FCC filing dated January 30, 2026: SpaceX mentions up to one million “data center” satellites for orbital AI with xAI. Costs, risks, impacts.
Summary
On January 30, 2026, SpaceX filed with the FCC to obtain a framework for eventually deploying up to one million satellites designed as orbiting data centers. The idea is simple on paper: space sensors and data streams generate too much data to send everything back to the ground, and solar energy is almost continuous in orbit. By bringing computing closer, SpaceX is aiming for “orbital AI” capable of sorting, compressing, training, and inferring faster, with laser links between satellites. The financial signal is also clear: xAI raised $20 billion in early January and SpaceX is preparing, according to the press, for a massive IPO. But orbital mechanics are no laughing matter: multiplication of objects, collision risks, spectrum constraints, light pollution, legal liability. The case therefore opens up a battle: energy and industrial innovation versus the sustainable viability of orbit and political acceptability.
The FCC filing that changes the scale of the debate
The raw fact is spectacular: a filing submitted to the Federal Communications Commission to authorize a system of one million satellites geared toward computing, not just telecommunications. The information was revealed by several Anglo-Saxon media outlets in late January and early February 2026, based on the filing and its initial public summaries.
This figure should be read as a ceiling, not as a fixed production plan. Constellation operators often ask for “large” numbers in order to negotiate later. SpaceX has already done this with Starlink, requesting very high volumes and then moving forward in stages, at the pace of authorizations and industrial constraints.
But even as a “ceiling,” the symbol is significant. We are moving from a world where low orbit hosts communication networks to a world where orbit would become an extension of the cloud, with distributed computing, storage, and potentially more sensitive processing functions. In other words: space as a server room. And this is no longer a metaphor.
The industrial logic behind AI in space
SpaceX’s reasoning can be summed up in three words: energy, latency, bottlenecks.
First, energy. In orbit, solar resources are abundant and more consistent than on the ground, with no weather variations, no seasons in the terrestrial sense, and different intermittency constraints depending on the orbit. The idea is to convert this energy into electricity, then into computing power, then into AI “value.” On paper, it sounds almost too good to be true: you plug in some panels, you get teraflops, and no one complains about the noise from the cooling units.
Then there’s latency and, above all, throughput. Much of the data produced by the space ecosystem (imagery, weather, surveillance, telemetry, science) ends up being brought back to the ground via stations, with visibility windows, spectrum limitations, and infrastructure costs. Moving some of the sorting, compression, annotation, and analysis into orbit reduces the volume that needs to be brought down and speeds up the “sensor → decision” cycle.
Finally, there is the bottleneck of terrestrial AI infrastructure. Over the past two years, the world has discovered that AI is not just a matter of algorithms. It is a matter of electrical power, land, local acceptability, networks, and water for cooling. Data center projects encounter resistance and very prosaic physical constraints. The bet on orbit is to bypass some of these frictions in exchange for other, much more… orbital frictions.
The technical building blocks that make the idea credible
The concept of orbiting data centers
The term orbital AI covers several functions, which are not equivalent:
- Edge inference: executing models to detect, classify, correct, and alert.
- Preprocessing: radiometric correction, mosaicking, intelligent compression, filtering.
- Fleet orchestration: optimizing the allocation of tasks to computing nodes.
- Training: much heavier, more energy-intensive, and more error-prone.
In a realistic short-term scenario, inference and preprocessing are the natural targets. Massive “GPU farm-style” training in orbit is theoretically possible, but this is where problems of mass, heat dissipation, and maintenance become brutal.
Inter-satellite links and distributed architecture
For a cloud of computing satellites to resemble a coherent system, rapid exchanges between nodes are required. SpaceX is relying on laser links already present on certain generations of Starlink. The benefit is twofold: offloading the ground and building a mesh network over oceans or poorly covered areas.
A credible architecture would resemble distributed computing: heterogeneous nodes, divided tasks, redundancy, and high fault tolerance. Because in orbit, failure is not an exception: it is an operating cost.
Thermal management, the real cost of vacuum
It is often believed that vacuum “cools.” This is false. Vacuum insulates. In orbit, there is no convection; dissipation occurs mainly through radiation, via radiators. The more you calculate, the more you heat up. The more you heat up, the more you have to radiate, so you need more surface area, which adds mass and complexity.
This is where the idea of a “satellite data center” hits an engineering wall. A communications satellite already supports demanding thermal management. A high-density computing satellite continuously converts electricity into heat. The design becomes a constant compromise between power, mass, radiating surface area, and lifespan.
Radiation resilience and computing reliability
Space damages electronics. Energetic particles cause errors (single-event upsets), degradation, and failures. Running intensive computing loads in a radiative environment requires fault-tolerant architectures, error correction, redundancies, and sometimes more rugged hardware… which is more expensive and less efficient for the same power consumption. In short, there are no miracles, only trade-offs.
The orbits under consideration and the question of cohabitation
Public summaries refer to an organization in narrow orbital shells (layers) and altitudes ranging from low orbit to significantly higher altitudes, with varying inclinations, including configurations close to solar synchronous. If these parameters are confirmed in the full public documents, they point to one objective: to fit into orbital corridors while leaving room for other systems.
The economic model and the question of who finances what
This is where we need to be frank: a million satellites, even “later,” is a project that cannot be financed on a shoestring.
On the xAI side, the news is clear: the company announced a $20 billion fundraising round in early January 2026. Reuters also reported a valuation of around $230 billion for this round, and discussions of a merger between SpaceX and xAI.
On the SpaceX side, Reuters mentioned a planned IPO that could seek to raise more than $25 billion, with a valuation exceeding $1 trillion according to sources close to the matter. Again, this is not a turnkey “orbital data center” budget, but it is a possible fuel source.
The question of unit cost remains. Precise manufacturing figures are closely guarded. We can therefore only work with public and conservative ranges. A useful benchmark: the Starlink 2024 progress report mentions a mass of approximately 575 kg (1,267 lb) for “optimized” V2 Mini satellites. Another industry source cited by Eurospace used cost assumptions per satellite of between $1 million and $1.5 million depending on the version, noting that the “marginal cost” announced in the public debate may be lower than the full cost.
Even taking a very low and highly theoretical assumption of $0.5 million per “all-inclusive” satellite (manufacturing + a share of the launch), we are talking about $500 billion for a million units. At 1 million, that’s $1 trillion. And that doesn’t include ground stations, operations, or replacement. The figure serves to illustrate one thing: the project is not a “product,” it is a strategy. And a strategy that only makes sense if SpaceX succeeds in industrializing launches at a radically lower marginal cost with Starship and in transforming orbital calculations into recurring revenue.
Pressure on orbit and the risk of backlash
Object density and the risk of collisions
The European Space Agency points out that the amount of debris and objects being tracked is increasing rapidly: approximately 35,000 objects are being tracked, only a fraction of which are active, with the rest being debris larger than 10 cm. ESA statistics also indicate catalogs of tens of thousands of tracked objects and more than 14,000 satellites still in operation, figures that depend on definitions and tracking networks.
A drastic increase in the satellite population automatically multiplies interactions and coordination complexity. Even if each satellite is capable of maneuvering, priorities, passing rules, failures, and the unpredictable must be managed. And the unpredictable, in orbit, ends up as space debris.
Cascade dynamics and the fear of a systemic effect
The dreaded specter is the Kessler effect: a scenario in which collisions generate more debris than is naturally deorbited, gradually rendering certain altitudes unusable. Analyses and opinion pieces remind us that the risk increases as low orbit fills up, and that current rules are struggling to keep pace.
Light pollution and friction with astronomy
Starlink has already crystallized a conflict with the astronomical community. Even with visors, coatings, and mitigation efforts, constellations alter the sky, complicate certain observations, and raise radio frequency issues. A “data center” constellation could add constraints, depending on its frequency bands, lasers, and density.
Responsibility and regulatory risk
The FCC filing is only one part of the picture. There is also international coordination (frequencies, interference), deorbiting rules, mitigation requirements, and political acceptability. In January 2026, the FCC showed that it could authorize, but also limit: SpaceX obtained 7,500 additional Gen2 satellites, not the nearly 30,000 requested, with strict deployment milestones.
In short: the authority knows how to say yes, but it also knows how to say “not at this speed, not on this scale.”
A million satellites would therefore be, at the very least, a marathon of paperwork, arbitration, and compromise.
The geopolitical context that is driving acceleration
This issue does not come out of a strategic vacuum. China has also announced ambitions for space computing centers and a “Space Cloud” by 2030, with large-scale solar infrastructure in the pipeline.
In this context, in-orbit computing becomes a matter of technological sovereignty: who controls the entire chain, from launch vehicle to platform, from network to AI models? The rumored merger between SpaceX and xAI takes on a particular significance here: integrating launch, constellation, and AI software into a single industrial strategy.
The final question that no one can avoid
The project is exciting because it tackles a real problem: the energy and material hunger of AI. It is worrying because it treats orbit as an infinitely expandable industrial terrain, when in fact it is finite, fragile, and already crowded.
The key point in the coming months will be less the figure “one million” than the trajectory of credibility: which orbits, which mitigation rules, what transparency on maneuvers, what guarantees of deorbiting, and what demonstrable small-scale economic model. If SpaceX can prove that an orbital “compute layer” really reduces costs and impacts on the ground, while remaining sustainable in orbit, then the idea could catch on.
Otherwise, this project will remain what it already is: a gigantic proposal, launched as a challenge. With one major difference from a punchline: in orbit, the consequences remain for a long time.
Sources
Reuters, SpaceX seeks FCC nod for solar-powered satellite data centers for AI, January 31, 2026.
Reuters, Exclusive: Musk’s SpaceX in merger talks with xAI ahead of planned IPO, January 29, 2026.
xAI, official Series E announcement ($20 billion), January 2026.
Reuters, Musk’s xAI raises $20 billion in upsized Series E funding round, January 6, 2026.
Reuters, SpaceX IPO could raise more than $25 billion, December 2025.
ESA, ESA Space Environment Report 2024 (tracked objects, debris, trends), July 19, 2024.
ESA, DISCOSweb Space Environment Statistics (statistics on objects in orbit).
FCC, FCC Approves Next-Gen Satellite Constellation (Gen2 authorization), January 9, 2026.
SpaceX, Starlink Progress Report 2024 (optimized V2 Mini mass).
Space.com, Starlink satellites: facts, tracking and impact on astronomy, December 2025 / January 2026.
DatacenterDynamics, SpaceX files for million satellite orbital AI data center megaconstellation, January 31, 2026.
The Verge, SpaceX wants to put 1 million solar-powered data centers into orbit, January 31, 2026.
Live a unique fighter jet experience