The Physics of the Premium

Buried in the twelve-tranche SpaceX founder compensation package are two milestones that no finance committee can value on a spreadsheet alone. The first conditions a tranche on the company delivering on the order of 100 terawatts of compute, explicitly including capacity off Earth. The second pairs a $7.5 trillion market capitalisation with a permanent, self-sustaining settlement of at least one million people on Mars. These are not financial targets. They are physical ones. Pricing them honestly means asking what they would take in radiator area, regolith depth, launch cadence, and capital — and identifying the wall each one hits first.

The principle the model is built on

The simulator does not return a flat possible or impossible. The honest answer to "can this be done?" is conditional, and pretending otherwise turns analysis into theatre. Every output names the binding constraint — the wall the design hits first — and labels what is demonstrated technology versus what is speculative. The user supplies the assumptions; the physics supplies the consequences. That separation is what makes the tool defensible to an institutional reader and useful to a household one.

Module A: the cooling wall

In vacuum, heat does not leave by air or water. It leaves only by thermal radiation, governed by the Stefan–Boltzmann law: Q = ε σ A (T⁴ − T_env⁴). Because essentially all compute power becomes waste heat, the radiator area required scales linearly with the compute budget. A reasonable engineering anchor is roughly 1,200 square metres of radiator per megawatt of waste heat at electronics temperatures near 20 °C — about four tennis courts per megawatt.

At Starcloud-class single-digit megawatts, this is large but not absurd. At hyperscale gigawatts, it becomes the first wall to break. At the 100-terawatt milestone scale — roughly thirty times the entire average United States electrical consumption — the radiator surface required is measured in millions of square metres per major facility and the launch mass is measured in megatonnes. The simulator does not editorialise this. It computes the area, mass, and cost, and flags the result as outside the demonstrated-technology envelope so the reader can see where aspiration has left engineering.

The honest output for the 100-terawatt milestone is not "impossible." It is "here is the radiator area, here is the launch mass, here is the cost per delivered kilowatt-hour — now decide what probability to assign."

The rad-hardening tax nobody on the call mentions

Even where the thermal physics is permitted to close, the compute itself pays a tax in orbit. Total ionising dose and single-event upsets corrupt and degrade modern processors. Radiation-hardened parts are an order of magnitude or more slower and less efficient than the GPUs that fill ground-side hyperscale facilities — peer-reviewed work puts the derate at roughly ten to fifty times. The implication is that one nameplate orbital megawatt of compute delivers a fraction of the useful work that a terrestrial megawatt delivers. The simulator surfaces this explicitly: as the rad-hardening derate grows, effective exaflops fall, and cost per delivered kilowatt-hour rises to a level only the most latency-tolerant workloads — training, batch, rendering — can absorb. Interactive workloads exclude themselves.

The reference number

Engineering analyses circulated in early 2026 put a single 1-gigawatt orbital data centre at roughly $42.4 billion, against roughly $14 billion for a 1-gigawatt facility on the ground. That three-times multiplier is the floor, not the ceiling — it assumes a Starship cadence and a launch cost near $500 per kilogram that does not yet exist. The simulator's cost-per-kilowatt-hour output is the right number to argue about. It reflects the binding economic constraint that decides whether an orbital data centre is a real business or a marketing milestone.

Module B: Mars receives sunlight; people do not

The popular image of a Mars colony is a glass dome under a pink sky. The physics points elsewhere. Mars does receive sunlight — about 43% of Earth's surface intensity, on a day only slightly longer than ours. The planet is not dark. The problem is not the Sun. The problem is what protecting a population from 240 to 300 millisieverts per year of background dose does to its access to the Sun.

A surface dome admits light but provides almost no shielding. A subsurface habitat under one to several metres of regolith provides shielding but admits almost no light. Engineering can soften the trade: overhangs and a shielded roof near a window can block roughly 70% of incident radiation, bringing that zone toward 20 to 25 millisieverts per year; light tubes and periscopic mirrors can pipe daylight inward. But these are rationed exceptions, not a sunlit life. The simulator models the dose-versus-light trade as a spectrum across surface-dome, hybrid, and subsurface habitats so the user sees the swap directly. The honest finding the design is built to surface is that, with current and near-term technology, a million-person Mars settlement is overwhelmingly an indoor, radiation-shielded, artificially-lit existence with natural-light access deliberately rationed.

The arithmetic of a million

Reaching one million people is a logistics problem before it is anything else. The Mars launch window opens roughly every twenty-six months. Settlers per window equals vehicles per window times seats per vehicle; total launches and years to one million follow directly. Musk's published target of roughly $200,000 per seat, against an industry-traditional figure closer to $10 billion, drives the transport cost. Habitat capital — life support, ISRU plant, structural mass under regolith — drives the rest. Independent academic work points in different directions: a Purdue Project Destiny student study modelled one million people over a century at the $200,000 ticket; an IEEE Spectrum analysis put a first batch of fifteen Falcon Heavy launches near $4.5 billion. The simulator lets the user pick a position on that spectrum and shows the years-to-one-million, the total launches, and the capital required as a range.

Self-sustaining means closing the loop

The milestone is not "a million people on Mars." The milestone is "self-sustaining." That word does the heaviest lifting in the package. Self-sustaining requires in-situ resource utilisation to close the loop on water, oxygen, propellant, food, and structural materials — to a percentage that today exists only as research-grade capability, not industrial. The simulator surfaces an ISRU closure percentage and reports the gap. At 60% closure, 40% of consumption is still resupplied from Earth, and "self-sustaining" is a marketing word, not an engineering finding.

What this means for the valuation

The valuation question is not whether either milestone is achievable. It is what probability a disciplined model should assign to each, and what discount the package as a whole should carry given the magnitude of the assumptions required. The companion piece — Pricing the Founder — treats Mars and orbital compute as real options in a probability-weighted overlay on top of a Starlink-anchored fundamentals stack. The physics outputs from this tool feed that probability directly. If the 100-terawatt milestone requires radiator areas measured in millions of square metres per facility and launch tonnages no Starship cadence delivers within a decade, the probability of the milestone vesting on any reasonable horizon collapses, and so does its option value.

That does not make the package worthless. It makes the narrative premium quantifiable. An allocator who can write down, in basis points, how much of the $1.77 trillion ask is paying for milestones whose physics requires step-change engineering is an allocator who can defend the position in front of a risk committee. An allocator who cannot is paying a premium they have not named.

What the editorial does and does not do

This tool is not a prediction. It does not assert that either milestone is achievable or unachievable. It returns, for each set of user assumptions, what the design would require and which constraint binds first. The defaults are sourced — SatNews on radiator physics, peer-reviewed work on rad-hardening derate, Marspedia and Curiosity-RAD on Mars dose, Project Destiny and IEEE Spectrum on Mars logistics, Fortune and Bloomberg on the compensation milestone language itself. Where popular coverage and the technical literature disagree, the technical literature wins. Where the assumptions exceed demonstrated technology, the output is flagged as such on the same line as the number.

The product principle is non-negotiable. The tool refuses binary verdicts because the honest answer is conditional. That protects the brand's authority and keeps the experience educational rather than polemical. It also gives the institutional reader something they can cite, and the household reader something they can use to make a smaller decision than the headline would otherwise invite.