There is a second case, made loudly by the European study and quietly by everyone else, that orbital compute is better for the planet. This chapter prices it: encouraging in ratio, small in absolute terms.
Energy and carbon
A terrestrial megawatt of compute draws about 1.25 MW from the grid once cooling and conversion are included, roughly 11 GWh a year. At a US average grid intensity near 370 gCO₂/kWh, that is about 4,050 tonnes of CO₂ per megawatt-year. An orbital megawatt draws none of it. Against that sits the launch. A methane-fuelled heavy vehicle burns roughly a thousand tonnes of fuel, producing on the order of 2,750 tonnes of CO₂. A 37-tonne node claiming its share of a hundred-tonne launch carries about 1,020 tonnes of embodied launch emission.
Figure 19.1 — The launch carbon is paid back in months
Payback is about three months. After that, each year of operation is roughly four thousand tonnes that did not happen.
Two caveats belong in the same breath. This counts only combustion CO₂; rockets also inject soot and water vapour directly into the stratosphere, where radiative effects are poorly quantified and not favourable, at hundreds of launches per gigawatt that is a research gap, not a footnote. And against a data centre powered by dedicated new renewables, the advantage narrows substantially.
Water and land
A terrestrial megawatt cooled evaporatively can consume on the order of twenty million litres of water a year, though modern closed-loop facilities have cut this by an order of magnitude. An orbital megawatt consumes none. Powering 1.25 MW continuously from terrestrial solar needs several hectares before the data centre is built; orbit uses none but it consumes orbital volume and cross-sectional area, which Chapter 16 established is scarce and unpriced. The environmental case tends to skip that.
The dividends that are not about energy
The sensor dividend is the most certain public benefit in this book and needs no gigawatt to materialise: compute next to the sensor turns Earth observation from a system that photographs and forgets into one that watches and reports. Resilience cuts both ways: compute detached from a national grid survives grid failure, and an orbital fleet is exposed to a severe solar storm in a correlated way a distributed terrestrial fleet is not. Spillover may be the largest effect of all. Megawatt-scale two-phase cooling, high-voltage distribution, radiationtolerant commercial silicon and fully autonomous operation are all things terrestrial data centres will want as rack densities pass six hundred kilowatts. Chapter 1’s slope does not stop.
Now the honest part Figure 19.2, Honest about scale
US data centre demand alone heads toward roughly 134 GW by 2030. Europe’s most ambitious public programme targets one gigawatt in orbit by 2050. A wildly successful decade might put a gigawatt up by the mid-2030s, well under one percent of demand. Orbital compute is not a climate solution and cannot become one at any plausible scale in the next fifteen years. It is a relief valve for capacity that cannot be built in time, and its environmental advantages are a bonus rather than a reason for it to exist. Any deck leading with the carbon argument is selling something, and the giveaway is that the carbon argument is the only one in this sector that requires no engineering to be true.
The reason to build in orbit today is a premium buyer and a long queue. The reason to build in orbit in 2032 is that it will be the cheaper way to add a gigawatt. Everything in this chapter is a dividend on a decision made for a different reason.
Part VI sets the horizon and then makes the case against everything argued here.