Offshore Is Now Overhead
By Andrew Horton
15 June 2026
The next tax haven will orbit the planet every ninety minutes at 27,000 kilometres per hour. It will carry more frontier compute than most national economies can field on Earth and under current settings, it will fly the flag of whichever state sells its jurisdiction cheapest.
This architecture is already assembling. China's state-backed mega-constellations began deploying orbital AI supercomputing nodes into low Earth orbit over the past year, aiming for an eventual ecosystem of thousands of satellites delivering exascale processing in space. Western tech giants are close behind: Google's Project Suncatcher is preparing TPU-equipped orbital prototypes, Nvidia silicon is already operating aboard commercial testbeds, and tech founders openly signal that orbital facilities will eventually beat the escalating cost of terrestrial data centres.
The capital is moving. The engineering is maturing. Yet the policy debate trails far behind both.
The conventional framing asks whether space-based data centres will escape regulation, taxation, and legal accountability - whether orbit becomes a lawless commons for rogue compute. The true danger lies not in the absence of law, but in the law that already governs orbit.
The Outer Space Treaty of 1967 fills the jurisdictional vacuum completely and that is precisely the problem. Article VI imposes on states "international responsibility for national activities in outer space," explicitly including the actions of private companies. Article VIII grants the state of registry continuing "jurisdiction and control" over the objects it launches.
Read carefully, the treaty binds every orbital platform to a sovereign. Crucially, it also allows that platform to choose its sovereign.
History has run this experiment before. Maritime law assigns primary jurisdiction to the flag state. Within decades of that legal codification, shipowners discovered that flags could be purchased. Today, more than half of global shipping tonnage sails under flags of convenience. Panama, Liberia, and the Marshall Islands - nations with negligible domestic maritime infrastructure - govern the world's commercial fleet because they competed on the single dimension that corporate operators valued: the lightness of their oversight.
The same auction is about to open five hundred kilometres up.
A multinational tech operator seeking to train an unaligned frontier model simply needs to register its orbital platform with a nation offering zero corporation tax, token data obligations, lax cybersecurity rules, and guaranteed regulatory silence. The 1967 treaty does the rest, legally bonding the platform to that registry. The privacy statutes, AI assurance frameworks, and critical infrastructure regimes of every other nation - from the European Union's AI Act to Washington's export controls to Canberra's security legislation - stop, quite literally, at the Kármán line.
This is the reality waiting to confront governments around the world: orbital regulatory arbitrage will arrive as a commercial product, sold by sovereign micro-states, using a Cold War treaty as its terms of trade. The first great strategic transaction of the intelligence age may not be the purchase of compute, but the purchase of jurisdiction.
Treasuries will see this as a question of lost revenue. National security agencies should see it as a question of lost control.
Consider what convenience-flagged orbital compute enables: advanced AI models trained and operated entirely beyond allied jurisdiction, processing data harvested globally, accountable only to a registry state that asked for a processing fee and nothing else. It enables intelligence collection platforms indistinguishable from commercial infrastructure, and autonomous cyber capabilities - the kind frontier laboratories are currently restricting on Earth hosted entirely beyond the reach of export controls, deployment safeguards, or domestic subpoenas.
The "harvest now, decrypt later" doctrine already assumes adversaries are stockpiling encrypted allied data against the arrival of quantum capability. Orbital compute flying a flag of convenience gives that doctrine a permanent, sovereign-proof sanctuary.
Furthermore, when convenience-flagged infrastructure causes kinetic or financial harm a manipulated market, a compromised energy grid, a hostile model deployed at machine speed - attribution collapses into a hall of mirrors. A satellite launched from one state, financed from a second, operated from a third, and registered in a fourth allows grey-zone actors to exploit the same legal ambiguities they have used at sea for a decade. Orbit offers deniability as a core design feature.
Nations that rigorously police foreign data centres on their own soil will soon find untrusted infrastructure operating, perfectly legally, directly above them.
There is, however, a strategically significant advantage for democracies: orbital compute depends upon infrastructure they collectively control.
Photons must come down. Every orbital data centre requires ground stations, optical downlinks, and terrestrial fibre networks to deliver value to paying customers. The ground segment is to orbital compute what cable landing stations are to the global internet and governments already possess the legislative power to license those.
The premium ground-segment geography of the planet characterised by electromagnetic quiet zones, deep-space communications heritage, and critical polar and southern-hemisphere coverage sits overwhelmingly within allied territory. It stretches from Australia and New Zealand to Norway, Canada, Japan, and the continental United States.
Jurisdiction ends at the atmosphere, but strategic leverage extends all the way up. Three moves should follow immediately:
First, build a trusted flag regime before the convenience flags fly. The G7, the OECD, and advanced mini lateral frameworks like AUKUS Pillar II should establish a Trusted Orbital Compute standard. This registry framework would mandate strict baselines for taxation, data sovereignty, AI assurance, and security auditing, with mutual recognition across like-minded states. Maritime history teaches that port-state control, applied late, clawed back only a fraction of what flag-shopping surrendered. Orbit offers the rare chance to dictate the standard before the race to the bottom begins.
Second, extend export controls to the launchpad. Advanced AI accelerators are already strictly controlled at national borders; they must be controlled at the atmospheric one. Allied-origin frontier silicon should reach orbit exclusively under trusted flags. Advanced chips are the one critical input the arbitrageurs must source from the very nations they seek to escape.
Third, make landing rights the primary instrument of enforcement. Critical infrastructure and telecommunications frameworks must be updated to capture orbital compute serving domestic markets. Downlink and ground-station licences should be conditional on strict jurisdictional compliance exactly how undersea cable landing licences condition market access today. Convenience-flagged compute must be treated as untrusted infrastructure: denied landing rights and priced out of every allied market it hopes to monetise.
Acting on this requires political recognition rather than new rockets. The contest has already started. The Outer Space Treaty has pre-written the rules, and the nations holding the ground segment hold a stronger hand than they realise provided they play it before the first orbital registries are sold.
The age of offshore finance cost governments revenue, and they spent forty years trying to recover it. The age of offshore compute puts something far harder to recover on the table: authority over the most consequential cognitive machines ever built.
If a hostile or unaligned AI system one day operates beyond allied jurisdiction from a platform flying an orbital flag of convenience, the defining political question will not be who launched it, but why the nations with everything to lose allowed the strategic high ground to be sold in the first place.