A million satellites, sun-synchronous orbits, near-constant solar illumination, and a thermal management engineering programme that nobody has solved yet. All described as a data center play. But there is a simpler, more coherent explanation for everything SpaceX filed in January - and it has nothing to do with running AI inference in orbit.
Start with what the filing actually specifies. SpaceX requested authority for satellites operating between 500 and 2,000 kilometres in sun-synchronous and 30-degree inclinations, with clusters spaced at 50-kilometre intervals. The company requested the orbital slots and spectrum without a deployment schedule, without a cost estimate, and without a waiver deadline - then asked the FCC to exempt it from the milestone rules requiring actual launches within six years. In regulatory terms, this is less a construction permit than a land claim. The territory being claimed is the finite slots and frequency assignments that determine who can operate what, where, in low Earth orbit, for the next several decades.
The question is not whether the data center story is true. As Part One of this series established, orbital data centres at commercial scale face physical constraints that have no engineering solution in the near term. The question is what the underlying infrastructure being built and the orbital rights being secured are actually for, if not for data centres. The answer points toward something both more ambitious and more strategically coherent than anything being publicly discussed.
Sun-synchronous orbit is a specific orbital geometry in which the satellite's orbital plane precesses at the same rate as Earth's movement around the Sun, keeping the satellite in a roughly constant relationship to the solar terminator. Satellites in SSO receive near-continuous solar illumination - typically more than 99 percent of their orbital period. This makes SSO the preferred orbit for Earth observation satellites that need consistent lighting conditions. It also makes SSO the optimal orbit for solar power collection.
SpaceX's FCC filing explicitly states that different orbital shells within the constellation will be designated for "workloads requiring constant compute" versus those handling "peaks in demand," mapping the solar illumination geometry directly to operational power availability. For a data center, this framing is almost meaningless - compute demand does not track the Earth's terminator. For a solar power generation network, the framing is precisely correct: the sun-synchronous shells are the base-load generators and the lower-inclination shells are the peaking capacity, exactly as a grid operator would describe a combination of baseload and peaking generation assets.
Aetherflux, a company that spent its first three years explicitly building what it described as "an American power grid in space," provides the clearest public confirmation of this alignment. In December 2025 it announced its "Galactic Brain" orbital data center project - and in the same press release, its COO described the company's laser power-beaming technology as "foundational" to the data centre initiative and noted that the same technology was being explored with Lockheed Martin for a NASA lunar energy system. A power-beaming company rebranding as a data centre company at the moment the data centre narrative becomes investable, while its core technology remains unchanged, is not a pivot. It is a costume change.
Every kilogram of hardware launched from Earth's surface must fight a gravity well that requires approximately 9.4 kilometres per second of velocity change to reach low Earth orbit. The dominant mass fraction of any spacecraft is propellant. Electric propulsion - ion drives and Hall-effect thrusters - achieves dramatically higher efficiency than chemical rockets but requires substantial electrical power to operate. A high-powered Hall thruster capable of moving significant masses around the inner solar system might consume 50 to 200 kilowatts continuously. At current space-grade solar panel performance, that thruster requires 250 to 1,000 kilograms of solar array mass just to power it - mass that itself must be launched.
Abundant, cheap orbital solar power breaks this recursion. If power is available from a distributed orbital grid rather than from panels that must be launched with each spacecraft, the mass budget of every vehicle operating in that power environment improves dramatically. Orbital tugs, propellant depots, in-space manufacturing facilities, and cislunar transport vehicles all become lighter and more capable when they can draw power from infrastructure rather than carrying it. The cislunar infrastructure market was valued at approximately $13.8 billion in 2025 and is projected to reach $24.8 billion by 2032 - and that projection assumes the current power constraint persists.
The second constraint that abundant orbital power removes is propellant supply. Water ice exists in permanently shadowed craters at the lunar south pole. Electrolysing water into hydrogen and oxygen produces rocket propellant. Electrolysis requires electrical power. A propellant depot at a lunar-orbit Lagrange point, supplied by ice mined from the lunar surface and processed using orbital solar power, becomes the filling station that makes the rest of the inner solar system economically accessible.
The third constraint is manufacturing. Varda Space Industries and Space Forge have demonstrated that in-space manufacturing is viable - Varda returned pharmaceutical compounds processed in microgravity to Earth in 2024. The constraint on scaling these operations is power. Smelting metals, processing chemicals, extruding structural components - all require sustained energy input. A distributed orbital solar power network changes that equation for every industrial process that benefits from microgravity, hard vacuum, or the specific thermal environment of space.
Separate from any specific application, the FCC filing secures something of durable and independent value: priority rights to orbital positions and spectrum allocations that no competitor can occupy once they are claimed. Peter Plavchan of George Mason University noted in commentary on the filing that "whoever occupies most usable orbits first will effectively prevent other companies or nations from hosting satellites in those orbits." The ITU's coordination rules operate on a first-come, first-served basis with respect to spectrum and orbital slot priority.
China clearly understands this dynamic. Its filings for nearly 200,000 satellites in two mega-constellations are widely understood within the industry to be as much about securing orbital and spectrum positions as about near-term deployment plans. SpaceX's request for a waiver of the FCC's standard deployment milestones is in this context entirely rational. The orbital slot rights are the valuable asset, regardless of what eventually populates them.
Read the FCC filing with this lens and the language choices become sharply informative. The filing does not say SpaceX will build the most powerful data centre network in history. It says the constellation is "a first step toward becoming a Kardashev Type II civilization - one that can harness the Sun's full power." A Kardashev Type II civilisation uses the entire energy output of its star. Data centres do not approach that scale. A distributed orbital solar power infrastructure that powers every human activity in the solar system does.
The S-1 confirms the scale: SpaceX's stated long-term goal is 100 gigawatts of annual orbital compute capacity, delivered by thousands of Starship launches per year. The S-1 also discloses Terafab - a proposed $20 to $25 billion semiconductor fabrication joint venture with Tesla and Intel - which would produce a space-hardened chip designated the D3, designed specifically to run at higher temperatures to reduce radiator mass in vacuum. This is genuine engineering work on a genuine problem. It is also work whose timelines the S-1's own risk language describes as uncertain, with neither Tesla nor Intel obligated to remain in the project.
The Anthropic contract - $1.25 billion per month for terrestrial compute, running to May 2029 - is the most honest near-term statement in the entire S-1, because it reveals what the company actually believes about its own orbital timeline: not two years, not three, somewhere past 2029 at minimum, and subject to breakthroughs the risk language explicitly says may never arrive.
The most defensible reading of what SpaceX is actually building is an orbital power and logistics infrastructure platform for the space economy of the 2030s and beyond. The orbital slots being claimed are genuinely valuable. The thermal management technology being funded is genuinely necessary. Terafab's D3 chip, if it is ever built, is the right engineering response to the radiation and thermal problems that prevent standard silicon from operating at scale in orbit.
What is fake is the near-term timeline, the AI data centre framing as the primary near-term application, and the implicit suggestion that investors buying the late-June listing are purchasing access to orbital compute revenue rather than to a decades-long infrastructure build whose payoff is contingent on a space economy that does not yet exist at commercial scale. The S-1's $18.67 billion in revenue and $4.94 billion net loss make the dependency on the orbital thesis for valuation explicit: without it, 94 times trailing sales is indefensible.
The bridge being sold in late June may lead somewhere worth going. The problem is the delivery date painted on the side. The how of that misdirection - the specific technical claim that makes the data centre story sound credible to a non-specialist - is examined in Part 2a of this series. The full mechanics of what the S-1 reveals about the distance between the roadshow and reality is Part Three.
Related Links| Subscribe Free To Our Daily Newsletters |
| Subscribe Free To Our Daily Newsletters |
