MOXIE

Start with what actually happened inside the box, because the process is simpler than its name. Martian air is about 95 per cent carbon dioxide, and a carbon dioxide molecule is one carbon atom holding two oxygen atoms. MOXIE pulled that air in through a baffle and a HEPA filter to keep the dust out, squeezed it with a scroll compressor which in the nominal case took the ambient 5 Torr up to about 567 Torr, roughly three quarters of Earth sea-level pressure, and heated it to about 800 degrees Celsius. At that temperature it passed the gas over a nickel cathode, which strips an oxygen atom off each CO2 and leaves carbon monoxide behind. The freed oxygen, now an ion carrying a negative charge, is pulled through a ceramic wafer of scandia-stabilised zirconia, a material that at 800 degrees will conduct oxygen ions and essentially nothing else. On the far side the ions give up their charge and pair off into ordinary O2. That ceramic sieve is the whole trick: it is what makes the product pure without any separation step. MOXIE had ten such cells, 22.7 square centimetres of working area apiece, 227 square centimetres in total for the entire experiment. Everything else in the 17-kilogram box, the heaters, the insulation, the 3D-printed heat exchangers, the sensors, the compressor, existed to hold those ten wafers at the right temperature and feed them at the right pressure.

Now the honest arithmetic. Sixteen runs spread over the better part of two and a half years produced 122 grams. JPL's own comparison is that this is about what a small dog breathes in ten hours; running the same conversion NASA used for the first run, where about 5 grams was called ten minutes of breathing for one astronaut, the entire career output would keep one person alive for roughly four hours, which is also what Hecht's separate figure of a tonne a year for four astronauts works out to. And nobody breathed it. MOXIE had no tank, no liquefier and no plumbing to anywhere; each batch was measured for volume and purity by onboard sensors and then vented back into the Martian atmosphere, along with the carbon monoxide. Perseverance itself has no use for oxygen: it is nuclear-powered and it has no crew. So MOXIE was a passenger that consumed rover resources rather than providing any, roughly 1,000 watt-hours per run, a full sol's payload energy allowance, of which only 30 to 60 watts was the chemistry and about 130 watts was the compressor, with at least two hours of preheating swallowing about half the total. Judged as a supply of oxygen it is a rounding error. Judged as an experiment it beat its own requirement twice over: the specification asked for 6 grams an hour and it reached 12.

Propellant is the reason anyone cared. Air comes a distant second. A rocket has to carry its own oxidiser, and for a methane engine that oxidiser outweighs the fuel by roughly three and a half to one, so oxygen is roughly 78 per cent of the propellant mass of a Mars Ascent Vehicle. The published figures for how much that is depend entirely on crew size and on which study you read: Design Reference Architecture 5.0 gives about 31 tonnes of oxygen for a six-person ascent vehicle massing around 50 tonnes; Polsgrove and colleagues give 22.6 tonnes for a crew of four and 33.2 tonnes for a crew of six; JPL's press release quotes about 25 tonnes for four astronauts; Rapp rounds to about 30. None of these is the number, and anyone quoting a single one without saying whose mission it is has flattened something. Against that, the same crew breathing for a year on the surface uses about one tonne between them. The propellant demand is twenty to thirty times the life-support demand, and it all has to arrive somehow. Getting mass to the Martian surface costs somewhere between 8 and 13 tonnes in low Earth orbit per tonne landed, depending again on whose figures you take, which turns 40 tonnes of ascent propellant into somewhere between about 320 and about 520 tonnes that must be launched from Earth for every single mission; the MOXIE team's own paper, using the higher multiplier, puts it at roughly 500 tonnes. Making the oxidiser on site deletes about three quarters of that.

So the sixteen runs were not about volume, they were about coverage and confidence. MOXIE ran at night and in the afternoon, at the annual maximum atmospheric density and at the annual minimum, at published run pressures from 631 to 768 pascals, and the performance differences were no larger than the changing air density predicted. It ran diagnostic runs that had nothing to do with making oxygen: one to pin down the resistance of the internal wiring, another to map how oxygen purity responds to the pressure difference between the two sides of the stack, which turned out to be the whole story of purity. Keep the oxygen side at higher pressure than the gas side and the carbon dioxide cannot leak across, and the product comes out at essentially 100 per cent. It also had to thread a needle every time it ran: too little voltage and no oxygen comes out, too much and solid carbon deposits on the cathode, blocking it and possibly cracking it. At MOXIE's reference operating point the safe band between those two limits is about 170 millivolts per cell, and the uncertainty in the wiring resistance alone ate 15 of them. The team ran conservatively on the assumption that the coolest cell in the middle of the stack, up to 10 degrees colder than its neighbours, was the one at risk. Cell resistance crept upward with every heating cycle, but slowly enough to project more than 60 cycles of useful life.

MOXIE was not the first attempt. The idea is 48 years old: Ash, Dowler and Varsi argued in 1978 that making ascent propellant on Mars beat carrying it, and coined the term ISPP for it. The first flight hardware was not MOXIE either. The Mars In-situ Propellant Production Precursor, MIP, was built for the Mars Surveyor 2001 lander with a one-kilogram zirconia cell designed to make 0.04 grams of oxygen an hour, using the same underlying technology. Its qualification unit was tested into August 2000 and its flight unit completed, by which point the lander had been cancelled because Mars Polar Lander had crashed, and MIP went into a box. Between 1978 and 2014 Mars ISRU work was, in Rapp's description, sporadic, intermittent and always at low technology readiness. What changed was an announcement of opportunity in late 2013 for a roughly 30 million dollar demonstration that would ride piggyback on a rover already going to Mars, avoiding the cost of a dedicated mission entirely. MOXIE cost perhaps 55 million dollars in the end, on Rapp's own uncertain estimate, of which about 6 million went into the electrolysis stack itself and the rest into the engineering needed to survive launch, cruise, landing and a rover's power budget.

What MOXIE did not settle is as important as what it did. It never ran continuously: a full-scale plant would operate for around 10,000 hours without stopping, and MOXIE had passed 1,000 minutes of production by its thirteenth run and went through more than twenty thermal cycles in all, so the degradation data it produced measures cycling and says nothing about endurance. It never liquefied, stored or transferred a single gram, and liquefaction and storage are where a real propellant plant becomes hard. Its compressor was not designed for long life and nobody knows what the full-scale replacement should be; scaling this one linearly would demand 42 kilowatts. The eleven stacks built for the programme varied widely from one another, which is a manufacturing question nobody has answered for the 60-cell and 65-cell stacks a full-scale plant would use. And there is a live argument about what should happen next. Rapp, a MOXIE co-investigator for nine years, argues that atmospheric processing is the only Mars ISRU that is ready, that the first crewed landings will be equatorial where near-surface water is very unlikely, and that NASA's pivot toward lunar ISRU and toward Martian water mining is putting the cart before the horse; he notes pointedly that a 2023 NASA ISRU overview did not mention MOXIE at all. The counter-position, which is NASA's programme of record, is that the Moon comes first and that lunar experience feeds forward. Meanwhile SpaceX's Mars architecture assumes methane synthesised from atmospheric CO2 and mined water ice, which needs exactly the water Rapp thinks an equatorial first landing site will not have. Both Starship and the broader case for sending people to Mars turn on that question. As of August 2026 we have found no announced flight successor to MOXIE, and no other machine anywhere that has made oxygen out of another world's own material.