How to read this appendix. It contains no valuations, no round sizes and no dated financials. That is deliberate: those numbers move every quarter and would make the book wrong within a year of printing. What follows is structural judgement, what each company does, why it might matter to orbital compute, where it is fragile, and the single event that would most change the assessment. Those things age in years rather than months. Everything here is the author’s opinion as of 27 July 2026, informed by his own fund’s research, and should be read alongside the disclosure at the front of this book. Companies are grouped by the chapter that covers their layer. One organising idea. Chapter 18 dates the moment orbital compute clears the commodity market to the early 2030s, and identifies the mechanism that gets it there: premium buyers funding the first nine doublings of manufacturing experience. The right question to ask of every company below is therefore not “is this profitable today” but “is this company positioned to be on the curve when it crosses, and can it survive the descent.” Companies with premium revenue now and manufacturing leverage later are the ones that compound through it.
Launch and mobility (Chapter 9) SpaceX. Sets the price everyone models against, holds the only publicly disclosed heavy-lift contract price, and, having filed for a very large compute constellation and merged with an AI company, competes directly with the operators who depend on it. There is no clean way to hold a view on this sector without holding a view on SpaceX. Fragile: nothing structurally; the risk it presents is to everyone else. Watch: whether Starship flies paying customers at contracted rather than target prices, and turnaround time between heavy flights. Rocket Lab. The credible second heavy provider, and since early 2026 also a power-layer entrant with silicon arrays aimed explicitly at orbital data centres. Its strategic significance exceeds its market share: a sector priced off one supplier has a single point of commercial failure. Fragile: competing in two layers against much larger incumbents in each. Watch: whether the silicon-array bet wins design slots against multi-junction suppliers, that contest decides Chapter 8’s open architectural question. Impulse Space. Last-mile orbital mobility, founded by SpaceX’s first employee, revenue-generating with flight heritage. Chapters 5 and 6 push compute constellations into specific awkward orbits, and something has to put them there. Fragile: the most mature private name in the theme is also the one where expectations are highest. Watch: whether compute constellations start buying orbital transfer as a line item rather than accepting the orbit the rocket gives them.
Power (Chapter 10) K2 Space. The satellite is the power system, a large bus designed around high power, high-voltage avionics and a high-power thruster, sized for what heavy lift now permits. The cleanest structural expression of the view that power, not production rate, is the durable differentiator in buses. Fragile: a competing supplier on the same government programme is direct evidence the layer is contestable. Watch: on-orbit confirmation of high-voltage bus performance at rated power. Star Catcher. Concentrates and beams solar energy onto client spacecraft’s existing arrays, no retrofit, no custom receiver. Shared infrastructure that sells to every operator rather than requiring you to pick one, with contracted customers ahead of first flight and a repeat founder who previously built and sold an in-space manufacturing company. Fragile: small team against a capital-intensive constellation; the hard version of the physics has never been done. Watch: a spacecraft-to-spacecraft beaming demonstration on orbit, it would create a category. Zeno Power. Radioisotope power. Terrible specific power, irrelevant for megawatts, and exactly right for anything that must survive the dark. The moat is not the physics, which is sixty years old but a recycled-isotope fuel supply chain that venture money cannot compress. Fragile: revenue is defence and civil space, not orbital compute; it is a dual-use holding, not a pure-play. Watch: whether the fuel supply agreements convert into repeat unit orders. Aetherflux. Generates power and consumes it in its own compute payload, the only name owning both ends. Strongest founder record in the group and, at the time of writing, nothing demonstrated in orbit. That combination is precisely why it should be held smaller than its narrative justifies. Fragile: weakest ratio of expectation to evidence in the sector. Watch: the first power-beaming demonstration from low Earth orbit.
Thermal (Chapter 11) Sophia Space. The closest thing to an investable pure-play in the layer that most determines whether the sector works. Its tile architecture makes thermal integration the product rather than a subsystem. Fragile: early stage, and strategically ambiguous, the same tile can be sold to operators or assembled into a competing data centre, and only the first configuration is a supplier business. Watch: a signed supply agreement with a named operator. Absent that, it is competing with its customers. Spacebilt. Thermal tiles already supplying commercial orbital data-centre nodes, the best technology fit in the layer and not investable at venture scale. Track it as an acquisition target rather than a position. Watch: who buys it. Celeroton and the turbomachinery suppliers. Oil-free high-speed compressors for cryocoolers and heat pumps: the exact hardware Chapter 11’s heat-pump argument requires. European industrial suppliers, no venture round. Watch: anyone packaging this into a space-rated heat pump, the highest-leverage unbuilt product in this book.
Compute platform (Chapter 13) Starcloud. Flight heritage leader: flew a commercial accelerator, trained a model in orbit, and its second vehicle is the first serious test of direct-to-chip liquid cooling into deployable radiators at rack scale with named cloud customers. The single most informative company in the sector whether or not you own it. Fragile: constellation scale requires capital far beyond what has been raised, and reported strategic interest from the dominant launch provider is both an exit path and a dependency. Watch: paying workloads running on orbit, and any published price for orbital compute. Google (Project Suncatcher). The only organisation designing its own silicon with orbit in mind, and the only one publishing its physics. Not investable at this scale; its progress is nonetheless the strongest de-risking event the sector has had. Watch: the prototype satellites, and any published on-orbit single-event data. Axiom Space. The hosted-platform route, with data-centre nodes already on orbit and a compute business crosssubsidised by existing revenue, a genuinely different risk profile in a sector full of pre-revenue narratives. Fragile: leadership churn and a binary government station selection sit underneath it. Watch: whether third parties choose to host compute on a station rather than fly their own free-flyer. Loft Orbital. Flies other people’s payloads on standardised shared satellites and lets customers run inference on hardware already in orbit, the lowest-friction route to orbital compute for a customer who does not want to own a satellite. A sovereign manufacturing joint venture gives it a distribution channel the US-centric names lack. Fragile: a shared-platform business is a margin business. Watch: how much of its bookings become compute rather than sensing.
Apex Space. Productised buses built ahead of demand, running toward high annual production. Excellent, and held small precisely because rate manufacturing is a real moat against incumbents and no moat at all against the next well-funded productiser. Its demand is proliferated national-security constellations, not orbital compute. Watch: whether a compute operator standardises on a bought bus rather than building its own.
Network (Chapter 14) Kepler Communications. A deployed optical relay constellation running as an IP mesh, with compute and storage on the relay satellites themselves. Simultaneously the backhaul for orbital compute and a distributed compute platform of its own and Chapter 14 showed the relay layer’s value is duty cycle, a twentyfold multiplier on the usefulness of identical hardware. Fragile: if the dominant operator opens its laser mesh as a commercial service, pricing compresses immediately. Watch: exactly that. Skyloom and the terminal suppliers. Microradian pointing mechanisms with lasers attached. Unglamorous, necessary, and sold to everyone. Watch: whether terminal supply becomes a bottleneck as constellation counts rise.
Outside the roster
ADA Space, Zhejiang Lab and the Chinese programmes (Chapter 17) are excluded by mandate rather than by capability, and they currently have more dedicated compute hardware in orbit than anyone else. The relevant question for a Western investor is not whether to buy them but what a state-backed constellation that does not need a commercial return does to price. Thales Alenia Space and the ASCEND consortium (Chapter 17) have produced the most rigorous public feasibility analysis in the field, attached to a 2050 timeline. Europe’s realistic roles here are customer, component supplier and regulator.
The two vacancies
Two positions in this appendix do not exist and should: 1. A megawatt-class thermal transport company with qualified two-phase hardware, selling to operators rather than competing with them. 2. An on-orbit hardware refresh business, servicing, module replacement, or a modular architecture sold as a refresh service. Chapters 11 and 15 argue that these are the sector’s two unserved constraints. If either company is founded and funded well, it should be assumed to be the most valuable new entrant in this appendix.
End of book.