The constant
Every argument for compute in orbit begins with one number: 1,361 watts per square metre. That is the solar irradiance at Earth’s distance from the Sun, above the atmosphere, and it is the closest thing this industry has to a free lunch. It is not free on the ground, because the atmosphere takes a share, clouds take another, night takes half, the seasons take more, and the panel spends most of the day pointing somewhere other than directly at the Sun. A good terrestrial site delivers perhaps 2,000 to 2,300 kWh per square metre per year onto a tilted panel. A spacecraft in the right orbit sees the full constant, almost all the time, pointed almost perfectly.
The comparison, built up
Take a 30% multi-junction cell. Facing the Sun, it produces 1,361 × 0.30 ≈ 408 watts per square metre. In a dawn, dusk sun-synchronous orbit, an orbit whose plane stays roughly perpendicular to the Sun line, so the spacecraft rides the day, night terminator, the array can be in sunlight essentially continuously for much of the year. Run that for a full year: 408 W/m² × 8,760 h = 3,577 kWh per square metre per year. A conventional LEO orbit, sunlit perhaps 62% of each revolution, yields around 2,200. A tilted panel at an excellent desert site, at 22% module efficiency and a realistic 0.85 performance ratio, yields about 430. A northern European site yields about 224.
Figure 4.1 — The same panel, in two different places
So the honest ratio is roughly eight to one against a very good terrestrial site, per square metre of panel. That is the figure Google’s own Project Suncatcher work arrives at independently, describing solar panels in the right orbit as up to eight times more productive than on Earth.8 Two adjustments, one in each direction. In favour of orbit: this compares panel area to panel area, and a terrestrial solar farm needs land between the rows, so the ratio per acre is larger still. Against orbit: the terrestrial panel is bolted to the ground and the orbital one had to be launched, deployed, pointed, and kept in an orbit that decays.
What 100 kilowatts actually looks like Divide through. A 100 kW node needs 100,000 / 408 ≈ 245 square metres of array, call it 280 m² once you allow for pointing losses, cell degradation, and the power lost between the array and the chips. That is a square roughly 17 metres on a side, or two wings each the size of a tennis court, folded into a launch fairing. Mass follows from specific power, the watts per kilogram a given array technology delivers. Rigid panels sit around 40, 100 W/kg; modern flexible roll-out arrays reach roughly 150 W/kg at beginning of life. At 150 W/kg, 100 kW of generation is about 670 kg of array. At a conservative 100 W/kg it is a tonne. Scale that to a megawatt and the array alone is seven to ten tonnes. Hold that number; Chapter 5 adds a bigger one to it.
Eclipse, and the freedom nobody talks about
If the orbit is not full-sun, the spacecraft passes into Earth’s shadow for roughly 35 minutes of each 90-minute revolution. Conventional spacecraft carry batteries. For a data centre, that is expensive: 100 kW through a 35minute eclipse is about 58 kWh of storage per orbit, which at 200 Wh/kg is nearly 300 kg of battery per 100 kW, before depth-of-discharge margin and before the batteries’ own thermal management. But a data centre has an option a weather satellite does not. It can simply stop computing. Terrestrial data centres are built for continuous availability because their customers demand it and their capital is idle when they are off. An orbital node running batch workloads, training runs, overnight rendering, scientific simulation, anything checkpointable can be a load-following machine: full power in sunlight, idle in eclipse, no batteries at all beyond housekeeping. That is a 35% haircut on utilisation in exchange for deleting a mass and a failure mode. Whether that trade is good depends entirely on the workload, and it is one of the sharper questions to ask any company in this sector: do you carry batteries, and if so, which customer forced you to? Choosing the full-sun orbit instead is the third answer, and it is the one Google’s design takes, a dawn, dusk sun-synchronous orbit around 650 km.8
Where power breaks Four things, and they are where the startups in Chapter 10 live. 1. Deployment. A 280 m² array is a mechanism, and mechanisms are the classic single point of failure on spacecraft. Nothing about the physics fails here; the hinge does. 2. High voltage in plasma. Moving a megawatt around a spacecraft at the tens of volts traditional buses use would require absurd conductor mass, so you want hundreds of volts. But LEO is not empty, it is a tenuous plasma, and high-voltage surfaces in plasma arc. This is a genuinely under-discussed constraint on scaling orbital power, and it is why high-voltage power management is a differentiator rather than a commodity. 3. Degradation. Arrays lose output to radiation, typically a low single-digit percentage per year. Over a five-year life that is a real haircut on the revenue model and it belongs in any honest unit-economics sheet. 4. Drag. A large array is a large sail. At 500, 650 km the atmosphere is thin but not absent, and a high area-to-mass spacecraft decays measurably. Station-keeping propellant is a consumable, and consumables set service life.