In 1956 a former trucking operator named Malcom McLean loaded 58 aluminium boxes onto a converted tanker in Newark and shipped them to Houston. The standardised shipping container that followed cut the cost of moving cargo across an ocean by something close to 90% — and in doing so it did not merely make shipping cheaper. It moved the binding constraint somewhere else, and everything downstream rearranged itself around the new bottleneck.
I keep returning to that story, because the same pattern is now playing out above our heads — and most of the commentary is staring at the wrong part of it.
1. The number everyone quotes
The figure that gets repeated, correctly, is the launch cost. NASA's own analysis put the Space Shuttle at roughly $54,500 to place a kilogram into low Earth orbit. SpaceX's Falcon 9 now advertises around $2,720 for the same kilogram — a reduction of about 20×. Starship, if it eventually does what it is built to do, is aiming below $100. Falcon 9 already flies every few days and carries the large majority of the world's commercial payloads. The chokepoint that kept space the preserve of governments for sixty years has, in commercial terms, been removed.
This is the genuine breakthrough of our era, and it earns the attention it gets. But the launch cost is the container ship. It is necessary, it is remarkable — and it is not where the most interesting value is going to settle.
2. The constraint has already moved
Once mass reaches orbit cheaply, the question stops being "can we afford to launch this" and becomes "what is now worth launching that was not before." The most serious answer on the table is compute. Data centres on Earth are hitting a wall, and the wall is power. In the right orbit a satellite sits in nearly constant sunlight, with no atmosphere to dim its panels and a cold vacuum to radiate heat into — abundant power and a free heat sink, which happen to be the two things the AI build-out is shortest of on the ground.
So the filings have started arriving, quickly. In early 2026 several companies filed plans, launched hardware or committed funding to put data centres in orbit within weeks of one another. SpaceX unveiled AI1, a solar-powered compute satellite with a 70-metre wingspan and a liquid radiator to shed waste heat into space, with prototypes slated for early 2027. Starcloud has filed for a constellation built to process data rather than relay it; Axiom has flown its first orbital compute nodes; and a clutch of startups are chasing edge inference on satellites that are up there anyway.
It is worth being honest about the physics, because the enthusiasm tends to outrun it. Cooling in vacuum is hard. Radiation degrades chips. Replacing failed hardware in orbit is expensive in a way that swapping a server in a rack is not. A sober reading is that meaningful orbital compute is a late-decade story, not a this-year one. None of that changes the underlying point: the bottleneck has moved off the launch pad, and the race is now about what you do once you are up there.
The cloud taught us this lesson once already. The servers became a commodity. The control plane did not.
3. The part almost nobody is solving
The contest is usually described as a land grab — who owns the route, who plants the most satellites. That mirrors the old shipping story, where the winners owned the ships and the ports. But the container era had a second act that mattered more than the hulls: logistics. The firms that captured the largest, most durable margins were the ones that worked out how to route, schedule and coordinate millions of containers across a planet so the right box arrived at the right place at the right time. The control layer outlived the steel.
Orbital compute has exactly this shape, only harder. A single satellite is not a data centre. A useful system is a constellation of nodes that are intermittently lit, thermally constrained, radiation-limited, linked by lasers and moving at about seven and a half kilometres a second relative to the ground. Deciding which workload runs on which node, when to compute and when to wait for sunlight, how to move data between satellites and back down without saturating the links, how to keep going when individual nodes fail — all of that is a distributed-systems problem of a kind we have not had to solve before. It is orchestration, not hardware, and it is the genuinely unsolved piece.
My bet is that the orchestration layer — the thing that turns a swarm of constrained satellites into something a customer can simply rent — will capture value out of all proportion to its share of the metal in orbit.
4. Who it is actually for
A cost collapse this large does not have a predetermined distribution. Cheap orbit could concentrate the world's compute and connectivity in two or three constellations owned by two or three companies. Or the same fall in cost could put real capability within reach of national space agencies, research consortia and operators across the Global South that were locked out entirely a decade ago.
That is not a technical question. It is a question of standards, access and how the orchestration layer is governed — and it is being decided now, quietly, in filings and frequency allocations rather than in headlines. At Maargin, that open, sovereign-by-design orchestration layer is exactly the part we are building toward.
The twenty-times fall in launch cost is the part that is finished. What we build on top of it — and who gets to build — is the part that is still open. The container was never really about the box; it was about everything the box made thinkable. Cheap orbit is the same. Getting there was the hard engineering. Deciding what it is for is the harder choice, and that one is still ours to make.
