The International Space Station has been running a closed-loop ammonia cooling system since 1998. Heat generated by onboard electronics transfers through internal water loops to external radiator panels, where ammonia loops reject it directly into the vacuum of space. No evaporative cooling. No water consumption. No cooling towers drawing from a municipal supply. The system works because space is the ultimate heat sink: an infinite thermal reservoir at roughly 2.7 kelvin available at no marginal cost.
On January 11, 2026, Axiom Space and Spacebilt launched two orbital data center nodes to the ISS, building on an initial Data Center Unit-1 demonstration deployed in fall 2025. The architecture enables satellites, spacecraft, and researchers to process and store data in orbit, running AI and machine learning workloads in an environment where the cooling problem is solved by physics rather than infrastructure.
Terrestrial liquid-cooled data centers still require facility-side heat rejection. The cold plate removes heat from the chip. A CDU transfers that heat to a facility loop. A cooling tower or dry cooler rejects it to the atmosphere, either by evaporation or convection. Each stage of that chain draws energy. The evaporative stage draws water. In water-stressed regions, that draw is becoming a permitting constraint before it becomes an engineering problem.
In orbit, heat rejection runs on radiator panels that emit directly to the vacuum. No working fluid is consumed. No water rights application is required. The ISS thermal control system, which NASA and its contractors developed across three decades, uses ammonia as the external loop fluid specifically because of its thermodynamic properties at orbital temperatures. The same two-phase phase-change physics that makes ammonia effective in space is what makes refrigerant-based two-phase cooling attractive for terrestrial high-density deployments.
Orbital data centers are estimated to cost roughly three times more to operate than equivalent terrestrial facilities. Space-grade solar panels run approximately 1,000 times the cost of ground-based panels. Launch costs, even at the rates SpaceX currently offers, remain orders of magnitude above terrestrial construction costs. Gartner has been direct about this: the technical arguments for space-based compute are interesting; the economic arguments are not yet there.
What the Axiom-Spacebilt deployment does establish is a proof point for the cooling architecture. The ammonia loop and vacuum radiator system that the ISS has operated continuously for nearly three decades can support data center density. That validation matters because the heat rejection problem for terrestrial liquid-cooled facilities is getting harder, not easier, as rack densities climb toward one megawatt per cabinet. An orbital demonstration of a zero-water cooling architecture is a real data point, even if the economics of orbital deployment remain unfavorable for most operators.
The more immediately relevant question is whether the loop heat pipe and two-phase ammonia technology developed for spacecraft finds its way into terrestrial dry-cooling architectures. Companies already working on zero-water or near-zero-water heat rejection, including the adiabatic dry cooler vendors and the closed-loop refrigerant CDU manufacturers, are working on the same thermal physics from the ground up. The ISS system is not a blueprint operators can specify today. It is a 25-year validation of the principle that high-density heat can be rejected without water consumption at commercially viable operating temperatures.
Regulators in Nevada, Arizona, and Texas are already moving against evaporative cooling. The zero-water pilot programs running in Phoenix and Mt. Pleasant are testing exactly this: whether the dry cooling penalty is acceptable at AI rack densities. The ISS has been running the answer for 28 years. The industry is still working out how to bring it to ground level at data center scale.