The old idea
In 1960, Freeman Dyson published a two-page paper in Science proposing that an advanced civilisation would eventually capture the full output of its star, and that we could search for such civilisations by looking for the infrared they radiate. Read that again, because it is the most under-appreciated sentence in the history of this subject. Dyson’s insight was not the sphere. It was the waste heat. He understood immediately that any structure gathering enormous energy must dump an almost equally enormous amount of heat, and that the dumping not the gathering would be the visible signature. The first person to think rigorously about civilisation-scale computing around a star concluded that you would find it by its radiators. Science fiction then spent sixty years forgetting this. Almost no spacecraft in film has radiators. They have engines, weapons and windows, and no way at all to get rid of the heat those imply. The omission is so consistent that it functions as a useful test: when someone shows you a rendering of an orbital data centre, look for the radiators, and if they are not obviously the largest thing in the picture, the picture is fiction.
The physics
Chapter 5 did the arithmetic; this chapter needs only its two conclusions. Rejection scales as the fourth power of temperature, so running hotter is the dominant lever. And radiator area is not really the hard part, getting a megawatt of heat from the silicon to the panel is. There is a third lever, largely unexploited, that follows directly from the fourth-power law: actively pumping heat uphill. A heat pump consumes electrical power to raise the rejection temperature above the source temperature. It sounds wasteful, you are adding heat to get rid of heat but because area falls as T⁴, raising the radiator from 27 °C to around 100 °C can cut radiator area by roughly sixty percent and system mass by roughly a third even after accounting for the pump’s own power and its own waste heat.13 In a vehicle where radiators are the largest mass item, that is an enormous trade, and it is available today in principle. That it is not yet a product tells you how immature this layer is.
What breaks and the number that frames the whole sector
Figure 11.1 — The gap that decides the sector
The International Space Station rejects 70 kilowatts, through two independent pumped ammonia loops and six deployed radiator assemblies.14 That is the largest active thermal control system ever flown by anyone. Reference Node A, a single modest compute satellite, half a rack of accelerators, needs 100 kW. The entrylevel vehicle in this sector exceeds the largest thermal system in the history of spaceflight. The megawatt node needs fourteen times the ISS. The gigawatt-scale ambitions in the sector’s public filings need tens of thousands of times. One more comparison, because it cuts both ways. The ISS rejects its 70 kW from roughly 420 square metres of panel, about 166 watts per square metre. Chapter 5’s numbers give a data centre radiator at 58 °C roughly twice that per square metre. The difference is simply temperature: the ISS runs its loops near room temperature because it contains humans. A machine that contains no humans can run hot, and that is the sector’s single largest structural advantage over every thermal system previously flown. The physics is on the sector’s side here. The hardware is not there yet. What specifically is missing: Two-phase pumped transport at megawatt scale. Heat pipes top out in the low kilowatts. Single-phase pumped loops are proven to tens of kilowatts. Two-phase is the right answer above a megawatt and has no flight heritage at that scale, and managing boiling flow in microgravity, where vapour does not rise is genuinely
hard. Deployment of a thousand square metres of thin panel, reliably, once. Micrometeoroid and debris puncture of the largest, thinnest, fluid-filled surface on the vehicle. The ISS radiators were designed with debris protection precisely because this is a certainty over years, not a possibility. Coating degradation, which attacks emissivity, which sits inside the fourth-power law.
Who is attacking it
Here the chapter becomes unusual, because the honest answer is: almost nobody you can invest in. Spacebilt has the best technology fit I have found, thermal tiles rejecting heat to deep space, already supplying orbital data-centre nodes for a commercial station operator. It has no visible venture round, no accessible cap table, and no mark. It is a supplier and an acquisition target, not an investment. Sophia Space is the closest thing to an investable pure-play. Its TILE architecture makes thermal integration the product rather than a subsystem, modules combining solar generation on one face with radiative cooling on the other, racked together into compute. It is seed-stage with first customer deliveries targeted for 2028.12 The strategic ambiguity is worth naming: the same tile can be sold to orbital data centre operators or assembled into a competing one, and only the first configuration is a supplier business. Celeroton and companies like it, oil-free high-speed turbomachinery for cryocoolers and heat pumps are exactly the technology the heat-pump argument above requires. They are European industrial suppliers with no venture round to buy into.
The investment stance, and why the emptiness is the finding
The layer that most determines whether this sector works has the fewest investable companies in it. That is not an oversight on my part; it is a structural observation, and it admits three readings. The optimistic reading: this is a gap, and gaps get filled. A well-founded thermal company with a megawatt-class two-phase loop would have every operator in the sector as a customer. The realistic reading: thermal is so central to the vehicle that operators will build it in-house rather than buy it. Cooling is not a component you bolt on; it shapes the whole spacecraft. If that is right, the value in this layer accrues to the operators, and the correct exposure is through them. The pessimistic reading: the layer is empty because the problem is harder than the sector admits, and the people who understand it best are not starting companies to solve it. I hold the middle reading, and I act on it by reserving capital against the layer conditionally rather than committing to it, funded only against a demonstrated supply agreement with a named operator, because a thermal company that is not selling to operators is competing with them.
What to watch
The first commercial deployable radiator and direct-to-chip liquid loop operating on orbit at rack scale. This is the nearest genuine test of the architecture and it is imminent. Anyone announcing a pumped two-phase loop qualified above a few hundred kilowatts. That announcement, from anyone, reprices the whole sector. Anyone shipping a space-rated heat pump. It is the highest-leverage unbuilt product in this book. Conversely: an operator quietly raising its published radiator area per megawatt is telling you the ΔT chain in Figure 8.3 is worse in practice than in analysis.
Chapters 12 to 17, radiation-tolerant compute, the compute platform itself, communications and ground segment, in-space assembly and servicing, debris and insurance, and innovation outside the United States, follow in the next instalment.