to OpenAI and Meta (recognized as an extremist organization and banned in the Russian Federation). But the most interesting thing is that the idea of "AI in space" is no longer unique.: Google has officially announced the Suncatcher research project, which examines solar satellite groupings with computing chips for machine learning.
And Jeff Bezos and Blue Origin recently revealed their orbital stake: Reuters wrote about plans for a TeraWave network of thousands of satellites focused on communications for data centers, businesses and government agencies. This is not an "orbital server", but it is the same race for infrastructure around big computing.
Naturally, the idea has a dark side. Astronomers have long warned that mega-groupings complicate observations, and security experts recall the "Kessler syndrome", when collisions generate an avalanche of debris and make the orbit dangerous for everyone. And while Musk is joking on social media in the spirit of "I misspoke, I wanted to say not a million, but a billion," reality will be determined not by jokes, but by the decision of the regulator and how ready the industry is for orbits clogged with devices.
