Space data centers: compelling physics or overhyped narrative ahead of a SpaceX IPO?
Dec 9, 2025
Key Points
- The space data center pitch rests on eliminating two major Earth data center costs: power via 24-hour solar exposure and cooling via radiative dissipation into vacuum, but chip cost, not energy, dominates inference economics.
- Operational barriers are severe: orbital repairs require astronaut intervention or retrieval missions, satellites pass through Earth's shadow periodically requiring battery backup, and thermal dissipation in zero-gravity remains an unsolved materials challenge.
- Gavin Baker's $900 million SpaceX investment creates a financial incentive to promote space data centers ahead of a potential IPO, while Delian Asparouhov's skepticism reflects the gap between first-principles physics and real-world space operations.
Summary
The case for data centers in space rests on a straightforward physics argument. Two of the three inputs to running a data center—power and cooling—become dramatically cheaper or free in orbit. Satellites can maintain solar exposure 24 hours a day, generating roughly 6x more irradiance than Earth-based systems. Cooling becomes passive when radiators on the dark side of a satellite dissipate heat directly into near-absolute-zero vacuum, eliminating the massive cost of HVAC, coolant distribution units, and liquid cooling infrastructure that dominates Earth data center weight and expense.
Inter-satellite laser links move data faster through vacuum than through fiber-optic cables, which matters for inference workloads where packets travel from user to computation and back. Gavin Baker frames this as a superior user experience for AI queries.
But the pitch faces hard pushback from people with actual space and data center experience. Delian Asparouhov, a Founders Fund partner who has launched hardware into orbit and sits on boards of space companies, remains skeptical. Energy cost is not the binding constraint on inference. Casey Anwar's analysis shows that for inference workloads, chip cost dominates total token cost while energy is a small slice. Eliminating energy savings does not move the needle meaningfully on unit economics.
The operational challenges are more severe. Swapping out a failed chip or performing mechanical repairs on an orbital data center requires either astronaut intervention or a retrieval mission. Baker claims 24-hour sun exposure, but satellites pass through Earth's shadow periodically depending on orbital mechanics, forcing the system offline or requiring backup power. Radiative cooling in vacuum is not passive but a material science problem. Joe Morrison, an engineer at Umbraco, noted that thermal engineers struggle with the casual treatment of radiative cooling in near-zero-gravity, large-mass environments.
The deeper tension is reputational and financial. Gavin Baker led Fidelity's estimated $900 million investment in SpaceX at a $12 billion post-money valuation. Trades held SpaceX as its largest VC position at 30-40% of portfolio as of 2022. If space data centers become a dominant narrative heading into a SpaceX IPO, it benefits Baker and SpaceX materially. Delian faces a conflict: Founders Fund holds a major SpaceX position, but Delian understands the gap between first-principles physics and operational reality from launching hardware. AST SpaceMobile stock rose 27% in five days on space data center enthusiasm, suggesting the narrative is already moving markets even as the underlying engineering remains disputed.