Who Was Paying for Mars?

If you bought a Starlink Mini, you were told what it was for. Not broadband at a fixed address — there were cheaper terminals for that. The Mini was the one you took with you: the boat, the truck, the farm at the end of a dirt track, the fortnight in a valley where the nearest mast is forty kilometres away and pointing the wrong direction. A hundred and fifty countries, the marketing said. Sixty days per trip. Take it anywhere.

As of 17 August 2026, it works at your house.

That is not a tightening of an allowance. It is the removal of the entire capability from every plan except one costing several times as much. The satellites did not move. The physics did not change. The Mini in your hand can still do exactly what it could do in July; it has simply been switched off at the billing layer and offered back to you at a higher price.

The date is the story. On 12 June 2026, SpaceX listed on Nasdaq at a valuation of $1.77 trillion. Thirty-two days later, the restriction was announced. To see why those two events belong in the same sentence, you have to understand what Starlink's customers had unknowingly been paying for.

I. The cash engine

SpaceX's S-1 prospectus made public, for the first time, the internal architecture of a company that had spent two decades declining to explain itself. Starlink generated $11.4 billion in 2025 — roughly 61% of total company revenue — up from $7.6 billion the year before. Active customers had passed 10.3 million across 160 countries, more than doubling in fifteen months. Margins on the satellite business ran above 60% at the EBITDA line.

Set against that: an accumulated deficit of $41.3 billion.

Read those figures together and the shape of the enterprise becomes clear. Starlink is not one division among several. It is the mechanism by which everything else is paid for. Falcon 9 revenue is real, but launch is a business with customers, competitors and pricing pressure. Starship has consumed capital on a scale with no revenue against it at all, and will not have any for years. The computing and AI ambitions consume more. The satellite broadband subsidiary, with its monthly direct debits and its extraordinary margins, is what covers the gap.

Which means that every Starlink subscriber — the farmer in Cumbria, the crew on a trawler, the man with a Mini in the back of a Hilux — has been supplying venture capital to the most ambitious engineering programme in modern history. Not by choice, and without a share certificate to show for it, but in substance that is what the arrangement was. Consumer broadband bills were being converted into reusable heavy-lift rocketry and an eventual attempt at Mars.

There is nothing scandalous in that. It was arguably the most productive use of a broadband bill in the history of broadband bills. It is also precisely the arrangement that a public listing destroys.

II. What changes when the shares start trading

While SpaceX was private, the cross-subsidy required no justification. A controlling founder decided that satellite revenue would fund Starship, and nobody had standing to object. The capital allocation was, in the most literal sense, unaccountable — and that unaccountability is what made it possible.

The listing changes the character of that decision in two directions at once, and both press on the same subsidiary.

The first pressure is on Starlink to yield more. A valuation of that order prices in the subscriber growth that has already happened; nobody paying $135 a share was surprised by the customer numbers. What is not yet in the price is revenue per user — and revenue per user is the only variable a satellite operator can move without launching hardware, hiring staff or waiting on a regulator. Adding subscribers costs money. Extracting more from the ones you have does not.

The second pressure runs the other way, against Mars. Every dollar of satellite margin diverted to Starship is now a dollar visible to shareholders who did not choose that allocation. Musk's super-voting control means he can keep making it, but he must now make it while being asked, quarterly, to justify it. The predictable institutional response to that scrutiny is not to spend less on the frontier. It is to squeeze harder on the thing funding the frontier, so that the diversion never shows up as a shortfall.

Consumers sit beneath both pressures. They are the only party in the structure with neither leverage nor representation.

III. Why roaming, and why so fast

Given a mandate to raise revenue per user without adding capacity, an operator looks for a capability that is plausibly costly, cheap to withdraw, valued intensely by a minority, and unavailable anywhere else. International roaming satisfies every condition.

It is plausibly costly because roaming users consume capacity in cells provisioned and paid for by residential subscribers somewhere else — which gives the operator a respectable reason to point at. It is cheap to withdraw because withdrawal is a configuration change. It is valued intensely because the people who use it have organised their working lives around it; the whole point of the Mini is that it makes certain jobs and certain places possible. And it is unavailable elsewhere because there is, at present, no alternative.

That last condition is what turns an ordinary pricing decision into something worth writing about. Amazon's Leo constellation has around four hundred satellites in orbit against Starlink's ten thousand, missed its regulatory deadline to have 1,618 aloft by July, has announced no mobility product of any kind, and offers no consumer service in most of the world. Iridium and Inmarsat are genuinely global and legally licensed almost everywhere, but they are low-bandwidth by design — adequate for email, useless for work. There is no third option.

So the customer choosing between the premium tier and losing the capability they bought hardware for is not really choosing. They are complying. Every one of them who pays is pure margin, arriving in the first full quarter reported to public markets.

IV. The case for the defence

There are honest arguments on the other side, and an argument that ignores them deserves to lose.

Network management is real. An operator has a legitimate interest in stopping customers provisioning service in a cheap market and consuming it in an expensive one, and roaming abuse is a genuine engineering problem rather than an invented one. Some restriction was always defensible.

Regulatory compliance is real too, and cuts more sharply than most commentary allows. A private company can tolerate terminals operating where it holds no licence. A listed one must disclose that exposure as a risk factor and be seen to be closing it. The new travel registration requirement — passport, live photograph — is not a customer experience measure and was never designed as one. It is a compliance artefact, and it exists because the company acquired shareholders.

Both defences explain some restriction. Neither explains this restriction. Capacity management does not require removing international use entirely from the mid-tier plans while preserving it, unchanged in every technical respect, on the expensive one. If the concern were network load, the remedy would be a data cap. The remedy chosen was a paywall.

Then there is the sequencing. Starlink published restriction language in May 2026 and pulled it. It published a revised version in June and pulled that too. Both retreats fell within the S-1 filing and the roadshow. The harsher final version appeared five weeks after trading opened. That is not proof of causation and I will not dress it up as such. It is exactly the pattern one would expect from a decision taken before the listing and held back until the book was closed.

V. What was actually taken

The word freedom has been so thoroughly annexed by people defending corporate discretion that it feels almost embarrassing to use it on the other side of the ledger. But it is the right word for what was withdrawn.

The Mini owner had bought a practical freedom: the ability to be somewhere without permission and still work. To take a contract in a place with no infrastructure. To spend a month somewhere a mast does not reach. That freedom was paid for — in hardware, on the strength of representations the seller made and the buyer relied on, and in every monthly subscription since.

This is how shareholder maximisation actually operates, and why it is so difficult to legislate against. It does not usually make products worse in ways anyone can see. It works by locating every place where a customer has been given more than the strict minimum, and reclaiming it. Optionality is expensive to supply and invisible on a balance sheet — right up until the moment you withdraw it and sell it back, at which point it appears as revenue growth.

A private firm can afford slack in the system: capabilities rarely used, allowances loosely policed, generosities that build the loyalty on which the business was founded. A listed firm cannot, because that slack is now legible as foregone revenue, and somebody will ask about it on the call.

VI. Growth and its counterfeit

SpaceX reached a $1.77 trillion valuation by doing things no publicly listed company would have been permitted to attempt. Landing orbital boosters was, for years, a capital sink with no demonstrated return and a chorus of incumbents explaining why it could not work. A ten-thousand-satellite constellation selling broadband in places with no broadband was a decade of expenditure against an unsized market. No investor committee approves either. No board signs off a market nobody has counted. Both happened because a private company with a controlling founder could ignore the people who would have stopped it.

The listing monetises that history. The first substantive act of the newly public entity was to reach into the customer base that had funded it and take something back.

The numbers will not show this, which is why the argument is easy to dismiss. Subscribers will keep climbing, revenue will keep climbing, the chart will look magnificent. But there are two ways to produce that line. One comes from doing something nobody has done before. The other comes from finding another eighty pounds a month in the pocket of somebody with nowhere else to go. They are indistinguishable on a graph for years at a time, and only one of them builds anything.

VII. The asymmetry

None of it would be possible without a contractual device so routine that nobody reads it: the unilateral variation clause, by which a customer agrees in advance that the counterparty may later change what was agreed.

British consumers are not defenceless. Section 50 of the Consumer Rights Act 2015 makes information a trader gives about a service — advertising included — a binding term where the consumer relied on it in deciding to contract. Paragraph 11 of Schedule 2 identifies as potentially unfair any term permitting unilateral alteration without a valid reason specified in the contract. There is a route: complain, wait six weeks, escalate to CISAS, then the county court.

But look at what that route demands. The individual must locate the archived webpage, quantify the loss, draft the complaint and wait months. The company need do nothing. It will settle some fraction of cases for some fraction of their value, and the arithmetic will still favour the change by an enormous margin. Consumer law in this country compensates the persistent. It does not deter the profitable, and it was never designed to.

Use it anyway. But recognise it for what it is: a mechanism chosen in place of prohibition, and one that functions as a rounding error on the balance sheet of anyone large enough to be worth suing.

VIII. The bill, itemised

South Africa has no Starlink licence and no near-term route to one, because SpaceX declines to meet the local ownership requirement by any means the regulator will accept. Farms, lodges and field camps across the country had nonetheless been getting online through terminals registered elsewhere — irregularly, but functionally. From 17 August, the passport-and-photograph regime is systematically identifying and disconnecting them.

A private company could tolerate an irregularity that connected people who had no other option. A listed company must close it, because it is now a disclosable risk. The consequence, in that specific case, is that places which had internet last month do not have internet now, and the reason is a line in a prospectus.

That is the price of shareholder primacy, denominated in the only unit that counts: somebody, somewhere, who could do something in July and cannot do it in August.

The customers built the rockets. The shareholders arrived afterwards and sent them the invoice