Lunar grid trades: why 3 kV AC beats lighter DC

What the work claims

This is a primary modeling-study presentation: every result below comes from simulation in NASA Glenn's Electrical Power System Sizing and Analysis Tool (EPS-SAT), a MATLAB-based power-system concept tool, using planetary surface cable and converter models built for the purpose.¹ Four claims carry it. First, for an Artemis-scale lunar grid connecting a 20 kW-load habitat with two 10 kW 120 Vdc sources, an ISRU plant with eight 10 kW 120 Vdc sources and a 68 kW load, and a 40 kW fission source, 3 kV AC is the mass-optimal feasible design point among voltages from 1.2 to 6 kV, DC and three-phase AC, and frequencies from 60 Hz to 1 kHz.¹ Second, DC transmission is significantly lighter than AC above 3 to 4 kV in the model, but a DC grid is practically limited to 1.2 to 1.5 kV by the lack of high-voltage rad-hard switches, which would force 4 to 5 series 300 V stages, with more stages judged prohibitive; this, not cable physics, drove the team to AC.¹ Third, fault tolerance has a steep mass price: adding a ring tie line costs about 50 percent more transmission mass than radial, and a full mesh about 100 percent more than radial and 50 percent more than ring.¹ Fourth, follow-on studies map voltage against distance and power for the grid beyond Artemis: at 100 kW on insulated 8 AWG cable, 3 kV reaches 600 m and 18 kV reaches 20 km within loss limits; at 1 MW, 9 kV reaches 1 km and 24 kV reaches 10 km; at 10 MW on bare 4/0 cable, 30 kV reaches 4 km and 240 kV reaches 400 km.¹

How it works

The physics is Ohmic loss against insulation and converter mass. For a given power, cable loss scales with the square of current, so raising voltage lowers loss and copper mass together; but higher voltage demands thicker insulation, more demanding switches, and eventually conversion hardware. EPS-SAT sweeps that trade across architectures (radial, ring, mesh), power types (DC and three-phase AC), voltages (1.2 to 6 kV), and frequencies (60 Hz to 1 kHz).¹² The stated assumptions matter for calibration: copper 10 to 14 AWG wire with ETFE insulation at roughly 90 percent design efficiency at 40 kW, skin and proximity effects and inductance and temperature modeled but regolith effects ignored; converters assumed at 95 percent for bidirectional DC-DC, 96.5 percent for bidirectional DC-AC inverters, and 98 percent for AC-AC conversion at fixed frequency; and a constant design efficiency per voltage, meaning converter mass and loss were not re-optimized at every operating point.¹

AC wins for a parts-availability reason, not an electromagnetic one. A transformer changes AC voltage for near-unit efficiency and lets converter designers standardize switch ratings on the transformer's low-voltage side. A DC bus has no such lever: every volt of bus voltage must be blocked by semiconductor switches in series, and rad-hard switch parts effectively top out near 300 V per stage, so a 1.2 to 1.5 kV DC bus already stacks 4 to 5 stages. The team judged more stages prohibitive, capping DC below the 3 to 4 kV regime where DC's cable advantage actually begins.¹ Among AC options, 1 kHz showed the lowest transmission mass in the radial sweep and was carried forward; the study flags that long-distance expansions may need to drop frequency to avoid stability and propagation problems, marked explicitly as to be determined.¹

The expansion studies apply the same loss accounting at larger scale, with a 3 percent cable-loss goal and a 10 percent threshold. The 100 kW numbers assume 8 AWG cable at 1 kHz; 1 MW uses 4 AWG; and the 10 MW inter-regional case assumes uninsulated bare 4/0 cable, where voltage regimes from 30 kV to 240 kV span 4 km to 400 km.¹ These are model outputs for notional systems: no lunar grid hardware existed to validate them when the study was presented, and the study itself notes that distances and power levels for the beyond-Artemis cases are all to be determined.

The strongest case for it

The study's most valuable move is refusing the textbook answer. DC-is-lighter is true in its model above 3 to 4 kV, and the team still chose AC, because a feasibility ceiling in the parts catalogue beats a theoretical mass optimum. That is exactly the kind of constraint that bites real programs, and naming it (no high-voltage rad-hard switches, 300 V per stage, 4 to 5 series stages) converts a vague supply-chain worry into a design driver.¹ The fault-tolerance pricing is equally usable: a program can now trade 50 percent more transmission mass (ring) or 100 percent more (mesh) against single- or dual-line-fault survival, and the study's evolutionary recommendation, start radial and add ties as assets become interdependent, gives a procurement path instead of a big-bang grid.¹ The voltage-versus-distance ladders are a planning scaffold that any lunar power architect can sanity-check a concept against in minutes.

Where a skeptic should push

The load-bearing assumption is the converter model. Fixed efficiencies of 95, 96.5, and 98 percent, held constant across voltage and power, flatter AC and DC architectures differently in ways a real hardware design would not tolerate: magnetics mass at 1 kHz, partial-load efficiency, and converter reliability all move the optimum. The cable model ignores regolith thermal contact, which on the Moon is a first-order radiator-and-ampacity effect, and the loss curves assume a clean 3 percent goal that a costed program may not afford. The architectures also omit protection hardware: a 3 kV lunar DC or AC bus needs switchgear, and none of its mass appears in the transmission totals. Finally, the whole edifice is simulation, presented once at a 2022 conference; the follow-on expansion numbers carry an explicit to-be-determined disclaimer on their inputs.¹ None of that makes the study wrong; it makes its numbers screening-grade, not design-grade.

What grid voltage trades mean for surface power

The non-obvious implication is that distribution voltage on the Moon is set by the semiconductor industry, not by cable physics: the ceiling that forced AC is a parts-availability fact, which means a qualified high-voltage rad-hard switch family would quietly rewrite the whole trade and hand DC back its theoretical advantage.¹ The genuine threat is institutional: if every surface program accepts 3 kV AC because that is what trades today, habitats, rovers, and ISRU plants will standardize around 1 kHz AC interfaces, and the resulting installed base could lock lunar power into an architecture that a future switch generation makes obsolete, the way terrestrial standards outlive their enabling constraints.¹ The opportunity runs the other way: the study's own ladders show that a single insulation and connector family, scaled from 3 kV to 24 kV, covers everything from an Artemis microgrid to a 1 MW industrial feed, so a vendor that qualifies a voltage-scalable lunar cable and switchgear line early would own the standard that the expansion studies assume.¹

The bottom line

Confidence is high that this is what the study found, and moderate that the conclusions survive contact with hardware. As a screening-level result it is sound and internally honest about its assumptions: 3 kV, 1 kHz AC is the mass-optimal feasible point for an Artemis-scale lunar grid because rad-hard switch availability caps DC near 1.2 to 1.5 kV, and every step up in fault tolerance costs 50 to 100 percent more transmission mass. What would confirm it: a breadboard lunar-grid power converter comparison, or a funded parts program delivering 1 kV-plus rad-hard switches, after which the DC case deserves a rerun. What would break it: measured converter or regolith-thermal data showing the constant-efficiency and cable assumptions shifted the AC-DC crossover by a kilovolt or more.