We usually think of the internet as a cloud, but it is actually a heavy industrial machine anchored to the ground by copper wire and power plants. That assumption broke this week. With a quiet regulatory filing and a corporate merger signed on Monday, the plan to move the physical hardware of artificial intelligence—the actual servers and processors—off the planet entirely has shifted from science fiction to a formal business proposal.
Key Takeaways
- SpaceX and xAI officially merged on Monday.
- SpaceX filed plans with the FCC for a network of one million data center satellites.
- The FCC accepted the filing and set a public comment schedule on Wednesday.
The timing here is not accidental. On Monday, Elon Musk formally merged SpaceX and xAI, creating a single entity that controls both the launch vehicles and the artificial intelligence software. By Friday, the new conglomerate had filed paperwork with the FCC to launch one million satellites designed to function as orbital data centers.
This is a pivot from simply beaming the internet down to Earth. The new plan involves doing the heavy computational work in orbit. FCC Chairman Brendan Carr has already shared the filing publicly, signaling a smooth regulatory path as long as the political winds remain favorable. Musk is now arguing that within 30 to 36 months, space will become the most economically viable place to process AI data.
The big deal
Data centers are hungry. As AI models grow, they consume massive amounts of electricity, straining local power grids and driving up costs. On Earth, you have to pay for land, cooling, and utility-grade power, which is often generated by burning fossil fuels. In space, the sun shines constantly.
Musk claims that solar panels in orbit generate about five times more power than they do on the ground because there is no atmosphere or night cycle to interfere. If this plan works, it effectively offloads the energy burden of the AI boom to space. It also suggests that the trillions of dollars currently earmarked for terrestrial data center infrastructure could be redirected upward. Musk predicts that in five years, his companies will launch more AI capacity annually than the total existing capacity on Earth.
How it works
The concept is to put the computer processors directly on the satellites, power them with high-efficiency solar arrays, and beam the completed data back down.
Think of it like cooking. Right now, we run AI like a massive commercial kitchen in the city: it is easy to get supplies delivered, but the rent is high and the electric bill is enormous. Musk wants to move the kitchen to a solar-powered food truck. The truck has no rent and free energy from the sun, but it has to carry everything it needs for the trip.
Instead of building concrete warehouses full of servers plugged into the grid, SpaceX intends to launch self-contained computing units. These units capture solar energy, run the calculations locally in the vacuum of space, and send only the results back to the user.
The catch
The math has some holes. While solar power is indeed more potent in space, power is not the only cost of running a data center. Launching mass is expensive, and radiation shielding adds weight. Critics have noted that if a GPU fails on Earth, a technician walks down the hall and replaces it. If a GPU fails in orbit, it is dead forever.
There is also a question of timing. The newly merged company is heading toward an IPO in a few months. Announcing a project of this scale right before a public offering is a convenient way to drive up valuation, regardless of the immediate technical feasibility. The logic assumes that the savings on power will outweigh the immense costs of launch and the inability to repair hardware, a claim that has not yet been proven in practice.
What now?
The FCC has opened the floor for public comment on the filing. Given the current political alignment between the FCC leadership and Musk, approval seems likely. The real test is the timeline: Musk has marked 2028 as the year orbital AI becomes cheaper than ground-based AI.
If you work in data infrastructure, you should watch the IPO prospectus for the new SpaceX-xAI entity later this year. It will likely contain the specific risk factors and technical roadmaps that are missing from the podcast interviews.
