The second thing to consider is the science itself, which can be problematic when your entire personal brand and adoration cult is based on the premise that you are the guy who turns science fiction into science fact.

Let’s take the two goals in turn. One million people on Mars by some undefined point in the future, and a planetary-scale orbital data centre powered by enough fissile material to make a Bond villain weep with envy.

The numbers don’t work. Not in the “this is going to be quite tricky” sense. In the “you are being sold a magic trick and the trick is that the numbers don’t work” sense.

A million people on Mars

Start with the headline number everyone quotes: it costs about $3,000 to put a kilogram into Low Earth Orbit on a Falcon 9. Down from $54,000 in the Space Shuttle era. That’s a genuine ~95% reduction and SpaceX deserve real credit for it.

Here’s the trick. An orbiting kilogram carries about 9 kWh of energy. At what you pay for electricity, that’s roughly a quid.

So Falcon 9, the greatest leap forward in launch economics in human history, is currently operating at about three thousand times the physics floor. Musk’s stated target for Starship is $10/kg, which is ten times the floor. Pick which of those two numbers you think qualifies as science fiction.

But fine. Let’s pretend $10/kg is real. Let’s pretend Starship works, reuses ten times, and hits its design targets. The Mars projections still require a thing called orbital propellant refuelling, in which a Starship launches to LEO, then waits while eight more Starships fly up one after another to top up its tanks with cryogenic methane and oxygen, and then it sets off for Mars fully fuelled. [INSERT JOKE: this has been demonstrated in space exactly zero times. The architecture for which Elon’s compensation is contingent rests on a propellant transfer technology that currently exists at the same technological readiness level as the [lightsaber/perpetual motion machine/Cybertruck wiper blade].]

Why do you need it? Because the rocket equation is a bastard.

To get from the surface of Earth to the surface of Mars you need to change velocity by about 13.9 km/s. Plug that into Tsiolkovsky with the best chemical rocket fuel currently in production and you find that 2.4% of your liftoff mass arrives at Mars. The other 97.6% is fuel, fuel tanks, fuel tank insulation, fuel pumps, and the slowly dawning realisation that the rocket equation does not care about your TED talk.

The cadence problem nobody mentions

Even if every single one of those assumptions held — the price, the reusability, the refuelling, the rocket equation handwaving — you’d still need to physically conduct the launches. A million people, at roughly a hundred passengers per Starship, means ten thousand crewed flights. Each one needs eight tanker flights to top up in LEO. Plus the infrastructure — of which more in a moment — needs another twenty-odd thousand cargo flights.

Total programme: about 250,000 rocket launches. Done in twenty years. And because Earth and Mars only line up for an efficient transfer every 26 months, almost all of those launches have to be crammed into nine three-month windows.

That works out at around 35 launches a day. Every day. For two decades.

The current global launch rate, across every space-faring nation and company on Earth combined, is roughly one launch a day. So we are talking about increasing global launch cadence by a factor of thirty-five and holding it there for twenty years.

[INSERT JOKE: At which point Boeing will presumably have got round to finishing Starliner.]

Falcon 9 currently runs at about 99.5% reliability, the best ever achieved. Nobody seriously believes that number holds when you’re launching 35 vehicles a day, when your quality control engineers have a median tenure of eighteen months, and when your boosters are on their fortieth flight. The honest expected reliability at that cadence is more like 97–98%, which is roughly where Soyuz sits.

At 97% reliability, across ten thousand crewed flights of a hundred passengers each, you lose about thirty thousand people in launch failures alone. Not in transit. Not on Mars. Just in the bit where the rocket leaves the ground.

For comparison: Challenger killed seven people and grounded the American space programme for thirty-two months. Columbia killed seven and grounded it for twenty-nine more. The notion that the American public — or any public — will tolerate the steady-state loss of a Boeing 737’s worth of would-be Martians every six weeks is [INSERT JOKE about the Overton window or perhaps Elon’s fondness for ketamine].

The colonists are a rounding error

Here’s the bit that genuinely made me laugh when I worked it out. A human being weighs about 60 kg. A million of them, therefore, weigh sixty thousand tonnes. That is the entire human payload of the colony.

The infrastructure to keep them alive — the nuclear reactors, the hydroponic farms, the life support, the heat radiators, the manufacturing seed equipment, the pharmaceuticals, the semiconductors — weighs around 2.75 million tonnes. Roughly fifty times the mass of the people themselves.

Why so much? Mars gets about 43% of Earth’s sunlight, and you can’t rely on it because the planet has dust storms that black out the sky for months at a time. So you can’t grow food in greenhouses; you have to grow it indoors under artificial light. Artificial light at the intensity plants need to actually grow burns about 15 kilowatts per person, continuously. Add heating (Mars’s average surface temperature is -63°C), pressurisation, life support, and the various industrial processes needed to produce fuel and water from the regolith, and you’re at roughly 30 kilowatts per person of continuous power demand.

For a million people, that’s 30 gigawatts. Roughly 1.5 times the entire UK electrical grid, on the surface of a planet with no infrastructure.

You can’t run it on solar (dust storms, see above). So it’s nuclear. About 250 small modular reactors. Total mass: 400,000 tonnes. That’s 400 kg of reactor per colonist — about seven times their body weight in nuclear plant, each.

Oh, and because Mars’s atmosphere is 0.6% the density of Earth’s, you can’t dump waste heat into the air. You need radiator panels. To reject 60 GW of thermal waste at Martian sky temperatures requires about 310 square kilometres of radiator surface. Roughly the size of Malta.

[INSERT JOKE: I’d like to see that on the SEC filing. Item 9.01, Exhibits: One Malta-sized radiator, F.O.B. Boca Chica.]

The bit where money stops mattering

The cost of all this, by the way, is somewhere around twenty-five trillion dollars. Roughly a quarter of one year of global GDP, or 1.1% of global GDP every year for twenty years. Expensive, but, weirdly, not impossibly expensive. The Apollo programme cost roughly 4% of US federal spending at its peak. Twenty-five trillion over twenty years is achievable in the narrow accounting sense.

Which brings us to the actual problem. The one that’s never in the investor pitch.

We do not know whether humans can live on Mars. Not in the “oh it’ll be hard” sense. In the “the relevant experiments have not been performed and there is no funded programme to perform them” sense.

The total cumulative time any human has spent in any gravity field other than Earth’s — not counting brief Apollo lunar surface sorties — is approximately zero. Across the entire history of spaceflight, every astronaut who has ever flown has been either in 1g (on the ground) or 0g (in orbit). Mars is 0.38g. We have literally no data on what happens to a human body in 0.38g for months, let alone years, let alone a lifetime.

The assumption baked into every Mars colony projection is that 0.38g will be kind of like 1g, only a bit lighter. Bones won’t waste away the way they do in microgravity. Hearts won’t atrophy. Eyeballs won’t deform. There is no evidence for this. It’s a hope.

No mammal has ever successfully reproduced in space. Mouse embryos flown to the ISS developed abnormally. Sperm motility is impaired in microgravity. Foetal bone development, neonatal cardiovascular adaptation, mammalian gestation in 0.38g — entirely uncharacterised. If it turns out that human reproduction doesn’t work on Mars, then your “permanent colony of a million people” is in fact a temporary colony with a half-life equal to the oldest colonist. The whole thing is a retirement village in a hostile environment.

The radiation dose on the Martian surface is about 230 millisieverts a year, unshielded. NASA’s career limit for astronauts is 600–1000 mSv across their entire working life. A five-year stay on Mars puts a colonist over that limit. A fifty-year residency puts them at roughly 11.5 sieverts. We have no human cohort data above about 1 Sv except atomic bomb survivors and Chernobyl liquidators, neither of which is remotely analogous to chronic, steady-dose, multi-decade exposure.

The single cheapest experiment that would actually answer the central question — can humans survive long-term in 0.38g? — would be a rotating module on the ISS producing artificial gravity by centrifugal force. It has been proposed, in various forms, for more than thirty years. It has never been built. It is not funded. It is not on anyone’s roadmap.

The most charitable reading of this is that nobody wants to know the answer.

The trick

Here is what I think is actually going on. The engineering questions — cost, cadence, reactors, hydroponics — are hard in the sense that they yield to money and time. Throw enough capital at them, iterate enough times, and they get solved. Twenty years, fifty years, some-day-but-eventually.

The biological questions — can humans live in 0.38g, can mammals reproduce off Earth, what does multi-decade radiation exposure do, can a closed-loop biosphere work at scale — are hard in a completely different sense. They are gated by experiments that take human lifetimes to run, that you cannot parallelise, and that you cannot buy your way past. Some of them may turn out to have the answer no.

The cost-per-kilogram framing is being used as a magic trick. It’s the metric that gets cited because it’s the one that can be computed. The metrics that actually determine whether any of this is possible — mostly biological, mostly experiment-limited — get rounded to “hard” because they aren’t computable. The numbers in the SEC filing are correct given their assumptions and irrelevant to the actual question.

It’s not that Mars is impossible. It might be possible. It is that the people pitching it are pitching the part of the problem they know how to do arithmetic on, and quietly hoping nobody notices that the binding constraints are in a category they cannot do arithmetic on at all.

Which, frankly, is on-brand. [INSERT KILLER LINE that ties it back to the Cybertruck/Autopilot/Thai cave throughline — something like: this is the same man who launched a car at Mars because he couldn’t think of anything else to put on top of a rocket.]