Mars Power

The gate

The Dust Storm Is the Whole Argument for Fission

Not the night, not the distance from the Sun, not winter. The storm.

Mars receives about 590 W/m² at the top of the atmosphere, roughly 43% of Earth’s. That is a manageable penalty. Seasonal variation adds another: eccentricity drives insolation from about 493 W/m² at northern summer solstice to 717 at southern, and latitude adds a further winter cut. None of that rules out solar.

The planet-encircling dust storm does. Roughly once every three Martian years, dust wraps the globe and surface insolation falls to about 59 W/m², which is dimmer than Jupiter’s orbit at 50 and darker than Ceres at 177. The 2018 event killed Opportunity. A 2022 regional storm ended InSight’s mission, and China’s Zhurong entered hibernation the same year and never woke up, with panel dust the leading explanation.

Sizing a solar-primary site against that floor requires roughly 8,500 m² of array to hold a 150 kWe bus — deployed, kept clean, and on a surface actively coating it at about 0.2% power loss per sol unmitigated. The storm arrives on its own schedule with no useful warning, it lasts weeks, and it is planet-wide, so no second site anywhere on Mars escapes it. Storm resilience is bought in fission and in stored energy, and it cannot be bought in dispersal.

This is the sharpest contrast with the Moon in the whole series. At the lunar south pole, an illuminated rim point sees sunlight more than 90% of the year and solar with a modest buffer covers the base case, so fission is not a capability gate there. On Mars it is.

What exists

Surface Fission Hardware: What Is Funded, and What Is Not

Reactors in the tens of kilowatts are real programs. Megawatt-class surface power exists nowhere.

The last two rows carry the architectural weight. Radioisotope units are probe hardware, not base hardware: at roughly 110 We each and a plutonium-238 supply measured in low single-digit kilograms a year, they will never power a settlement. And the absence of any megawatt-class surface program is precisely why the architecture choice matters — direct-to-surface propellant production needs 438–912 kW, which is that missing program, while the orbit-staged ferry needs 68–142 kW, which is the hardware already under contract.

Underneath all of it sits a supply constraint the space program does not control: these reactors run on high-assay low-enriched uranium, and HALEU enrichment capacity is a terrestrial industrial-policy question. It gates the schedule regardless of how the flight hardware progresses.

Split the reactor

Several Small Units Beat One Large One, and the Reason Is Landing Attrition

Same expected power, very different distribution — and a monolith cannot be split after it is designed.

At the working assumption of 50% landing survival for uncrewed cargo, a single 100 kW reactor gives a coin-flip between full power and none. Dividing the same capacity across units transforms the odds.

Two units is the minimum and four is better if the capacity divides cleanly; three is the awkward case, where the parity of the half-power threshold works against you. The practical conclusion is that the requirement is not a bigger reactor but several of the one that already exists — the 40 kWe fission surface power class is already near the right unit size. Lockheed has argued the same position publicly on program-risk grounds, which is independent industry confirmation of a conclusion this analysis reaches from attrition alone.

The redundancy argument and the attrition argument point at the same decision, which is unusual and worth taking seriously. It also has to be made before the unit is designed, because a monolithic reactor cannot be subdivided afterward.

One correction to the naive reading. Those probabilities assume independent landings. Common-mode failure — a shared entry system, a shared descent profile, a shared software load, a dust event over the ellipse — erodes the benefit fast: at moderate correlation the four-unit advantage falls by roughly a factor of eight. Split the units across launch windows, not across landings within one window.

Reactors don’t throttle

Buy Energy With Reactors, Buy Peak Power With Storage

Fission wants constant output. Load-following costs xenon transients, drum actuation, thermal cycling and wasted reactor-hours.

A reactor is a capital asset whose cost is incurred whether or not the power is used, and whose operating life is shortened by cycling. N coupled units deliver N times rated output continuously — not a larger peak. Every kilowatt of peak capability beyond the average load should therefore come from a buffer, not from installed nuclear capacity.

Where the crossover falls depends on the buffer’s specific energy against the reactor’s specific mass, at roughly 100–200 kg/kWe:

The resulting rule is short: load under a day, use a battery; one day to a week, use a regenerative fuel cell; continuous beyond a week, build another reactor. Sodium-ion rather than lithium, because lithium sits far below ore grade off Earth and the chemistry substitution is nearly free.

Peaky industrial loads turn out to be almost free. A plasma cutter drawing 100 kW for two hours is 200 kWh — 0.2 t as a regenerative fuel cell, 1.4 t as sodium-ion, against roughly 15 t if it were sized as installed reactor capacity. That is a 10 to 75-fold mass advantage, and it means the arc melter, the plasma torch, the sintering head and the drill are not reactor-sizing loads. Only the continuous ones are.

Dispatch

The Propellant Plant Is the Site’s Load-Follower

Large, continuous, genuinely interruptible, and its product is storable. Nothing else on the site has all four properties.

Grid precedence on a Martian site falls out of the physics rather than from policy. Reactors run flat out. Life support and habitat take what they need off the top. Peaky industrial loads come from the buffer. The propellant plant absorbs the remainder and is curtailed first, because electrolysis and Sabatier throttle in seconds and a tonne of methane not made today can be made next week.

That reframes how the plant should be specified. Propellant production is an energy integral over a 780-day window, not a power rating — the 68–142 kW and 438–912 kW figures are window averages, and the plant should be designed to a duty cycle rather than to continuous operation. With several trains the curtailment is granular: shed a train, no fast control loop required.

One asset is already paid for and rarely counted. A tank farm holding hundreds of tonnes of methalox is a 206 MWh chemical energy store — far larger than any battery bank the site will ever build, and recoverable through a fuel cell or a generator set if the reactors ever go down together. The propellant inventory is the site’s deep reserve as well as its ticket home.

Solar

Solar Is a Co-Primary, Not a Survival Layer

Ruling out solar-primary is not the same as ruling out solar.

Everything above says solar cannot carry the storm case. It says nothing against solar carrying a large share of the nominal case, and the arithmetic favors doing exactly that: photovoltaic arrays are cheap per kilowatt, arrive divisible, tolerate partial losses on landing, and can be expanded incrementally by machines already on site. Under the orbit-staged architecture, where the whole site draws a couple of hundred kilowatts rather than a megawatt, solar plus storage can be co-primary with fission rather than a backup to it.

Two operational facts shape the installation. Dust deposition costs roughly 0.2% of output per sol if nothing intervenes — the figure that ended Opportunity, Zhurong and effectively InSight — and dust devils clean panels stochastically, which the Mars Exploration Rovers survived on for years but which no program should plan around. Active cleaning, electrodynamic dust removal, or simply oversizing the array and accepting the loss are the three honest options, and nobody has demonstrated the first two at scale.

The other fact is geometric: at these distances and latitudes the Sun is low for much of the year, so tracking and vertical mounting recover more than they cost. None of it changes the storm conclusion. It changes how much reactor the site has to land in the first window.

Distribution

Moving Power Across a Site That Sits in the Paschen Minimum

High-voltage spans are fine. Terminations are the problem, and that inverts the usual advice.

Conductor mass scales as the inverse square of transmission voltage, and the difference is not marginal: carrying 100 kW over 2 km at 5% loss needs 212 mm² of aluminum, about 2.3 t, at 1 kV — or 24 mm², about 255 kg, at 3 kV. On a site where every kilogram was landed at great expense, that is a real argument for going high.

Mars pushes back. At 6 mbar of carbon dioxide the Paschen minimum sits at 292 V across a 1.39 mm gap, so a few hundred volts can strike an arc across a millimeter. The early form of this finding said no high-voltage distribution on Mars; the corrected form is narrower and more useful. At 3 kV the required clearance in open Martian air is only about 7 cm, so overhead and surface-laid spans are perfectly safe — it is every connector, feedthrough, switchgear enclosure and junction box that sits at millimeter scale in dust-fouled air. The rule is no unsealed high-voltage terminations: pot them, pressurize them, or design clearances an order of magnitude past the minimum.

Aluminum rather than copper throughout, incidentally. Per unit mass at equal conductance aluminum is about 2.1 times better, and it is 9 wt% of the regolith against copper at parts per million — the substitution is free now and locally sourceable later.

The layout conclusion is to cluster rather than scatter. Two or three generation-and-load groups — an industrial and propellant node, the habitat, perhaps the ice mine — beat five separated singletons, because every additional run costs conductor mass and every additional termination is a failure point in an environment that is trying to arc across it.

Heat rejection

Every Kilowatt Generated Is a Kilowatt to Throw Away

The atmosphere is too thin to help much, so rejection is radiative and the radiator is a major structure.

A fission plant rejects several times its electrical output as heat, and at 0.020 kg/m³ the Martian atmosphere carries away only a small fraction of what air does on Earth. Rejection is therefore radiative, and radiator area at ordinary rejection temperatures runs about 2–3 m² per kilowatt of heat — a figure that barely varies across the solar system, because radiated power goes as the fourth power of temperature and the sink temperature hardly matters by comparison.

Two Mars-specific complications apply. Dust settles on radiators and degrades emissivity, so radiators need the same cleaning strategy as arrays and rarely get one in published designs. And the daily 80–100 K swing means the radiator sized for the worst case is oversized for most of the sol, which argues for variable-conductance designs or for simply accepting the mass.

One useful coupling: the same buried-regolith trick that protects cryogenic tanks does not work for radiators, because burial is insulation and a radiator needs the opposite. Heat rejection wants surface area in the open, exactly where the dust is. Thermal design and dust mitigation are the same problem wearing different hats, and on this site they always will be.

Scaling

Kilowatts Per Resident, and Where the Megawatt Threshold Falls

Reactor class is a function of population, and the crossing arrives about four times earlier than the propellant argument alone suggests.

Counting propellant alone, a megawatt supports 200–450 residents. Counting everything, 1 MW supports 40–103 residents, 10 MW supports 400–1,000, and 100 MW supports 4,000–10,000. The two figures are not in conflict; they are calculations of different scope, and the all-in number is the one a settlement has to plan against. A 500-person settlement is a 10 MW program, not a 1 MW one — a different reactor class entirely from the 40 kWe units now in development.

Food lighting dominating the stack is the finding underneath that. It comes straight from photosynthetic efficiency: 2,500 kcal a day is 10.5 MJ, and at 1–3% light-to-edible conversion with 50–70% LED wall-plug efficiency, feeding one person takes 4–12 kW continuous. Sunlight does not rescue it, because installed capacity is sized by the storm. The crewed settlement is a farm with a reactor, in the same way the uncrewed outpost is a power plant with an excavator.

Record

Provenance and Data Sources