SpaceX has filed an application with the U.S. Federal Communications Commission seeking permission to deploy up to 1 million solar-powered data center satellites in low Earth orbit. The proposal would shift parts of AI computing away from ground-based facilities, immediately putting two questions at the center of the debate: rising power demand from AI workloads and whether orbital infrastructure can absorb such a massive increase in satellite traffic.
Orbit is being pitched as an answer to power and cooling limits
According to the source material citing PCMag, the filing was submitted this week on the 30th. The core idea is to use space-based solar energy to power computing nodes outside terrestrial grid constraints. The article also cites World Economic Forum analysis estimating that space data centers could reach electricity costs of $0.005 per kWh, roughly one-fifteenth of average ground wholesale pricing. In addition, vacuum conditions would remove the need for cooling water, a major operating burden for conventional large-scale facilities.
In its filing, SpaceX said the plan represents “the first step toward a star-faring civilization” and framed the effort as more than a fix for current bottlenecks. The company is also tying the proposal to long-term control of solar energy resources.
Starlink has already tested part of the technical foundation
The concept is not being presented in a complete vacuum. Starlink has already placed more than 9,600 satellites in orbit and has validated OISL, or optical inter-satellite laser links. Citing Time, the source says future Starlink nodes could exchange data and perform some real-time computation in orbit, sending only summaries or backups to Earth. That would reduce reliance on terrestrial fiber backhaul.
Google’s Project Suncatcher and Blue Origin’s TeraWave are described as exploring similar directions, but SpaceX’s filing stands out because of its scale. The issue is no longer just whether orbital computing can be attempted. It is how much regulators would ever allow.
The 1 million figure may be a negotiating anchor
Critics have focused on the size of the request. Engadget, as cited in the source, noted that SpaceX applied in 2022 to launch 30,000 Starlink satellites, while the FCC ultimately approved 7,500. That history has led to the view that the new 1 million-satellite figure may function as an opening anchor in a regulatory negotiation. Even after major cuts, the company could still preserve approval for a deployment counted in the hundreds of thousands.
Bloomberg is cited as saying the Trump administration leans toward easing reviews for large infrastructure projects, which could raise the chances of approval. The final number, though, would still depend on hearings and on details such as disposal orbits, active collision-avoidance protocols, and debris-removal systems.
Orbital congestion and collision risk are at the heart of the case
There are currently about 15,000 active satellites worldwide, according to the source. If regulators approved even 10% of the application, low Earth orbit would gain 100,000 additional data nodes. That is a sharp jump. The safety implications are obvious.
Astronomers and environmental groups cited in the source warn that higher satellite density would increase debris-collision risk and could even trigger the Kessler syndrome, a chain reaction of impacts that could compromise access to entire orbital bands. For the FCC, the review is shaping into a balance between support for AI infrastructure and the need to prevent breakdown in space traffic management.
Five hard limits stand between the filing and real deployment
The source lays out several practical barriers. First is launch cost. It says Falcon 9 has pushed the cost to orbit down to about $2,700 per kilogram, with Starship aiming lower, but a computing-capable satellite node would carry servers, solar panels, thermal systems, and communications modules, making it far heavier than a standard communications satellite. At large volumes, launch requirements and aggregate cost remain enormous.
Second is hardware capability in space. Standard data center GPUs and high-bandwidth memory were not built for orbital conditions. Cosmic radiation can cause single-event upsets, while extreme temperature swings threaten stability. The source says current radiation-hardened space chips lag consumer-grade commercial chips by about two to three generations.
Third is heat management. Vacuum eliminates water cooling, but it also removes convective cooling, leaving radiation as the only path. That requires large radiator surfaces, which add mass and size. Fourth is bandwidth and latency. One-way low Earth orbit latency is estimated at 4 to 20 milliseconds, but OISL bandwidth per link remains in the Gbps range, while submarine cables can carry tens of Tbps. That makes orbital compute more suitable for latency-tolerant batch inference than for distributed training that needs heavy parameter synchronization.
Fifth is maintenance and upgrade difficulty. Ground data centers can swap drives, replace GPUs, and repair failed nodes. Satellites in orbit cannot be serviced that way once deployed. When performance falls behind or components degrade under radiation, replacement means launching new hardware and retiring old satellites.
The FCC is still months away from a final decision, but the filing has already moved the idea of space-based data centers from speculative discussion into an active policy process. What SpaceX is asking for is not only spectrum or launch room. It is a claim on future AI infrastructure in orbit.