A Million Satellites for the Machine: The Orbital Compute Land Rush
The United States Federal Communications Commission, the FCC, is now holding applications for more than one million, one hundred and thirty-nine thousand orbital compute satellites. Three companies filed them between January and June this year. The total number of operational satellites of every kind currently in orbit, after seventy years of spaceflight, is about ten thousand. The filed number is roughly one hundred and fourteen times the entire working population of objects in space today.
No treaty caps it. No statute caps it. No regulator anywhere holds a mandate to reject one of these applications on the single ground that it is too big.
That is the story underneath the headlines about orbital data centres. Not the engineering, which works. Not the money, which is arriving. The story is that the assets are being filed at a scale governance has no answer for, and that every month they age through regulatory review is a month closer to becoming permanent.
What landed on the docket this year
Three filings make up that 1.14 million figure. SpaceX filed for an Orbital Data Center System on 30 January, and the FCC Space Bureau accepted it for review on 4 February under document DA-26-113. The application seeks authorisation for up to one million solar-powered satellites between five hundred and two thousand kilometres up, each carrying onboard machine-learning accelerators and optical links to its neighbours. SpaceX also asked the FCC to waive the standard deployment milestones, the usual rule that half a constellation must be flying within six years, citing the novel nature of the request.
Blue Origin filed in March for Project Sunrise: up to fifty-one thousand, six hundred satellites in sun-synchronous orbit for in-space computing. Starcloud, the company that ran Google's Gemini model on an Nvidia H100 chip in orbit last year, filed for eighty-eight thousand. These are not letters of intent. Each has been formally accepted by the FCC and entered into review.
Then, on roughly 24 June, the picture changed shape. Amazon, Iridium, Globalstar and Telesat formed a new Washington trade body, SpaceConnect, to coordinate positions on licensing, spectrum and the next World Radiocommunication Conference. SpaceX, which operates more than ten thousand non-geostationary satellites, around twenty-two times the combined fleets of all four founding members, was not invited. The new body appointed a former head of the United States telecommunications policy agency as its executive director and a former inaugural chief of the FCC Space Bureau as its general counsel. Its leadership understands every procedural lever the system has. Read structurally, and the membership list reads clearly enough, SpaceConnect is a regulatory counterweight to a single dominant operator, formed two years before the spectrum coordination that matters most.
There is institutional friction inside the docket too. NASA filed a formal objection to Project Sunrise in early May, on the grounds that the proposed orbital shell overlaps human spaceflight paths and that the launch market cannot realistically absorb the satellites on the proposed timeline. The objection does not block anything. It enters weight into a review process that has no authority to reject on scale alone.
The pirate radio problem, in orbit
This series has used the pirate radio parallel before, and it has never fit more precisely. In the nineteen sixties, radio operators anchored ships in international waters and broadcast into national markets from outside any broadcaster's jurisdiction. They did not need to win a legal argument. They needed to be in place before the regulators arrived. By the time the law caught up, the audience, the advertising and the habit were established facts.
The orbital compute filings are the same move at a different altitude. File at million-satellite scale now, occupy the orbital slots and spectrum positions, and let the assets become harder to remove with every passing month. The mechanism is jurisdictional arbitrage, and it works because the governance gap is real and specific.
Consider what is actually missing. The Outer Space Treaty of 1967 assigns responsibility and liability to launch states and prohibits any nation from claiming territory in space, but it says nothing about data processing. The FCC's five-year deorbit rule governs what happens to a satellite after its mission ends; it is a disposal obligation, not a launch cap. The International Telecommunication Union coordinates spectrum and orbital positions but does not limit how many satellites one applicant may fly. United Nations debris-mitigation guidelines exist, and they are voluntary. At no point in that stack is there a body that can look at an application for a million satellites and say no on the basis of size.
The scale matters because the orbital environment is already crowded. The European Space Agency's debris modelling tracks roughly fifty thousand objects larger than ten centimetres, around 1.2 million between one and ten centimetres, and on the order of one hundred and forty million smaller fragments. Adding more than a million satellites from three operators alone, over operational lives of fifteen years or longer, in altitude bands already carrying debris, raises the risk of a collision cascade, the self-sustaining chain that can render an orbital zone unusable. That risk has a name in the literature and no corresponding answer in governance.
Why file at this scale at all? Because the economics, on paper, point upward. SpaceX has assembled the full vertical stack, what this series has called Corporate Kardashev: its own silicon, its own model and compute, its own launch, and now an orbital data-centre constellation. The pull is the energy ceiling. Terrestrial data centres run into hard limits: cooling water, grid capacity, planning consent. Orbit sidesteps all three with near-constant solar power and the vacuum as a heat sink. The planned first-generation satellite, described around the SpaceX public listing, is designed to carry a compute payload in the order of one hundred and fifty kilowatts at peak, with a large deployable liquid radiator and a module meant to be swapped across chip generations. It is engineering in service of a single proposition: move the workload to where the constraints do not apply.
The reality check sits in the launch cost. A Google research paper, reported by CNBC in June, put the threshold for orbital compute to compete with the ground at roughly two hundred United States dollars per kilogram to orbit by the mid-2030s. That figure rests on a single source chain and should be read as one analyst benchmark, not settled fact. Current rideshare launch costs sit far above it, in the thousands of dollars per kilogram. The current production benchmark is smaller still: Kepler Communications runs forty processors across ten satellites for paying customers today. The filed constellations are a scale-up of roughly one hundred thousand times that. The engineering question is not whether orbital compute works. It is whether it works at the filed scale, under economics that do not yet exist, governed by frameworks no one has drafted.
That is the regulatory clock, and it is the one driving this story. The filings are racing ahead of any framework that could answer them, exactly as the pirate broadcasters raced ahead of the law.
What it means in New Zealand
Here is where the abstraction becomes a board-level question for organisations in Aotearoa New Zealand.
New Zealand has built sensible legislation for the part of this it can reach. The ground-based space infrastructure regime, with its authorisation deadline of 29 July this year, a date this series has tracked before and one that falls inside this article's publication window, governs the hardware that tracks, controls and transfers data to and from spacecraft. It governs the ground segment. It does not, and was never written to, govern what happens aboard the satellite.
That boundary is the whole point. Under the Outer Space Treaty, a satellite is governed by the law of the state that registered it. Data processed aboard a United States-registered compute satellite is therefore processed under United States federal law, not the New Zealand Privacy Act 2020, not the European General Data Protection Regulation, not any New Zealand statutory instrument. A New Zealander's personal data, routed through a local ground station to an orbital compute node, leaves New Zealand jurisdiction at the point of uplink and returns only on downlink. This is not a criticism of the ground-segment legislation, which addresses a specific national-security risk in ground infrastructure and addresses it well. It is a factual description of where the law stops. There is, at present, no New Zealand governance lever over the orbital processing layer.
Picture a technology director at a mid-sized New Zealand financial services firm reviewing a cloud-strategy paper. One recommendation is to evaluate orbital compute from a United States provider by 2028 for latency-sensitive inference. She asks her privacy officer where the Privacy Act obligation sits if personal data is processed in orbit. He checks the Act: no provision for orbital processing. He checks the ground-segment regime: it covers the uplink and downlink, not what happens in between. He checks the Outer Space Treaty: processing aboard a United States-registered satellite happens under United States law. The recommendation goes through. The workload migrates. Nobody has done anything wrong. There is simply no law yet that covers this.
The dependency question compounds it. As this series covered a fortnight ago, with no operationally available competitive satellite alternative for around two years, the New Zealand market is already concentrated at the connectivity layer; one provider holds roughly one-fifth of the satellite broadband market, on the order of fifty-seven thousand connections by 2025 to 2026 reporting. Now look at the compute layer. Amazon has committed several billion dollars over the life of its Auckland cloud region, with government among its early customers and a pledge to train tens of thousands of New Zealanders. The same company is an emerging orbital broadband competitor and a founding anchor of SpaceConnect, the body that will shape orbital spectrum and licensing positions at the World Radiocommunication Conference. New Zealand's infrastructure dependency on a single commercial entity now spans terrestrial cloud, satellite connectivity and the lobbying architecture for orbital governance. That is the sovereignty exposure this series and Gen AI Tuesday have traced through the CLOUD Act and data residency, lifted one layer higher, off the planet.
New Zealand has, by some reporting, raised its own marine-consent threshold for launches, a signal that it intends to grow its launch profile. Treat that as reported rather than confirmed. Either way, it addresses launch, not the orbital processing layer, and the country runs no orbital compute programme of its own.
What to watch
Three markers will tell you how this resolves. The first is the World Radiocommunication Conference in 2027, the arena where spectrum coordination for the next generation of constellations is decided, and the reason SpaceConnect formed now rather than later. The second is the NASA objection, which sets the first formal governmental record of safety concern from inside the United States executive branch and will test whether due-process weight can matter in a review with no authority to reject on scale. The third is the prototypes: first launches are targeted for 2027. Watch whether the timeline holds, because the filings are a bet that occupation precedes regulation.
The defence and sovereignty implications run directly through this. The same orbital compute being filed commercially is what allied military and intelligence systems will increasingly rely on. A bipartisan United States Senate bill, the NEW HORIZON Act, directs a pilot at the Pentagon's Defence Innovation Unit, the DIU, to evaluate commercial orbital data centres for military and intelligence operations and to study their resilience and vulnerability. Set that beside dedicated military constellations such as Starshield. When sovereign defence functions come to depend on processing owned by a handful of commercial operators under one jurisdiction, capability and dependency arrive together. For a country whose space contribution is its ground segment and its allied data-sharing arrangements, the structural question is not whether to use orbital compute. It is whose law governs the processing layer when the workload is overhead, and whether the answer was ever a national choice.
The pirates did not win an argument. They won a position, and the law arrived to find the facts already set. A million satellites have been filed for the machine. The governance question is not whether they should fly. It is whether anyone will decide the rules before the orbits fill.
If you have mapped where your organisation's most sensitive workloads are actually processed, have you found the point where they leave New Zealand jurisdiction, and what did you decide to do once you saw it?
The views expressed in this article are entirely my own, informed by more than 30 years of professional experience in architecture, security, and technology leadership in New Zealand. They do not represent the views of my employer, any government agency, or the New Zealand government. My commentary on legislation and policy is analytical, drawing on publicly available sources and my professional expertise in architecture, security, and AI governance. I follow the Public Service Commissioner's Code of Conduct for the Public Sector and social media guidance.
Andreas Hamberger is a New Zealand leader in Architecture & Security and Associate Member of the Institute of Directors. Space Mafia examines the sovereignty implications of orbital compute infrastructure.
I use AI tools, including Sudowrite, Claude, Perplexity AI, DeepSeek AI, ChatGPT, Grok, Copilot, Openart and Gemini, as deliberate production tools, not ghostwriters. This is consistent with my position: AI amplifies human judgement; it does not replace it. The frameworks, arguments, and editorial decisions in this series are original work. AI accelerated the process. The thinking is mine.
[1] Federal Communications Commission. "DA-26-113: SpaceX Orbital Data Center System, accepted for filing." 4 February 2026. https://www.fcc.gov
[2] SpaceNews. "SpaceX, Blue Origin and Starcloud file for orbital compute constellations." 2026. https://spacenews.com
[3] SatNews. "NASA files formal FCC objection to Blue Origin Project Sunrise." 5 May 2026. https://www.satnews.com
[4] Payload Space. "Amazon, Iridium, Globalstar and Telesat form SpaceConnect; SpaceX not invited." June 2026. https://payloadspace.com
[5] CNBC. "Orbital data centres and the economics of the Google Suncatcher research paper." 21 June 2026. https://www.cnbc.com
[6] European Space Agency. "ESA Space Environment Report and Space Debris Model." 2026. https://www.esa.int
[7] United Nations Office for Outer Space Affairs. "Treaty on Principles Governing the Activities of States in the Exploration and Use of Outer Space (Outer Space Treaty)." 1967. https://www.unoosa.org
[8] Bell Gully. "New Zealand ground-based space infrastructure: the authorisation regime and the 29 July 2026 deadline." June 2026. https://www.bellgully.com
[9] Ministry of Business, Innovation and Employment. "Ground-based space infrastructure: guidance for operators." 2026. https://www.mbie.govt.nz
[10] Air and Space Forces Magazine. "The NEW HORIZON Act and the Defence Innovation Unit pilot for commercial orbital data centres." June 2026. https://www.airandspaceforces.com