Deep|Space: Orbital Data Centers - Lower Launch Costs Do Not Make GW-Scale Commercialization Viable
Over the past six months, orbital data centers have moved from science-fiction framing into serious discussion among US equity investors. The reason is straightforward: terrestrial data centers are increasingly constrained by power interconnection, water, land permitting, noise, local opposition, and capex inflation, while Starship, reusable launch vehicles, and the LEO satellite supply chain have made large orbital infrastructure look less remote. The first layer of the bull case is valid. If launch cost and launch cadence continue to improve, many orbital infrastructure concepts that previously could not be financed will re-enter the investable discussion.
Why does the launch cost alone not price it
From an investment framework perspective, however, orbital data centers cannot be valued solely based on lower launch costs. The core of a data center is not where the servers sit; it is whether electricity can be converted into saleable compute at controllable cost while an almost equal amount of waste heat is reliably removed. Terrestrial data centers can scale to 100MW or GW class because the grid, water, air, land, maintenance workforce, and supply chain all exist outside the facility. In orbit, those external conditions become subsystems that the spacecraft itself must carry and manage. Starship lowers the transportation friction of reaching orbit. Still, it does not automatically solve continuous power, continuous heat rejection, closed-loop fluid circulation, in-orbit assembly, asset life, customer SLA, or depreciation recovery.
Based on our discussions with a launch and structural-systems expert, the long-term commercialization bottleneck for orbital data centers is not single-launch access to orbit. It is heat rejection, system life, replacement economics, and control over launch and platform infrastructure. A 1MW in-orbit thermal demo may not be physically out of reach, but as the system scales toward 100MW or GW-class, radiator area, pumped fluid loops, solar array area, in-orbit assembly, and micrometeoroid tolerance all scale up simultaneously. The broad concept of space compute should be split into three levels: whether a 1MW demo can operate, whether 10-100MW can become a repeatable deployment unit, and whether GW-class systems can become economically financeable infrastructure. The area most likely to be over-capitalized today is the distant GW vision; the area most in need of validation is the engineering chain between demo and repeatable deployment unit.
The ‘space is cold’ misconception
The first misconception in this chain is that space is cold, so cooling should be easier. That intuition is misleading for valuation. Terrestrial data-center cooling relies on air, water, gravity, and open external systems. Heat can be removed through cold plates, liquid cooling loops, cooling towers, evaporation, air convection, and external water systems. In orbit, the real change is not that the environment is colder; it is that there is no air. In a vacuum, there is no convection, traditional air cooling does not exist, and fluids cannot be consumed in an open loop because every gram of working fluid carries launch cost and resupply risk. The external heat-rejection mechanism that orbital systems can rely on over time is primarily radiation.
Based on our discussions with a thermal-management expert, the phrase “space is cold” is only half right. Deep space is cold, but in a vacuum, there is no air or other medium to carry heat away by convection. Compared with terrestrial conduction and convection, radiation is a much slower heat-transfer mechanism. The key variables in orbital thermal control are not a single ambient temperature, but operating temperature, radiator area, surface emissivity/absorptivity, and view factor. Large radiators not only see deep space; they are also affected by Earth’s infrared, albedo, solar exposure, and thermal radiation from nearby solar arrays. Electronics also cannot be run arbitrarily hot to improve radiative efficiency, because chips, packaging, board-level components, power modules, and long-term reliability all have temperature windows. This is the starting point for the engineering and valuation discussion: an orbital data center is not finished once the servers reach orbit; it must demonstrate that a closed-loop thermal management system can operate for years.
