The Billion-Dollar Bet on Space Data Centers That No Law Can Touch

Founders Fund has a theory about what SpaceX, OpenAI, Anthropic, and a two-year-old startup called Starcloud have in common. The short version: they are all pieces of the same bet, and the bet has a regulatory gap the size of a launch window.

That theory — laid out by Partners Delian Asparouhov and the firm's portfolio — is what brought roughly 200 people to a SpaceNews event in Washington last week. Asparouhov, who is also cofounder and president of returnable spacecraft maker Varda Space Industries, spent most of his remarks mapping adjacencies: lunar ice mines as a power source for orbital compute, Crusoe Energy as an AI infrastructure customer and Founders Fund holding, Starship as the launch vehicle that makes the whole thing financially coherent. The legal problem he did not dwell on, but which lawyers and policy researchers have been tracking for two years, is this: the infrastructure he is describing sits beyond any court on Earth.

SpaceX filed an application with the Federal Communications Commission on February 4, 2026, seeking authority to deploy one million satellites as orbital data centers. The filing was accepted. The public comment period is underway. And it exposed, for the first time at regulatory scale, a gap that the Outer Space Treaty of 1967 was never designed to bridge: it was written to prevent nuclear weapons on the Moon, not to regulate encrypted petabyte compute clusters in low Earth orbit.

Article VI of the treaty holds that a launching state retains jurisdiction and control over space objects regardless of where they operate. For a communications satellite, this was always manageable. For a data center processing the personal data of EU citizens, Brazilian users, or Indian customers, it creates what Vishal Sharma, an AI governance researcher at JURIST, calls a "Jurisdictional Mirage" — simultaneously under the legal authority of the United States and beyond the reach of GDPR, India's DPDP Act, or any other privacy framework on the ground. A regulator cannot board an orbital server. No warrant can be served on a satellite at 500 kilometers altitude.

"Every country without launch capacity becomes a data consumer rather than a data sovereign," Olubayo Adekanmbi, founder of EqualyzAI, told Rest of World. "If you do not have launch equity, you are just renting intelligence."

That framing captures the asymmetry more cleanly than any market-sizing slide. The countries best positioned to build and operate orbital compute are the same ones that already dominate terrestrial AI infrastructure. The rest are potential customers.

Founders Fund has placed bets on both sides of the divide. The firm backed SpaceX before Starship existed and continues to hold a stake. It invested in OpenAI and Anthropic when both were research labs. It put money into Crusoe, an AI infrastructure provider that is also Starcloud's earliest customer. Asparouhov was not speaking abstractly when he described lunar ice mining as an orbital compute adjacency — he was describing infrastructure his own firms are positioned to build.

The technical constraints are real and the timeline is not short. Starcloud, the two-year-old venture that has raised $200 million at a $1.1 billion valuation, launched its first satellite with an Nvidia H100 GPU in November 2025. The H100 worked. A companion Nvidia A6000 GPU did not survive the launch. Starcloud's next spacecraft, planned for October 2026, will need the largest deployable radiator ever flown on a private satellite to handle the thermal load — a problem that has not been solved at this scale in orbit before.

The cost case depends entirely on Starship reaching roughly $500 per kilogram to low Earth orbit. Falcon 9 currently runs about $3,600 per kilogram. Starship has not yet reached commercial operational cadence. Starcloud's own CEO told TechCrunch he does not expect cost-competitive orbital power pricing until 2028 or 2029 at the earliest. Musk, speaking at Davos, called orbital data centers a "no-brainer" given solar availability and cooling advantages. He did not give a date.

The regulatory timeline may move faster than the technical one. SpaceX's FCC filing has initiated a public comment process that will force regulators to confront questions they have been deferring. Sharma's analysis, which frames the problem as "Digital Soil" — territory claimed not by flags but by GPU clusters — is the clearest articulation of what is actually being built.

The history of internet infrastructure offers a rough analogy. The open architecture that made the web work was a policy choice, not a technical inevitability. The same is true of orbital compute governance. The question is whether multilateral frameworks will be built before the first movers have already defined the operational norms.

For now, the race is to the launch pad. The thesis that ties SpaceX rockets to AI compute to off-world data sovereignty is not portfolio coincidence. It is a single bet with multiple payoff vectors — and the regulatory gap is a feature, not an oversight, for whoever gets there first.