Why SpaceX Wants to Move Data Centers Into Orbit
Power is hard to find on Earth, and computers are hard to lift into space. Elon Musk is betting that cheaper Starship launches can flip that equation.
1. Why Musk wants data centers in space
The biggest bottleneck in AI is shifting from chip supply to power and infrastructure.
A ground-based data center needs more than GPUs. It needs large-scale power, cooling systems and a grid connection, and building one takes time.
Musk wants to solve that constraint in orbit. The plan is to place satellites carrying large solar arrays and AI chips in sun-synchronous orbit. They would compute in space and send results to Earth over laser links and Starlink.
Key specifications of the AI1 satellite, as published by SpaceX:
| Spec | Figure |
|---|---|
| Average compute power | 175 kW |
| Peak power | 250 kW |
| Wingspan including solar panels | 75 m |
| Deployed height | 30 m |
These are SpaceX's published design figures. They should be kept separate from the performance of any mass-produced unit or the net compute actually sellable to customers.
The design is interesting because each satellite acts as a small data center. About 5,714 satellites running at an average of 175 kW would add up to a nominal 1 GW.
But a failed GPU is hard to replace in space, and a vacuum rules out convective cooling. Generating power from sunlight matters, and so does a design that radiates the resulting heat away.
2. The number that matters more than 1 million satellites: 2028
SpaceX has filed with the FCC for a plan to operate up to 1 million satellites at altitudes of 500 to 2,000 km. The FCC accepted the filing and gathered comments. Acceptance does not mean final approval to operate.
What Musk stressed on Oct. 31 was not the total satellite count but the pace at which new AI compute is installed.
On his forecast, by the end of 2028 the compute newly deployed in space in a single month could exceed new deployments on the ground. In 2029, he argues, the gap widens.
One distinction matters here. This does not mean total compute in space would exceed existing ground data centers. It is a forecast that the pace of new supply would flip.
The first test is whether an initial launch at the end of 2027 and large-scale deployment in 2028 go as planned.
3. SpaceX's real edge is the rocket
Ground data centers build a great deal of infrastructure to get power. A space data center collects sunlight directly, but heavy chips and power equipment must be hauled into orbit.
That difference explains why Starship matters.
If Starship's mass transport and reuse become cheap enough, SpaceX could tie launch, satellite production, communications and AI compute into one business.
SpaceX is also developing a space-optimized system based on Nvidia's Vera Rubin NVL72, in cooperation with Nvidia. Its Gigasat factory in Bastrop, Texas, is the key facility preparing for mass satellite production.
Still, cheaper electricity than on the ground does not make AI compute cheaper overall. The comparison has to use total cost: launch, manufacturing, failures, replacement, communications and financing.
4. Who gains and who is challenged?
| Company or industry | Likely impact |
|---|---|
| SpaceX | Chance to expand from launch and Starlink into an AI compute provider |
| Nvidia | Possible demand for space-specific AI chips on top of ground demand |
| Tesla | New supply chain opportunity if a joint chip production project takes shape |
| Ground data centers | Long-term competitive pressure if some large AI workloads move to space |
| Power and cooling industries | Could keep benefiting from rising ground AI investment, though the very long-term demand mix may change |
Not all compute will move to space. Workloads that move a lot of data, or that need very low latency, favor the ground.
Space may first prove economical for AI jobs that run heavy computation and send back only the results.
Insight Times Editorial Desk
