Permanently Shadowed Regions as a Shared Compute / ISRU Hub

Thesis

A permanently shadowed regionPermanently Shadowed RegionA crater floor near the poles that never sees sunlight, sitting at ~40-110 K: both the water resource and an ideal cold sink. (PSRPermanently Shadowed RegionA crater floor near the poles that never sees sunlight, sitting at ~40-110 K: both the water resource and an ideal cold sink.) is, at the same time, (1) the best lunar location to reject compute waste heat and (2) where the water resource is. Those two facts are usually discussed in separate literatures. Put together, they suggest co-locating a surface compute facility with a water-mining ISRUIn-Situ Resource UtilizationProducing usable materials (oxygen, water, propellant) from local resources instead of shipping them from Earth. plant inside or adjacent to a PSRPermanently Shadowed RegionA crater floor near the poles that never sees sunlight, sitting at ~40-110 K: both the water resource and an ideal cold sink., and cascading the compute's rejected heat into ice sublimationSublimationA solid passing directly to vapor without melting; water ice sublimes at ~273 K under vacuum, the basis of the low-temperature water route. before dumping the remainder to the cryogenic PSRPermanently Shadowed RegionA crater floor near the poles that never sees sunlight, sitting at ~40-110 K: both the water resource and an ideal cold sink. sky. The PSRPermanently Shadowed RegionA crater floor near the poles that never sees sunlight, sitting at ~40-110 K: both the water resource and an ideal cold sink. turns a data center's largest liability (heat rejection) into a shared asset, and amortizes the one hard thing both systems need: power delivered into permanent shadow.

Claim 1: A PSR is a superior compute heat-rejection environment

Radiator performance is set by what the radiator sees. On the sunlit equatorial surface a radiator absorbs direct and reflected sunlight and exchanges with ~250-400 K terrain, so its effective sink is warm and its area is large. Inside a PSRPermanently Shadowed RegionA crater floor near the poles that never sees sunlight, sitting at ~40-110 K: both the water resource and an ideal cold sink. there is no direct solar load and the surrounding terrain sits at ~40-110 K, so a radiator can run cold and small, and a given compute load rejects its heat with far less radiator mass. Spacecraft thermal practice already treats a cold, sunless view as the ideal rejection condition (see the spacecraft-thermal-control literature); a PSRPermanently Shadowed RegionA crater floor near the poles that never sees sunlight, sitting at ~40-110 K: both the water resource and an ideal cold sink. is the most accessible place in cislunarCislunarThe region of space between Earth and the Moon (and lunar orbit); the market lunar propellant is meant to serve. space that offers it on the ground. This is a genuine, under-exploited siting advantage for compute, independent of ISRUIn-Situ Resource UtilizationProducing usable materials (oxygen, water, propellant) from local resources instead of shipping them from Earth..

Claim 2: The waste heat then lands exactly where ISRU needs low-grade heat

Per the companion paper, the only ISRUIn-Situ Resource UtilizationProducing usable materials (oxygen, water, propellant) from local resources instead of shipping them from Earth. heat demand a ~315-350 K waste stream can serve is the PSRPermanently Shadowed RegionA crater floor near the poles that never sees sunlight, sitting at ~40-110 K: both the water resource and an ideal cold sink. water route's sublimationSublimationA solid passing directly to vapor without melting; water ice sublimes at ~273 K under vacuum, the basis of the low-temperature water route. chain (~273 K target), and it serves all of it (~1.8 kWhKilowatt-hourA unit of energy (one kilowatt sustained for one hour). The paper reports kWh per kg of oxygen./kg O2Molecular oxygenThe primary product; energy is reported per kg of O2 delivered to cryogenic storage. after the exchanger pinch and effectiveness, ~11% of the route). Because the resource is in the PSRPermanently Shadowed RegionA crater floor near the poles that never sees sunlight, sitting at ~40-110 K: both the water resource and an ideal cold sink., co-location removes the transport problem that would otherwise kill the idea: the heat source, the low-gradeLow-grade heatHeat available only at modest temperature; useful for ice sublimation but useless for high-temperature reduction. sink (icy regolithRegolithThe loose, broken-rock surface layer of the Moon; the feedstock for every non-water route.), and the final cold sink (PSRPermanently Shadowed RegionA crater floor near the poles that never sees sunlight, sitting at ~40-110 K: both the water resource and an ideal cold sink. sky) are all in the same place. A ~12 kW compute load covers a 50 t/yr water plant; on energy grounds a 100 kW facility could serve ~500 t/yr, though delivering heat into a granular icy bed is conduction-limited, so that is an upper bound, not a design. The cascade is: compute silicon → coolant loop → ice sublimationSublimationA solid passing directly to vapor without melting; water ice sublimes at ~273 K under vacuum, the basis of the low-temperature water route. → residual to radiators.

Claim 3: Co-location amortizes the power-into-shadow problem

The decisive unsolved constraint for both systems is the same: a PSRPermanently Shadowed RegionA crater floor near the poles that never sees sunlight, sitting at ~40-110 K: both the water resource and an ideal cold sink. receives no sunlight, so power must come from fission (a reactor sited on a sunlit rim or in the PSRPermanently Shadowed RegionA crater floor near the poles that never sees sunlight, sitting at ~40-110 K: both the water resource and an ideal cold sink.) or be beamed in (demonstrated electrical-to-electrical efficiency only ~11.5% at 10 m). Whichever is chosen, a co-located compute + ISRUIn-Situ Resource UtilizationProducing usable materials (oxygen, water, propellant) from local resources instead of shipping them from Earth. hub pays for that infrastructure once and shares it, rather than two separate programs each solving permanent-shadow power independently. The compute load and the ISRUIn-Situ Resource UtilizationProducing usable materials (oxygen, water, propellant) from local resources instead of shipping them from Earth. load are also complementary in time: compute is a steady baseload, ISRUIn-Situ Resource UtilizationProducing usable materials (oxygen, water, propellant) from local resources instead of shipping them from Earth. can be throttled, which smooths the demand on a reactor.

Quantified benefit, and the probability it is realized

The two benefits are distinct and should not be summed; conflating them oversells the cascade (python -m lpem --benefit, reference: a 50 t/yr PSRPermanently Shadowed RegionA crater floor near the poles that never sees sunlight, sitting at ~40-110 K: both the water resource and an ideal cold sink. water plant + co-located compute):

• Cascade benefit (reuse compute heat in ISRUIn-Situ Resource UtilizationProducing usable materials (oxygen, water, propellant) from local resources instead of shipping them from Earth.): saves ~2.3 t of landed reactor mass (the low-gradeLow-grade heatHeat available only at modest temperature; useful for ice sublimation but useless for high-temperature reduction. heat offset, avoided fission power), against ~1 t of heat-integration hardware (exchanger, transport loop, dust mitigation).
• Siting benefit (put the compute in the PSRPermanently Shadowed RegionA crater floor near the poles that never sees sunlight, sitting at ~40-110 K: both the water resource and an ideal cold sink. for its cold sink): saves radiator mass, scale-dependent. Under an explicit radiator energy balance (IR emission minus absorbed solar minus environmental IR, with a sky view factor for a competently oriented vertical two-sided panel, F_sky nominal 0.5), a PSRPermanently Shadowed RegionA crater floor near the poles that never sees sunlight, sitting at ~40-110 K: both the water resource and an ideal cold sink. saves, over the feasible sampled cases, a median ~10 t of radiator per MWMegawattOne thousand kilowatts of power. of compute (IQRInterquartile RangeThe spread between the 25th and 75th percentiles of a distribution. ~7-16 t/MWMegawattOne thousand kilowatts of power.). An earlier version of this paper reported ~29 t/MWMegawattOne thousand kilowatts of power. and claimed ~30% of sunlit designs could not reject at all; both figures were artifacts of implicitly modeling a horizontal panel staring at warm terrain with zero sky view. With realistic vertical-panel geometry, only ~0.1% of sampled sunlit designs truly cannot reject at 330 K (net flux non-positive at any area) and a further ~0.4% can reject only with a prohibitive area (more than 10x the PSRPermanently Shadowed RegionA crater floor near the poles that never sees sunlight, sitting at ~40-110 K: both the water resource and an ideal cold sink. panel); the corrected advantage is the ~10 t/MWMegawattOne thousand kilowatts of power. mass saving, not widespread sunlit infeasibility. At data-center scale this still exceeds the cascade and remains the actual driver of co-location. It is a siting benefit, not a cascade benefit: a PSRPermanently Shadowed RegionA crater floor near the poles that never sees sunlight, sitting at ~40-110 K: both the water resource and an ideal cold sink. already offers a cheap radiative sinkRadiative sinkA cold surrounding (the ~40-110 K PSR sky) into which a radiator can dump heat; the colder it is, the less radiator mass needed., so the cascade itself does not save radiator mass. (Model: src/lpem/benefit.py, net_rejection_wm2; parameters, emissivity, radiator temperature, absorbed solar, environmental IR, sky view factor, are explicit and tunable.)

Estimating the probability of the benefit. Rather than assert subjective probabilities (P that lunar-surface compute exists, etc.), we compute the break-even joint probability the enabling chain must clear for the cascade hardware to pay for itself: P* = cost / benefit, ~44% at nominal and ~50% once propagated (the ratio is right-skewedRight-skewedA distribution with a long high-side tail, so its mean sits above its median; true of the reactor-loss-bearing routes., so the nominal under-reports it). Two readings follow:

• Conditional on co-location already happening (compute and a water plant both sited at the PSRPermanently Shadowed RegionA crater floor near the poles that never sees sunlight, sitting at ~40-110 K: both the water resource and an ideal cold sink. for their own reasons), the cascade is worthwhile in ~78% of trials (the integration probability clears the propagated break-even), so it is a favorable, positive-expected-value add-on, though a closer call than the earlier ~85% estimate (the cascade prize shrank once the offset model gained a pinch and effectiveness). Design it in.
• As a standalone speculative bet, the full enabling chain (surface compute exists, co-located, water route pursued, integration works) under wide illustrative priors is only ~9%, well below the break-even, so its expected value is negative (~-0.8 t). Do not justify co-location by the cascade alone.

The decision structure is therefore clear: co-location is justified (if at all) by the compute siting economics (roughly ten tonnes per MWMegawattOne thousand kilowatts of power. of radiator mass and shared power-into-shadow infrastructure) and the ISRUIn-Situ Resource UtilizationProducing usable materials (oxygen, water, propellant) from local resources instead of shipping them from Earth. heat cascade is a cheap, sensible bonus to capture once you are already there, not a reason to go. The break-even model (src/lpem/benefit.py) makes every assumption explicit and tunable.

What this is not, and what would sink it

This is a siting argument, not an engineering design, and it rests on assumptions a skeptic should press:

• Why surface compute at all? The thesis is empty unless there is an independent reason to run compute on the lunar surface (autonomy and real-time ISRUIn-Situ Resource UtilizationProducing usable materials (oxygen, water, propellant) from local resources instead of shipping them from Earth./robotics control, proximity to a future lunar data economy, or latency-tolerant batch compute that wants a free cold sink). If all useful compute lives on Earth or in orbit, there is no waste heat at the PSRPermanently Shadowed RegionA crater floor near the poles that never sees sunlight, sitting at ~40-110 K: both the water resource and an ideal cold sink. to cascade. This is the load-bearing assumption.
• Power into shadow is genuinely unsolved. Co-location amortizes it but does not solve it; if neither in-PSR fission siting nor efficient beaming matures, the whole hub is blocked, and so is standalone PSRPermanently Shadowed RegionA crater floor near the poles that never sees sunlight, sitting at ~40-110 K: both the water resource and an ideal cold sink. water miningPolar water-ice miningSublimate water ice out of permanently shadowed regolith at ~273 K and cold-trap the vapor; the only low-temperature route..
• Heat transport over the last tens of meters (racks to the working face) still costs mass and pumping power, and the regolith-side heat exchanger operates in abrasive dust at cryogenic temperatures with no flight heritage.
• The energy prize is modest (~11% of one route); the real argument is shared infrastructure and a free, ideally-located heat sink, not a large kWhKilowatt-hourA unit of energy (one kilowatt sustained for one hour). The paper reports kWh per kg of oxygen. saving.
• Dust on radiators in a worked mining environment degrades the very rejection advantage Claim 1 depends on.

What would make this a real study

A defensible follow-on needs: a concrete surface-compute demand case (kW and why it is on the Moon); a co-located mass model (reactor + radiators + compute + mining, shared vs separate) showing the amortization quantitatively; a heat-transport design from racks to the sublimationSublimationA solid passing directly to vapor without melting; water ice sublimes at ~273 K under vacuum, the basis of the low-temperature water route. face; and a comparison against the obvious alternative, solar-thermal process heat from a sunlit rim with uphill resource haul. The lpem energy and power/landed-mass models are the natural backbone for the mass-amortization piece.

Relationship to the other papers

• WASTE-HEAT-OFFSET.md, the narrow, quantified, Second-Law-safe energy claim this paper builds on. That paper stands without this one.
• MANUSCRIPT.md, the underlying common-basis energy model and the route rankings.