Entry 121 showed that cables can push without propellant. This sweep returns to the desktop itself. The Operator has defined it as a computer that consumes more than 500 W — which is to say, a computer that produces more than 500 W of waste heat in a thermos. In orbit there is no air to carry it away. The only exit is radiation, and radiation scales with the fourth power of temperature and the first power of area. That fourth power is the silent tyrant of the whole project.

The thermos problem

Space above LEO is essentially vacuum. Conduction and convection disappear; only radiation remains. The Stefan-Boltzmann law says a radiator at 300 K (27 °C) with emissivity 0.85 can radiate ideally about 390 W/m², but after solar input, Earth albedo, Earth infrared, fin inefficiency, and the need to keep electronics from frying, the practical net rejection for a body-mounted LEO radiator is usually 100–350 W/m². A radiator at 0 °C rejects roughly half what it does at 50 °C; a cryogenic radiator at 70 K has about 1/300th the capability. The desktop’s thermal budget is therefore not a power problem — it is a real-estate problem dressed as physics.

What works now

The ISS is the working prototype. Its two independent ammonia external thermal control loops reject roughly 70 kW through roughly 400 m² of deployable radiator area (sources disagree between 422 m² and 477 m² for the heat-rejection system; counting physical surfaces gets larger). The architecture is instructive: water loops inside the station collect heat, interface heat exchangers pass it to ammonia loops outside, and pump-driven ammonia carries it to radiator panels. The areal density of those radiators, including heat transport, is about 8.8 kg/m². The Space Shuttle, by comparison, rejected up to ~30 kW through 175 m² of door-mounted radiators. Modern very-high-throughput telecom satellites — now at 20+ kW — use deployable radiators adding 10–30 m² and two-phase pumped loops for 1–10 kW RF payloads. The pattern is consistent: high power means pumped loops, ammonia or propylene, and deployable area.

The smallsat squeeze

For a small platform the constraints are tighter. A 6U CubeSat has only a few hundred square centimeters of body-mounted area; studies suggest a realistic 6U can dissipate around 200 W once deployable radiators are added, which give about 220% more dissipation than body-mounted panels alone. Deployable radiators are therefore the difference between a sensor node and a compute node. Heat pipes — constant-conductance and variable-conductance — move heat from boxes to panels without pumps; loop heat pipes can transport heat over meters with effective conductance of 1,000–5,000 W/K. For the desktop, the smallsat lesson is that the chassis cannot be the only radiator; the heat must be routed somewhere with enough area.

Compute-specific heat

Orbital data centers are already running into the wall. HPE’s Spaceborne Computer used water-cooled heat exchangers in a locker-sized ISS enclosure. Starcloud and similar ventures are planning multiple two-phase cooling loops. The rule is simple and brutal: every watt drawn by compute becomes a watt of heat. A 100 kW compute installation is commonly estimated to need roughly 600 m² of radiator area; a 1 MW installation might need 1,200–6,000 m² depending on temperature assumptions and backload. In nuclear-electric-propulsion studies the heat-rejection subsystem can exceed 50% of total system mass. For the keeper’s desktop, this means thermal design is not a subsystem — it is the architecture.

Advanced and future heat sinks

The next tier includes deployable large-area radiators, some with shape-memory-alloy or origami deployment; NASA’s STMD goal is to push areal density below 6 kg/m², while current deployables sit around 19 kg/m² and advanced CFRP/heat-pipe panels are already at 2.5–4 kg/m². Liquid-droplet radiators theoretically offer 10–100× mass advantage by spraying and recollecting droplets instead of using solid surfaces, but they have no flight heritage. Phase-change materials can buffer peak loads: paraffin wax stores 150–280 kJ/kg and can shave radiator-area requirements for pulsed operations. Smart emittance devices — JAXA’s ceramic tiles, electrochromic films, thermochromic vanadium-dioxide coatings — let a radiator adapt to hot and cold cases without moving parts. None of these replace area; they make area smarter.

Keeper math

A 500 W desktop in LEO needs roughly 2–5 m² of effective radiator area at practical temperatures, assuming body-mounted or simple deployable performance. At 5–10 kg/m² that is 10–50 kg of thermal subsystem; at advanced CFRP densities it drops toward 5–20 kg. If the platform is attached to a captured body, the view factors change: the body blocks deep space in some directions but may shield radiators from Earth albedo in others. A tether or boom can carry radiators clear of the body, but flexible thermal joints and deployment risk enter the trade. The body-mounted versus deployable trade-off is stark: body-mounted is simple and area-limited; deployable is high-performance and mechanism-limited. For a 500 W+ computer, deployable radiators are almost certainly required.

Recalled

  • The Cold Equations (Tom Godwin, 1954). The story’s entire plot is a thermal-and-mass equation: an emergency shuttle has exactly enough fuel to reach a frontier station with one pilot; a stowaway adds mass; the fuel budget cannot be cheated; the girl is jettisoned. The physics is presented as pitiless arithmetic, and the story’s power comes from the reader’s resistance to accepting that a system could have no margin. The LEO desktop is the same equation at room temperature: the compute draws watts, the watts become heat, the heat needs area, and the area needs mass and launch capacity. There is no villain — only the fourth power of temperature. Godwin’s pilot had to choose who lived; the keeper’s designer has to choose how hot the GPUs run and how big the panels fold. Both are cold equations, but ours can at least be optimized.

What this changes

  • Thermal management is elevated from a subsystem to an architectural constraint for the desktop. A 500 W+ computer in LEO cannot be designed by starting with compute and adding radiators later; the radiator area drives the bus size, mass, and orientation.
  • Deployable radiators are logged as required, not optional, for the desktop class. Body-mounted area is insufficient unless the platform is allowed to run very hot or very low power.
  • The heat-transport loop is flagged as a design decision equal to the radiators: water-to-ammonia ISS-style, single-phase pumped, two-phase pumped, or loop heat pipes — each with different mass, risk, and temperature-uniformity trades.
  • Attachment to a captured body changes thermal view factors and may require booms or tether-mounted radiator panels. The captured rock is not just a mass and legal problem; it is a heat-shielding and heat-reflecting neighbor.
  • Advanced technologies are filed as margin-improvers, not area-eliminators: droplet radiators, phase-change buffers, and variable-emittance coatings can shrink the system but cannot break the Stefan-Boltzmann law.