The Lunar Mass Driver

How it works

Electromagnetic Launch: Coilgun Physics on a Kilometers-Long Track

A maglev sled, a sequence of superconducting coils, and no atmosphere to stop it.

A mass driver is a linear electromagnetic accelerator built on a surface. A payload sits on a magnetically levitated sled inside a sequence of superconducting coils. The coils fire in rapid sequence, accelerating the sled down the track. When the sled reaches the target velocity — 2.38 km/s for lunar escape, roughly 8,570 km/h — the payload separates and continues into space on its own momentum. The sled decelerates and returns to the start for the next launch. No propellant is consumed. The energy comes from the electrical grid. The track is reusable. The sled is reusable. The only consumable is the electricity.

The physics are the same as a coilgun or a naval electromagnetic aircraft launch system (EMALS, deployed on the USS Gerald R. Ford carrier class). The U.S. Navy’s EMALS accelerates aircraft of up to 45,000 kg to roughly 130 knots over a 91-meter stroke. The Navy’s railgun program reached muzzle velocities around 2.5 km/s with small projectiles in ground testing before it was shelved in 2021 — above lunar escape, on hardware that already existed. The unsolved part is not the speed. It is the track length, the pulse-power system, the construction on the lunar surface, and what catches the payload at the far end.

Why not Earth

The Atmosphere Is What Rules Out a Terrestrial Mass Driver, Not the Track Length

Earth escape at 1,000 g needs 6.4 kilometers of track. That part is buildable. The other part is not.

It is often said that an Earth mass driver would need an impossibly long track. Run the arithmetic and that turns out to be the weakest of the objections. Earth escape velocity is 11.2 km/s, and at 1,000 g the track works out to 6.4 kilometers — longer than the lunar case but not absurd, and shorter than several particle accelerators already in the ground.

What actually kills it is the air. A payload leaving a ground-level track at 11.2 km/s is travelling at roughly Mach 33 through the densest part of the atmosphere. The stagnation heating and the drag losses are both catastrophic, and no thermal protection survives the ascent at that speed and that density. Building on a high mountain reduces the problem without solving it, and the payload still has to be aimed through the whole remaining column.

The Moon removes the objection entirely. Gravity is 0.165 g, so escape velocity is only 2.38 km/s — a track of 2.9 km at 100 g. The atmosphere is effectively zero at 10⁻⁹ bar, so there is no drag, no heating and no shock wave, and the payload exits the track directly into space with no transition to manage. Electromagnetic surface launch is not merely possible on the Moon; it is architecturally simpler than chemical launch. The same is true, and more so, at Ceres, where escape velocity is 516 m/s and the launch energy is a twenty-first of the Moon’s.

One free lunch the Moon does not get: rotation. A body’s own spin contributes to launch velocity, and the Moon, turning once in 27.3 days, offers 4.6 m/s at the equator — two parts in a thousand, and worth nothing. Fast-rotating small bodies collect a third of their launch energy this way. The Moon pays for every meter per second. Observed

O’Neill to SpaceX

From the 1977 Ames Summer Study to the 2026 SpaceX Roadmap

Gerard O’Neill proposed the lunar mass driver in the 1970s. In February 2026, SpaceX put it on its corporate roadmap.

Princeton physicist Gerard K. O’Neill proposed the lunar mass driver in the mid-1970s as the supply mechanism for orbital space habitats. His team at the 1977 NASA Ames Summer Study designed a launcher operating at about 1,000 g, firing payloads of roughly 10 kg at a cadence of several per second to supply raw material to construction sites at the L5 libration point. The design was technically rigorous — optimized coil parameters, energy storage and system mass are documented in the NASA Space Resources proceedings — but it had no customer, no launch vehicle capable of delivering the hardware, and no political constituency. It remained a study for five decades.

Things changed in February 2026. In a corporate update following SpaceX’s merger with xAI, Elon Musk stated that SpaceX would prioritize a “self-growing” lunar city over immediate Mars colonization, and that an electromagnetic mass driver on the Moon would be the export mechanism for AI-satellite manufacturing. In March, Musk posted publicly: “Mass drivers on the Moon!” and described solar-powered AI satellites manufactured on the lunar surface and launched into a distributed orbital compute network. Building

SpaceX has not released engineering specifications. What exists is a declared corporate priority from the company that operates the world’s highest-cadence launch vehicle and is building the only vehicle capable of delivering 100+ tonnes to the lunar surface. The mass driver moved from a 1970s academic concept to a funded company’s stated roadmap item in one announcement. The physics were never the obstacle. The logistics vehicle was. Starship is that vehicle.

Track geometry and power

Track Length, Acceleration, and Power: The Engineering Trades

A shorter track means higher acceleration. Higher acceleration means cargo-only. The trades are sharp.

The fundamental relation is d = v² / 2a, where d is track length, v is launch velocity and a is acceleration. At 2.4 km/s, essentially lunar escape:

Ordinary spacecraft qualify to roughly 10–20 g quasi-static, so the 30 g row is already demanding for anything with deployable structure, large optics or populated circuit boards. A satellite launched by mass driver is one designed from the start for the track, not a conventional bus put on a sled. The baseline design uses a 3.2-km track at roughly 100 g.

Power. A 100-kg payload at 2.4 km/s carries 288 MJ of kinetic energy, which at 90% electrical-to-kinetic efficiency means 320 MJ drawn. Spread across the 2.45-second launch at 100 g, that averages 131 MW. The instantaneous figure is worse, and the reason is easy to miss: under constant acceleration the force is constant while the velocity rises linearly, so power rises linearly too. Peak power at the muzzle is twice the average — about 261 MW, reached in the last instant before release.

No plausible lunar power system delivers that continuously, and none has to. The energy per shot is modest; it simply has to arrive very fast. The launcher runs on a store that charges slowly from the grid and empties in seconds. At one launch per hour the average charging draw is 89 kW; at ten per hour, 890 kW; at O’Neill’s several-per-second cadence with 10-kg slugs, around 134 MW sustained, which is a power-plant problem rather than a storage problem and is the real reason that architecture assumed an orbital solar power station. At any realistic cadence the track is an energy-storage problem as much as a magnetics problem. Projected

The catcher problem

Aiming Is Harder Than Launching: Six Millimeters per Second at L2

The gun is proven hardware. Hitting a target 400,000 kilometers away with an unguided slug is not.

A mass driver throws payloads; something has to receive them. O’Neill’s architecture put a mass catcher at a libration point — a large structure that intercepts the stream and accumulates it. Every lunar mass-driver architecture since has needed the same component, and it is consistently the harder half of the system, for a reason that has nothing to do with energy.

An unguided slug is a ballistic object from the moment it leaves the track. Over a flight of roughly 400,000 km to a libration point — about 46 hours at 2.4 km/s — small errors at the muzzle become large errors at arrival:

One arcsecond of pointing error is 1.9 km of lateral miss at 400,000 km. The millimeter-scale track alignment quoted in most designs is not fastidiousness — it is precisely what a sub-arcsecond aim demands, and it has to hold through thermal cycling, moonquakes and the recoil of every previous shot.

There are three ways out and every architecture picks among them. Build the catcher large — O’Neill’s were kilometers across, which is its own construction program. Guide each slug with a small mid-course correction, which puts a thruster, a receiver and a power source into every payload and destroys the economics of throwing bulk regolith. Or hold the muzzle to parts per million, which no launcher of any kind has ever demonstrated. Pick two; nobody has shown all three.

The arrival velocity is the second half of the problem. A catcher on a matched trajectory still meets the stream at whatever relative velocity the transfer leaves, and absorbing that without destroying either the payload or the catcher is a structural problem at tonnage scale. The gun is old technology. The catcher has never been built at any scale, anywhere.

It is worth stating how much of this is specific to the Moon rather than to mass drivers. The lever arm is what hurts: the same one-arcsecond pointing error that misses by 1.9 km at a libration point misses by under 5 meters at a receiving station 1,000 km up. A launcher on a small body throwing to a nearby orbit faces a rounding error of this problem — see The Ceres Mass Driver. The Moon is where electromagnetic launch first becomes possible and also where its aiming problem is worst. Projected

The pulse store

Superconducting Magnetic Energy Storage: Powering the Launch Pulse

Charge slowly from the grid. Discharge in milliseconds. The coils that accelerate the payload are also the storage.

The natural energy-storage technology for a coilgun is Superconducting Magnetic Energy Storage (SMES). In O’Neill’s original design, the superconducting drive coils themselves served as the energy store — a “bucket brigade” where each coil charges magnetically, dumps its energy into accelerating the payload carrier as it passes, then transfers the remainder forward to the next coil in the sequence. The coils serve double duty: accelerator and storage. This is inherently a SMES architecture — the same principle used in terrestrial SMES for grid stabilization, but distributed along the track rather than concentrated in a single toroid.

Modern designs would add dedicated SMES banks at intervals along the track. Each bank charges slowly from the solar or fission grid between launches and discharges in milliseconds during the pulse. SMES has two advantages here: it stores energy in the magnetic field of a superconducting coil with near-zero resistive loss and essentially unlimited cycle life, and it discharges at the rate the coilgun needs. Ultracapacitors can supplement it for the fastest switching transients. Batteries and flywheels are too slow for the primary pulse but serve well as the intermediate store between the grid and the SMES charge cycle.

Superconducting coils require cryogenic temperatures — below roughly 77 K for high-temperature superconductors (HTS), below 10 K for low-temperature superconductors (LTS). The lunar surface offers a seemingly obvious solution: the permanently shadowed crater floor at 40–88 K provides free cryogenic cooling. Early analysis explored exactly this — siting the SMES in the cold trap to exploit the ambient temperature.

That approach was rejected. The same thermal budget constraint that governs ISRU extraction governs any heat-producing system placed in a cold trap. A surface at 40 K radiates about 0.15 W/m² and nothing else carries heat away, so the trap’s entire capacity to shed added heat is fractions of a watt per square meter. A SMES bank cycling at multi-megajoule scale exceeds that by orders of magnitude, and the ice inventory it would sit among took two billion years to accumulate. Using the cold trap as free cooling for industrial equipment would consume the cold environment that makes it valuable — destroying both the volatile inventory and the cryogenic computing case.

The corrected architecture places the SMES banks along the track itself, on the sunlit plain where the mass driver is sited — not in the cold trap. The coils require active cryocooling rather than passive environmental cooling, which adds mass, power draw and complexity. HTS coils operating at 40–77 K are the pragmatic choice for this application, because the refrigeration penalty scales steeply as operating temperature falls: removing a watt at 4 K costs several hundred watts of input, against a few watts at 77 K. Modern REBCO tape has closed the field gap that once favored low-temperature coils — high-temperature magnets have demonstrated fields above 20 T, so the old trade of storage density against refrigeration cost no longer bites the way the literature assumes. The remaining penalty is cost and conductor length, not stored energy per unit volume. Projected

What it launches

Export Manifest: Water, Oxygen, Metals, Regolith, and AI Satellites

The mass driver does not care what is in the container. It cares about the container’s mass, shape and acceleration tolerance.

Any payload that survives the acceleration and fits the sled can be launched. The export list, roughly ordered by industrial timeline:

Water and liquid oxygen. The highest-demand export in the near term. Polar water or regolith-derived oxygen launched to cislunar propellant depots eliminates the tanker chain from Earth for every mission beyond LEO. Water goes as frozen blocks — dense, inert, unpressurized and indifferent to 100 g.

Bulk regolith and sintered feedstock. Raw material for orbital construction. At 1,000 g, regolith slugs are the simplest payload class — no packaging, no fragile components. O’Neill’s original design was built for exactly this, and it is the one class where a large catcher and an unguided stream genuinely work.

Metals. Iron, ferrosilicon alloy, aluminum-silicon alloy — the co-products of oxygen extraction. Cast as ingots and launched to orbital manufacturing sites.

AI satellites. SpaceX’s stated use case: solar-powered compute satellites manufactured on the Moon from local materials and launched into a distributed orbital network. This is the payload class that strains the architecture hardest. It needs the low-acceleration track, and it needs guidance, because a satellite is too valuable to throw at a catcher and hope. A guided payload is a payload with a thruster and a power supply, which is most of a small spacecraft. Projected

Helium-3. High value per kilogram, low mass per launch. See the Helium-3 page for why the demand case rather than the launch case is the problem.

The mass driver does not launch anything to the lunar surface. It does not replace the transportation architecture for inbound cargo and crew. It is a one-way, surface-to-orbit export mechanism. The Moon imports by rocket. It exports by rail.

Site requirements

Where to Build It: Long, Level, Uniform Ground

A 3-km track needs 3 km of aligned, graded surface. The Shackleton rim is a ridge a few hundred meters wide.

The mass driver needs a long, level, straight surface with uniform bearing strength. As the catcher arithmetic shows, the alignment tolerance over the full length is millimeters, because that is what sub-arcsecond pointing means over three kilometers. A 3.2-km track at the south pole requires terrain that the narrow Shackleton rim does not provide.

Candidate sites from the Artemis candidate regions include Mons Mouton and Slater Plain — both broad and relatively level, at roughly 80–130 km from the pole. The habitat wants the rim. The mass driver wants the plain. They are not in the same place. That separation creates a logistics link — processed material must move from the extraction site to the launch site — and a power link, since the track draws pulse power the local grid must supply. See the Lunar Power page.

Track orientation also determines what is reachable. A fixed track launches in one direction, and the Moon’s slow rotation means the aim point sweeps through the sky over 27.3 days rather than 24 hours — launch windows to a given destination are correspondingly rare and long. Payloads aimed at different inclinations require a curved terminal section, a second track, or an on-payload thruster. And a payload launched below escape velocity follows a ballistic arc back into the surface: a mass driver cannot place anything in lunar orbit without a circularization burn carried by the payload itself. Projected

The strategic dimension

Dual Use, Golden Dome, and the Weaponization Question

A system that can fling 100 kg at 2.4 km/s in any direction is a weapon as well as a launcher.

A mass driver is inherently dual-use. The same track that launches water to a cislunar depot can launch a kinetic impactor at a target. The Outer Space Treaty (1967) prohibits military bases, installations and fortifications on celestial bodies under Article IV — but a mass driver is mixed use, and regulating dual-use hardware is difficult when the civilian application is identical to the military one except for the trajectory.

The connection to space-based missile defense has been noted explicitly. The Golden Dome concept envisions a space-based intercept layer, and a mass driver could launch kinetic interceptors from the lunar surface without consuming propellant. China treats mass drivers and related technologies as central to lunar industrialization and long-term space development. The RAND Corporation and other defense analysts have flagged the dual-use problem in published assessments. Projected

One technical point sharpens the policy one, and it is the same finding that appears at Ceres and in the redirection argument: nothing about the hardware distinguishes a delivery from an attack. Only the aim point does, and the aim point is software. An inspection regime that certifies a track as civilian certifies nothing that survives the next command load. This page does not resolve that. It notes that a system capable of launching payloads at escape velocity from a sovereign-controlled surface is a strategic asset regardless of what its operators call it, and that the Outer Space Treaty was written before any nation could build one. The strategic context is on the Lunar Contest page.

Record

Provenance and Data Sources