2030, 2035, 2040 — The Data Centers in LEO

How to read a horizon

Predictions in this sector age badly and publicly. So this chapter does not make them. It sets out conditions, things that must happen for the sector to reach a given scale by a given date and leaves the forecasting to you. The distinction matters. “There will be a gigawatt in orbit by 2036” is a claim I cannot support. “There cannot be a gigawatt in orbit by 2036 unless a two-phase thermal loop is qualified at megawatt scale, a spacecraft factory reaches constellation economics, and a heavy vehicle sustains a launch cadence measured in hundreds per year” is a claim I can support completely, and it is far more useful, because each clause is something you can watch.

Figure 20.1 — Conditions, not predictions

2030, the proving decade By 2030 the sector should have answered whether the machine works at all. What that requires: Condition

Why it mattersRack-scale compute operating on orbit with paying customersProves the vacuum-packaging and thermal architecture of Chapter 8Deployable radiators and a direct-to-chip loop demonstrated inChapter 5’s easy regime, confirmed rather than modelledflight A published price for orbital computeUntil this exists, every model including mine is theoryA published dollars-per-watt for a mass-produced orbital arrayChapter 8’s largest unresolved numberThe first insurance policy written on a compute nodeChapter 16’s financing gate opens or does notFirst megawatt-class node in flight or in buildThe step from Chapter 8’s Node A to Node B

My expectation for 2030 is tens of megawatts in orbit, not hundreds, a handful of operators, most revenue from sovereign, defence and sensor-adjacent buyers rather than the commodity cloud, and the megawatt-scale thermal question closing. That sounds modest and it is the most important rung on the ladder. Chapter 18’s crossover requires roughly nine doublings of manufacturing experience, and tens of megawatts delivered to customers who pay a premium is precisely how the first four of them get funded. A sector at 40 MW in 2030 is not a sector that failed to reach a gigawatt. It is a sector on schedule to.

2035, the manufacturing decade Everything after 2030 is a factory question. Part V showed why: at bespoke spacecraft costs the business is absurd, and at constellation-scale manufacturing it works. Nothing between now and 2035 matters as much as which of those worlds arrives. Condition

Why it mattersA spacecraft production line at $1,000, 2,000/kg for a 30-tonneThe single decisive variable in Part Vclass vehicle Two-phase thermal transport qualified above a megawattThe unretired technical riskA demonstrated payload swap or module replacement on orbitChapter 15’s answer to the refresh problemA second heavy-lift provider at sustained cadenceRemoves the sector’s single-supplier riskDebt financed against orbital compute assetsThe capital structure the sector needs to scaleRegional constellations under separate jurisdictionsChapter 17’s sovereignty logic playing out

If all six land, a gigawatt in orbit becomes a 2036, 2038 proposition. If the first one does not, none of the others rescue it, and the sector stabilises at a few hundred megawatts serving specialised workloads.

2040, the infrastructure decade Two futures, and they are genuinely far apart. If the manufacturing question resolved well, orbital compute by 2040 is boring infrastructure: several gigawatts across multiple sovereign constellations, routine on-orbit refresh, in-space assembly of structures too large to launch, possibly the first fission power in orbit, and a price for orbital GPU-hours that a procurement officer compares against terrestrial without excitement. The interesting companies by then are the ones nobody writes about, the terminal supplier, the radiator manufacturer, the servicing operator. If it did not, orbital compute in 2040 is a real but modest industry doing exactly one thing extremely well: processing sensor data next to the sensor. Every Earth-observation constellation carries meaningful compute; nobody trains frontier models in orbit; the gigawatt filings are a historical curiosity of the mid-2020s. That outcome is not a failure. It is a useful industry that was oversold, which is the most common ending for a technology of this kind. I do not know which one happens. I hold the position I hold because the first outcome is worth many multiples and the second is not worth zero.