Nearly everything in this book so far has been American. That is a fair reflection of where the private capital and the launch capacity sit, and it is also a distortion, because the country with the most flown orbital compute hardware is not the United States.
China
China treats orbital computing as infrastructure policy rather than as a venture opportunity, and it started earlier. In May 2025 a Long March 2D placed twelve satellites in orbit as the first batch of the Three-Body Computing Constellation, led by Zhejiang Lab with the Chengdu company ADA Space, described as the world’s first dedicated orbital computing constellation, with a combined capability of 5 peta-operations per second and 30 terabytes of onboard storage across the twelve.18 The stated plan reaches 100 satellites by 2027, within a wider “Star-Compute” programme targeting 2,800 satellites and a combined 1,000 POPS.19 It is not the only one. A 300-satellite Tiansuan constellation runs out of Beijing University of Posts and Telecommunications; a further roughly 1,000-satellite network, Xingshu, was announced in 2026 by Fudan University and a Shanghai partner; Beijing established a dedicated space-computing innovation centre in mid-2026 whose priority areas include heat-tolerant, radiation-tolerant chips designed specifically for orbit.20 Two things in the Chinese approach are worth taking seriously on the engineering merits. The first is the stated rationale, which is not energy at all. Chinese researchers frame the problem as downlink: that a very large share of the data generated in orbit is never transmitted or processed because bandwidth to the ground is insufficient.18 That is precisely Chapter 7’s strongest near-term argument, put the compute next to the sensor and it is notable that the programme with the most hardware in orbit is the one pursuing the workload the physics most clearly supports. The second is that Chinese teams are converging on the same COTS-plus-software answer as everyone else: commercial chips rather than slow rad-hard parts, with software-defined fault tolerance and ground-based digital twins doing the reliability work.20 When independent programmes under different constraints arrive at the same architecture, that architecture is probably right.
Figure 17.1 — Filings are cheap. Hardware is not
.
I hold no Chinese exposure, by mandate, and I would note that this is a decision about jurisdiction and not about capability. The relevant question for a Western investor is not whether to buy these programmes but what they do to price. A state-backed constellation that does not need to earn a commercial return can set the price of orbital inference in the markets it serves, in the same way state-backed manufacturing has repeatedly done in terrestrial industries.
Europe
Europe’s contribution has been a serious feasibility study rather than hardware. ASCEND, coordinated by Thales Alenia Space with a consortium including ArianeGroup, Airbus, DLR, HPE and Orange was funded by the European Commission under Horizon Europe and concluded in 2024 that space-based data centres are technically, economically and environmentally feasible, framed around EU net-zero goals and data sovereignty. Its stated ambition is one gigawatt deployed before 2050, against a projected 23 GW terrestrial data centre market by 2030, with modular infrastructure assembled in orbit by robotic systems.21 Read that carefully and Europe’s position becomes clear. The target date is 2050. The American programmes are talking about 2027. Europe has produced the most rigorous public feasibility analysis in the field and paired it with a timeline a generation behind the people building hardware. That is not merely a criticism. Europe’s realistic roles are as a customer, sovereign, regulated, willing to pay for data jurisdiction, as a supplier of optical terminals, robotics and thermal engineering, and as a regulator whose rules on data location will shape where orbital capacity can legally serve. All three matter. None of them is a constellation.
Russia, and everyone else
Russia is not a factor in this sector. Its launch industry has contracted sharply and there is no visible orbital compute programme. The Gulf states are more interesting than their space heritage suggests, because they combine three things this sector needs: sovereign capital with a long horizon, cheap energy and a strategic interest in owning digital infrastructure rather than renting it. Gulf sovereign funds already appear on the cap tables of Western station and compute companies, and regional manufacturing joint ventures are being built. Japan and India appear as component suppliers and investors rather than as programme leaders.
The stance, and the wrinkle worth thinking about
The sector will not remain a single market. Data jurisdiction is the reason: a data centre in orbit is under the jurisdiction of the state that licensed the satellite, not of the territory it flies over, which makes orbital compute an unusually clean instrument of digital sovereignty. That argues for several regional constellations rather than one global cloud, which is good for suppliers who sell to everyone and harder for operators hoping for winnertake-all economics. And a genuine legal oddity that I have not seen resolved anywhere: advanced AI accelerators are subject to export controls in most jurisdictions. What happens when the accelerator is launched into orbit and then sold, leased, or served to a customer in a restricted country? Where the chip physically is, who controls it, and which regime applies are not settled questions. They will be, and the answer will shape this industry more than several of the engineering problems in this book.
What to watch
Whether the Chinese constellation reaches its stated 2027 scale, and whether it publishes performance. Any European decision to fund ASCEND beyond study phase, which would pull its timeline forward by a decade. The first export-control ruling on accelerators in orbit.
Part IV is complete. Part V builds the full economic model, cost per GPU-hour, satellites per gigawatt, and what Earth gets back and Part VI sets the horizon and the falsifiers.