Economics of In-Space Industry and Competitiveness of

Lunar-Derived Rocket Propellant

Author:

Philip T. Metzger (corresponding author)

Florida Space Institute, University of Central Florida, 12354 Research Parkway, Suite 200,
Orlando, FL 32826; philip.metzger@ucf.edu.

Key Words: lunar mining, space mining, lunar resources, rocket propellant, space economics

Abstract

Economic parameters are identified for an in-space industry where the capital is made on one
planet, it is transported to and teleoperated on a second planet, and the product is transported off
the second planet for consumption. This framework is used to model the long-run cost of lunar
propellant production to help answer whether it is commercially competitive against propellant
launched from Earth. The prior techno-economic analyses (TEAs) of lunar propellant production
had disagreed over this. The “gear ratio on cost” for capital transport, 𝐺, and the production mass
ratio of the capital, 𝜙, are identified as the most important factors determining competitiveness.
The prior TEAs are examined for how they handled these two metrics. This identifies crucial
mistakes in some of the TEAs: choosing transportation architectures with high 𝐺, and neglecting
to make choices for the capital that could achieve adequate 𝜙. The tent sublimation technology
has a value of 𝜙 that is an order of magnitude better than the threshold for competitiveness even
in low Earth orbit (LEO). The strip mining technology is closer to the threshold, but
technological improvements plus several years of operating experience will improve its
competitiveness, according to the model. Objections from members of the aerospace community
are discussed, especially the question whether the technology can attain adequate reliability in
the lunar environment. The results suggest that lunar propellant production will be commercially
viable and that it should lower the cost of doing everything else in space.

Primary Symbols

𝐶fin Cost of financing

𝐶K Cost of developing and fabricating capital
𝐶ops Cost of operations over the life of the capital
𝑐R Reliability cost factor
𝐶tr Cost of transporting capital to the lunar surface
𝑐X Cost per kg of lunar propellant delivered to location X
𝐸𝑅 Scale of effort required to increase reliability
𝑓 Specific finance cost
𝑔 Surface gravity of Earth
𝐺 Gear ratio “on cost” for delivering lunar capital
 𝐺Y−X Gear ratio from location Y to location X
IMF Inert Mass Fraction
𝐼sp Specific Impulse
𝐿K Cost per kg of launching capital from Earth to LEO
𝐿p Cost per kg of launching propellant from Earth to LEO
𝑀K Mass of capital
𝑀p,X Mass of lunar propellant delivered to location X over the life of the capital
𝑅 As-built reliability of the capital
𝑅max Maximum achievable reliability of the capital
𝑅0 Baseline reliability of the capital
𝑇𝐾 Transportation rate (cost per kg) of capital to the lunar surface
𝑥 Launch-normalized equipment cost
ΓX Propellant use ratio for delivery to location X
∆𝑣Y−X “Delta-v” (from the rocket equation) from location Y to location X
𝜁 Capital development and fabrication cost rate ($/kg)
𝜁D Capital development cost rate ($/kg)
𝜁F Capital fabrication cost rate ($/kg)
𝜅 Specific capital fabrication cost
𝜆 Specific labor cost
𝜉 Launch-normalized finance cost
τ Specific capital transportation cost
𝜙 Production mass ratio
𝜒 Launch-normalized capital cost
𝜓X Cost ratio of lunar vs. terrestrial propellant at location X
𝜓0 Pre-delivery cost ratio of lunar (on LS) vs. terrestrial (in LEO) propellant
𝜔 Launch-normalized operations cost

1. Introduction

Mining lunar water ice to produce rocket propellant may be an early business in cislunar
industry. An example customer is a firm that owns and operates geostationary spacecraft and
purchases boost services using lunar-derived propellant to place their satellites in operational
orbit faster and at lower cost. Many see this “cislunar water economy” as vital for space
development because it will have the scaling power of Earth’s consumer markets, unshackling
space development from the slower pace of government-led space exploration. Several techno-
economic assessments (TEAs) of lunar-derived propellant have made either positive [1–4] or
negative [5–7] predictions for its near-term economic viability. Others have expressed skepticism
whether lunar ice mining can ever be economically viable because reusable launch vehicles will
lower the cost of launching resources into space directly from Earth. Charania and DePascuale
wrote,

The price per kilogram for delivery of propellant from the lunar surface to a GEO
[geostationary orbit] satellite tug customer does not provide an attractive
alternative as compared to launch from Earth. The price per kg to send propellant
from the lunar surface to GEO is extremely prohibitive…[T]he results of this
 study indicate a significant challenge in competing with propellants delivered
from Earth [5].

Jones et al. agreed,

… lunar ISRU [in-situ resource utilization, i.e., lunar-derived] propellant is 97%

more expensive than Earth-based propellant… [I]t is unlikely ISRU is a cost-
effective approach to providing propellant in cis-lunar space in the near term [6].

On the other hand, Bennett et al. reassessed the Jones et al. study by improving the economies of
scale of the lunar surface hardware and concluded,

We have proposed four alternative architectures that…reduce the cost of

propellant at the cislunar aggregation point to well below the cost of commercial
transport posited in Jones et al… We have demonstrated that there is a large space
of architectural choices and that one region can be an order of magnitude better
than another. This supports the potential for lunar ISRU propellant to make a
significant contribution to future large-scale operations in cislunar space [2].

The divergent TEAs are difficult to synthesize. Shisko reviewed them and wrote,

…the wide range of quantitative results is a consequence of the different

assumptions, parameters, and analysis methods used in the studies. Because of
this, it is difficult to compare results from different studies on a truly equal
footing, i.e., apples-to-apples [8].

Part of the confusion is because these TEAs were not really on the same topic: they were
assessing four different technologies (strip mining, borehole sublimation, tent sublimation, and
excavation with beneficiation). Some of them concluded that lunar propellant manufacture is not
economic, but that was illogical because they did not study the economics of lunar propellant
manufacture per se but only the economics of one possible technology operating in that sector.
The studies sometimes made basic scaling errors or used inadequate estimating methods that
skewed the results, or they made choices for transportation systems that would guarantee a
negative outcome, as I discuss below. Another basic problem is that the studies implied that the
economics are static or that they change over a timescale that is too long for decisionmakers to
care about, so they studied only point-designs at a single moment in time yet made longer-term
pronouncements on that basis.

I think this confusion arose because we skipped an important step: studying the economics of the
in-space propellant economy as a sector before doing TEAs on the possible technologies. We
need a “spherical cow model” before adding the hoofs and the horns. This paper will take that
approach by writing the equations for the primary, empirically supported economic factors that
will determine the cost of space-mined propellant versus Earth-launched propellant and studying
how those forces will drive in-space economics. This should produce greater confidence than a
focus on the technological details to decide whether space mining is commercially feasible or
not.
This study has the following goals. First, it will identify parameters that give insight into in-
space industrial development. Rather than simply the costs of capital and labor, it recognizes the
large role of mass and the rocket equation, so it focuses on mass-specific capital and mass-
specific labor that have been normalized by launch cost, and it incorporates gear ratios from
orbital dynamics. Second, this paper will assess the long-run trends in these metrics according to
four economic factors: the cost of reliability, the experience or learning curve, economies of
scale, and economies of scope. Third, it will try to provide a clear answer to the question whether
lunar ice mining can be economic or not as a sector, and more narrowly, whether the decreasing
cost of launch from Earth will inherently drive space resources out of business. Fourth, it will
look at some of the details in the prior TEAs to try to explain why and how they disagreed. Fifth,
it will hopefully create insight into what space companies need to do to make lunar mining more
profitable. Sixth, it will show that policymakers must not ignore lunar resources because their
economic potential will become disruptive to the status quo.

2. Assumptions and Ground Rules

For simplicity and clarity, some major issues will need to be ignored. This paper will focus only
on liquid oxygen/liquid hydrogen (LOX/LH2) propellant, although SpaceX is building their
transportation architecture around LOX/methane. Methane could be brought from Earth even
while LOX is sourced from the Moon. LOX is 89%wt of the stoichiometric mass of water and as
low as 80% of the mixing ratio in rocket engines, and an 11% to 20% error is not bad for a
spherical cow model so this analysis is still valid. It is also possible to manufacture methane and
other space-storable propellants like hydrogen peroxide using space resources, and some
companies are already working on the technologies. Furthermore, if LOX/LH2 turns out to be
economically superior when using lunar resources, it is not too late for companies to switch to
these propellants for many future applications. On the other hand, the United Launch Alliance
and Blue Origin have already focused on future use of lunar-sourced LOX/LH2. Introducing
architectural variations would make the model too complex for the present purposes.

This paper will also ignore the possibility of transportation architectures using propellant depots
in cislunar space [9–12]. The business case for depots is broader than just propellant sale and
would complicate the analysis. For specificity this paper will focus on the concept of launching
one load of propellant off the Moon at a time to provide a boost to a spacecraft then returning to
the Moon for another load [13]. The model predicts cost savings from economies of scale in
future years as the propellant market grows, and those economies would likely be achieved
because companies shift toward building propellant depots with their broader scope, greater
capital infrastructure, and better scaling, but exactly how these economies of scale and scope are
achieved does not need to be explicit in this type of model and would introduce too many
variables that make the analysis less certain.

This paper will furthermore ignore the many architectural choices for cislunar space
transportation in addition to propellant depots, including the use of aerobraking, non-chemical
propulsion technologies, mass drivers or sling launchers from the lunar surface, etc. Solar
electric propulsion (SEP) will be included but only as an example of a high specific impulse
system to show how this affects the economics. The details of SEP or other propulsion systems
would introduce too many complicating variables. The goal of this paper is not to analyze
architectures but only to study the economic feasibility of lunar resources using the simplest
cases.

This study will also restrict its focus to the costs to produce and deliver propellant, not the
pricing of the propellant and resulting profit, which involves additional economic dynamics.

Additionally, this paper will ignore the possible role of humans on the Moon working in this
industry. The Jet Propulsion Lab’s recent Robotic Lunar Surface Operations 2 study [14]
developed lunar ice mining architectures in which humans visit periodically for short stints only
to repair hardware. That may be a good model if the cost savings from repairing rather than
replacing hardware is greater than the extra expense. There are companies developing business
cases that include humans on the Moon, whether for tourism or industry, and national space
agencies may pay the expense of humans on the Moon for national goals and exploration. It
would be strategic for national space agencies to focus on developing lunar surface industry as
one of the activities for their lunar astronauts. Although I am an advocate for humans working in
lunar surface industry, including humans in this analysis would complicate it. To justify the cost
of habitats, life support systems, construction systems for radiation shielding, higher safety
standards in transportation, etc., there must be a larger business case than just propellant mining,
and that introduces many additional variables. Instead, this paper will assume pure teleoperation
of the robotics with modularized assets that can be replaced by teleoperation recognizing that as
the industry grows there will be increasing opportunities for synergy between human and robotic
lunar activities.

Autonomy of the robotics will therefore be important for the development of the industry. Some
initial level of autonomy is needed for smooth operation despite the 3-second round trip time
delay for communications with Earth, then loosely supervised or full autonomy can be phased-in
over time for greater labor efficiency. This is discussed in Section 9.2.

This model assumes that each of the proposed lunar propellant production systems will perform
as described by their individual TEAs. The technologies are mostly at low technological maturity
(some that are at mid-range technological maturity are mentioned in Section 6.6) and there is risk
that some may not work or may not perform as well as expected. On the other hand, ongoing
innovation could improve performance to better than currently expected. There is additional risk
derived from uncertainties in the geology and physical state of the lunar ice. Missions to the
Moon to study the resource are already planned, and many groups globally are working to
mature and test the technologies, so the technology assumptions in this model can be replaced by
clarity over time.

3. Initial Considerations

3.1. Plausibility in Physics

Because of constraints in physics, the best payload mass fraction for conventional rocket
technology launching off the Earth to GTO is about 2%. For launching off the Moon with its
weaker gravity to GTO, the payload mass fraction can be about 48%, or 24 times higher. If this
were the only difference between the Earth and the Moon, lunar propellant would be 24 times
cheaper than Earth-launched propellant in GTO. However, on Earth it is trivial to acquire the raw
materials that go into rocket propellant (e.g., water), whereas on the Moon they must be mined as
ice in an extremely harsh environment. The Moon’s capital (including its launch vehicle to carry
propellant off the Moon) can be made on Earth and transported to the Moon, so the Moon does
not need its own industrial supply chain, but there will be capital transportation costs. Humans
can teleoperate the rockets and robots from Earth, so the Moon does not need its own workforce.
(See the Discussion section about teleoperation.) However, operations will be less efficient when
it comes to troubleshooting and repairing broken hardware with no humans on-site. With fewer
successful repairs there will be more replacement of hardware, which adds to the capital
acquisition and transportation costs. Maybe the lunar capital can be designed for better
modularity with telerobotic replacement to minimize the extra mass, but still it must be designed
to operate with adequate reliability in a harsher environment than Earth.

The question whether lunar propellant can be competitive comes down to this: can we (1)
transport capital to the Moon, (2) tele-support its operation on the Moon, and (3) work with
difficult raw materials in a harsh environment, with a total economic penalty that is less than a
factor of 24 so it does not eat up the entire positive margin afforded by the physics? A factor of
24 is quite high (that is, a 2,300% cost increase over building and launching rockets from Earth),
so there is no a priori reason to reject its plausibility.

3.2. Comparative Advantage and Environmental Externalities

Before beginning it is important to expose a common fallacy: the belief that lunar-derived
propellant needs to be cheaper than Earth-launched propellant to be competitive. It needs only a
comparative advantage, not an absolute advantage, since launching rockets entails a significant
opportunity cost and buying propellant in cislunar space from a lunar mining operation may be
comparatively less expensive than losing the opportunity to launch a more lucrative payload. In
general, Earth-launched propellant does not compete against lunar propellant; it competes against
the value of other payloads that can be launched from Earth.

This may not be relevant in the near or mid-term for a company like SpaceX, which is
developing a super-sized rocket and expects a glut of launch capacity so they may want to launch
the propellant themselves. However, it will still be important for companies competing with
SpaceX and it becomes strategically crucial for SpaceX in the longer term. This paper is not
based on SpaceX’s plans, but they are part of the context for assessing lunar resources so the
reader may benefit from a brief overview.

SpaceX’s overarching goal is to build a city on Mars. Founder Elon Musk set the timeline at one
million settlers on Mars by 2050 [15]. He said, “I’m only personally accumulating assets in order
to fund this” [16]. Once people and manufacturing equipment start landing on Mars, it will take
years of buildup before a city can break even financially, so the effort must be financed at
extreme cost. Musk has stated this Martian city may cost between $100B and $10T [17],
possibly beyond his ability to finance so every advantage is needed. He said, “ultimately this will
be a huge public-private partnership” [16]. Supporting a population of a quarter million to a
million people who cannot yet fully support themselves will be expensive, so accelerating the
shipments of capital to make Mars self-sustaining sooner is highly leveraged to reduce net costs.
The time-value of cargo to Mars also includes intangibles such as extra security for the
population by having more industrial capacity on-site sooner, creating progress faster before
political opposition can galvanize on Earth (a benefit from the perspective of the investors), and
seeing the project completed in the lifetimes of the investors.

As a hypothetical illustration, assume that lunar propellant has a cost in LEO that is 8% more
expensive than launching the propellant from Earth. The modeling discussed below shows that,
if SpaceX applies 50% mark-up for profit in their launch price for the lunar capital, that increases
the cost of the lunar propellant by only 16% (because launch cost is only one of the cost drivers).
If the lunar mining company marks up their sale price another 12% for their own profit, then
SpaceX can purchase the propellant at 40% higher cost (1.08 × 1.16 × 1.12 = 1.40) than
launching it themselves. Thus, by launching one load of capital in lieu of one load of propellant,
SpaceX obtains a net 50% - 40% = 10% profit from the deal even though they could have
launched the propellant at lower cost. The lunar capital then produces on the order of 100x more
mass of propellant than the mass of capital (discussed below), so SpaceX can buy the additional
propellant and use their freed-up launch capacity to send more capital to Mars. If the extra loads
of capital each create only 20% annual return in value on Mars, which compounds to 48% per
synodic period (Mars departure windows from Earth), then sending a load of capital one synodic
period sooner is an immediate net gain even if the lunar propellant were 40% more expensive
(48% - 40% = +8%). This benefit compounds continuously because the extra capital on Mars
produces value through many years, and simple modeling shows this can dramatically cut the
cost of Mars settlement. Capital on the Moon enables production of rocket propellant; capital on
Mars enables production of goods that Mars residents consume plus materials to make more
capital, so SpaceX avoids shipping costs. In both cases, the capital results in a greater mass of
goods than the capital itself, which enables trade so comparative advantage is a factor.

Also, lunar propellant mass is obtained higher in Earth’s gravity well, so staging for Mars can be
done in Moon Distant Retrograde Orbit (DRO) instead of LEO (discussed below), which can
reduce the propellant cost by 30% compared to the cost of bringing all the propellant from the
Moon to LEO for refueling.

Furthermore, the launch rate in Musk’s Mars settlement plan is perhaps 7x higher [18] or more
[19] than rates predicted to cause significant damage to Earth’s ozone. There are also concerns
with carbon black emissions of hydrocarbon-fueled rockets affecting the thermal budget of the
upper atmosphere (this effect with methane has not been quantified, yet) [20,21]. These concerns
are a risk to Mars settlement because governments might limit the launch rate or might
internalize the external costs by imposing environmental impact fees that drive up the launch
cost, erasing any cost advantage of terrestrial propellant. The calls for this have grown
exponentially in the past three years [22–27]. Reducing the launch rate by up to 9x by using
lunar propellant will therefore reduce the political and environmental risks of a Mars settlement
campaign. For these reasons, even if SpaceX expects a glut of launch capacity, it is strategic to
plan on using lunar resources. Understandably, proponents of Mars settlement will reject putting
the complexity of lunar mining into the critical path to Mars, but if other companies (and/or
governments) develop lunar propellant at their own risk so it becomes available for Mars
settlement, then failing to take advantage of lunar propellant would only make Mars settlement
more costly with higher risk. Furthermore, this paper will show that lunar propellant can gain an
absolute advantage even without these considerations.

Of course, lunar propellant does not rely on Mars settlement as its only business case. There are
other space transportation companies like the United Launch Alliance (ULA) and Blue Origin
that have identified lunar propellant as a near-term business strategy in cislunar transportation
[28–30]. OrbitFab is already providing in-space water to NASA as a customer at the
International Space Station and is signing on other customers for their services, including
Astroscale’s LEXI spacecraft that will be equipped with OrbitFab’s refueling port. Daniel Faber
(CEO of OrbitFab) points out that in the past 10 years about 200 spacecraft worth an aggregate
$100B had to be discarded in space due to running out of propellant. There are also geopolitical
considerations that should create government customers for in-space propellant at higher than
launch market prices [29]. Listner [31] recently argued the geopolitical importance and urgency
in developing lunar resources. The business risk for a lunar propellant firm would be lowest if
they gain an absolute advantage over the Earth-launched propellant, but these broader
considerations suggest they could weather a short-term glut of launch services even without it.

4. Model Development

4.1. Basic Approach

A complication arises because the amount of product sold, both terrestrial and lunar, affects the
economy of scale and learning curve for each business, but the amount sold by each depends on
relative pricing that differs with location across cislunar space and depends on other economic
factors. Parsing the market demand by location would be too complicated for this simple model,
so instead the demand for both rocket launches and in-space propellant are applied exogenously
using three cases: one with optimistic growth of the launch market, the second with moderate
growth, and the third pessimistic. The analysis assumes that the launch market and the in-space
propellant market are related: when launch rates increase, then in-space activity and need for
propellant increases in rough proportion. The launch costs and the lunar industry costs then
develop independently based upon their own market models through learning curves and
economies of scale and scope. I attempted to choose reasonable index parameters for each of
these effects based on a thorough literature review given in the Supplementary Material. The
model then calculates the costs of lunar and terrestrial propellant for “year 1” (the first year of
product sales) and predicts which one’s cost will drop faster per those index parameters over the
next thirty years. This will indicate whether the competitiveness of lunar propellant may change
over a timescale that is important to policy planners. For example, if lunar propellant is predicted
to be more costly than terrestrial propellant in the first year of commercial sales, but the
modeling shows it should become much less costly than terrestrial within only five or ten
additional years, then it would be in a national space agency’s interest to continue developing the
technologies to spin them off to commercial firms to provide this service when the agency needs
them. The agency may gain deep cost savings for (e.g.) Mars missions ten or twenty years in the
future despite the poor cost comparison in the first year of commercial sales.

4.2. Cost Equation

This model focuses on Long Run Average Cost (LRAC) so appropriate averaging methods are
employed. The average cost ($/kg) for lunar-derived propellant when the product is sold at the
lunar surface may be approximated as,

𝑐LS = {𝐶K + 𝐶tr + 𝐶ops + 𝐶fin }⁄𝑀p,LS

= 𝜅+τ+𝜆+𝑓
(1)

where 𝐶K = cost of capital development and fabrication, 𝐶tr = cost of transporting the capital
from Earth to the lunar operations site, 𝐶ops = operating cost over the life of the hardware,
𝐶fin = finance costs, and 𝑀p,LS = the total mass of propellant (p) produced at the lunar surface
(LS) over the operational life of the hardware. Finance costs could have been included where
incurred as part of the capital acquisition, transportation, and labor, but it was written separately
to provide some insight. For a robotic lunar industry where humans operate the hardware
telerobotically from Earth, 𝐶ops is mainly just labor with overhead. The purchase of
telecommunications bandwidth between the Earth and the Moon is not a separate cost center
because it is counted as part of the overhead on labor. The labor supply is assumed to have a
constant wage rate and to scale inelastically as required to match the propellant demand. The
aerospace workforce moves fluidly between companies so both the lunar propellant industry and
the competing Earth-launched propellant business share the labor pool and wage rate. Spares
procurement has been included in capital costs rather than operations costs since optimizing
reliability entangles the amount of spares with the cost of capital (see Section 4.3). Energy is not
a separate cost center because the individual TEAs for lunar ice mining always include energy
generation equipment as part of the capital and the labor to operate that equipment is part of the
labor. The (product mass-) specific capital, labor, transportation, and finance costs are 𝜅, 𝜆, 𝜏,
and 𝑓.

Transporting the product from LS to other locations in cislunar space is considered here part of
the firm’s business, so transportation hardware such as a reusable lunar lander (RLL) and
possibly a space tug or orbital transfer vehicle (OTV), their fabrication, initial delivery to point
of operation, operational costs, and associated finance costs are already included in 𝜅, τ, 𝜆, and
𝑓, but the act of transporting the product in those vehicles uses up a portion of the product –
often the vast majority of it – so this requires an additional factor that does not occur in terrestrial
economics. Simplistically, the mass of delivered product becomes,

∆𝑣LS−X
𝑀p,X = 𝑀p,LS [(1 + IMF) Exp (− ) − IMF] = 𝑀p,LS /𝐺LS−X
𝑔 𝐼sp
(2)

per the Tsiolkovsky rocket equation, where 𝑀p,X = the mass of propellant at location X after
transport, IMF = the inert mass fraction of the spacecraft, ∆𝑣Y−X is the delta-velocity from
location Y to location X according to orbital dynamics, 𝐺Y−X is the “gear ratio” from Y to X,
𝐼sp = the specific impulse of the rocket propellant, and 𝑔 =9.8 m/s is Earth’s surface gravity,
which serves to define the customary units of 𝐼sp . The notation X and Y can be replaced by LS =
lunar surface, LEO = low earth orbit, GTO = geostationary transfer orbit, GEO = geostationary
orbit, EML1 = the Earth-Moon Lagrange point 1, LLO = low lunar orbit, DRO = Moon distant
retrograde orbit, etc. For an OTV and an RLL hauling rocket propellant or capital sequentially,
the gear ratios for each are multiplied together. To account for the propellant the RLL and OTV
need to return from their points X where they unload the goods back to their points of origin Y,
replace IMF with the effective IMF,

∆𝑣X−Y
IMF′ = IMF Exp ( )
𝑔 𝐼sp
(3)

The gear ratio in orbital dynamics is the ratio of the mass of hardware and propellant before
versus after moving from location Y to location X per the rocket equation. If you want to deliver
1 kg to location X, you need to start with 𝐺Y−X × 1 kg at location Y. The cost of propellant at
location X becomes,

𝑐X = (𝜅 + τ + 𝜆 + 𝑓) 𝐺LS−X
(4)

𝐺Y−X is a mass ratio through Eq. 2, but it can be interpreted as a cost ratio through the price per
mass of propellant. Generalizing it as such enables analysis of architectures using different and
mixed transport systems with varying base costs, propellants, reusability, etc., as discussed
below.

The condition for lunar derived propellant to have lower cost than terrestrial propellant at
location X is,

(𝜅 + τ + 𝜆 + 𝑓)𝐺LS−X < 𝐿p 𝐺LEO−X

(5)

where 𝐿p is the (mass-)specific launch cost of propellant from Earth to LEO in $/kg. The cost of
manufacturing and handling propellant on Earth and the operations cost of propellant transfer in-
space are negligible, so the right-hand side of Eq. 4 is only the transportation cost.

Capital equipment costs for spaceflight hardware are usually estimated by mass, so 𝜁D =
development cost of equipment per kg (mass of the equipment, not of the product it produces),
and 𝜁F = fabrication cost of equipment per kg. For the first hardware set, the cost is 𝐶K =
(𝜁D + 𝜁F )𝑀K , where 𝑀K = the mass of the equipment, but for additional production including
spares and replacements it is 𝐶K = 𝜁F 𝑀K. For now, we write 𝜁 = 𝜁D + 𝜁F , but we address the
ratio of spares to originals below. Then, 𝜅 = 𝜁𝑀K /𝑀p,LS .

The capital’s transportation rate (cost per kg) is 𝑇𝐾 = 𝐿K 𝐺K,LEO−LS and the (product mass-)
specific transportation cost is τ = 𝐿K 𝐺K,LEO−LS MK /𝑀p,LS. The launch cost 𝐿K and the gear ratio
𝐺K,LEO−LS refer to the transportation system that transports capital. This could be different than
the transportation system that launches the competing terrestrial propellant and delivers it to
location X. We define the capital’s “gear ratio on cost,”

𝐿K 𝐺K,LEO−LS
𝐺=
𝐿p
(6)

which reduces to the ordinary “gear ratio on mass” (as we could call it) when a common
transportation system is used to launch and deliver both capital and propellant, 𝐿K = 𝐿p .

We normalize the equation by 𝐿p since it is a convenient cost standard. Substituting the capital
and transportation terms into Eq. 5,

𝜁𝑀K 𝐺 𝑀𝐾 𝜆 𝑓
{ + + + } 𝐺LS−X < 𝐺LEO−X
𝐿p 𝑀p,LS 𝑀p,LS 𝐿p 𝐿p
(7)
which becomes

[(𝑥 + 𝐺) 𝜙 −1 + 𝜔 + 𝜉] ∙ ΓX < 1
(8)

where we introduced dimensionless parameters: production mass ratio, 𝜙 = 𝑀p,LS /M𝐾 ; launch-
normalized equipment cost, 𝑥 = 𝜁/𝐿 p ; launch-normalized operations cost, 𝜔 = 𝜆/𝐿 p; launch-
normalized finance cost, 𝜉 = 𝑓/𝐿 p ; and propellant use ratio, ΓX = 𝐺LS−X /𝐺LEO−X. For brevity
we may refer to these quantities without writing “launch-normalized” every time, the context
making it clear. If 𝑥, 𝜔, and 𝜉 are set to zero, the remaining term 𝐺𝜙 −1 ΓX is the inverse of the
“propellant payback ratio” defined by Pelech, et al [7]. The technology parameters are 𝑥, 𝜙, and
𝜔. The transportation parameters are G and ΓX.

We further define the (launch-normalized) capital cost,

𝜒 = (𝑥 + 𝐺) 𝜙 −1
(9)

which includes both fabrication and transportation of the capital and is specific to the mass of
product rather than the mass of capital. The cost ratio of lunar vs. terrestrial propellant at
location X is

𝜓X = (𝜒 + 𝜔 + 𝜉) ΓX
(10)

or we can write the pre-delivery cost ratio, 𝜓0 = (𝜒 + 𝜔 + 𝜉). Lunar propellant has absolute
advantage at X when 𝜓X < 1. Sometimes it is more convenient to plot 𝜓0 so the absolute
advantage is when 𝜓0 < 1/ΓX . The rest of this paper will study how these dimensionless
parameters behave.
4.3. Cost of Reliability

Low launch costs help make lunar propellant more competitive by making it inexpensive to
replace failed hardware, so ultra-high reliability is not needed for lunar mining where high
reliability would be expensive to achieve. Empirical data show that the hardware cost rate 𝜁
scales exponentially with reliability. A common saying is that NASA pays 90% of the cost for
the last 5% of reliability (the numbers vary with the telling). For space agencies, the politics of
mission failure drive even higher reliability than economics do. To help alleviate this, NASA has
been implementing commercial provider programs that create some political insulation,
including Commercial Lunar Payload Services (CLPS), which aims to put higher risk but lower
cost uncrewed landers on the Moon. This is expected to increase the science return on the dollar
despite a higher failure rate [32] due to the non-linearity of the reliability cost function. On the
other hand, a commercial operation needs to attract customers, who will want assurance of
reliable boost or reliable delivery of propellant. Here the assumption is that a commercial space
mining company will allow lower reliability at the hardware element or subassembly level as a
cost savings but will manage spares and repairs to provide high reliability in delivery of the final
product.

Mettas [33] wrote a cost-reliability model by fitting to empirical cost-reliability data,

𝑅 − 𝑅0
𝑐R = exp [(1 − 𝐸𝑅 ) ]
𝑅max − 𝑅
(11)

where 𝑐R is the reliability cost factor that is multiplied onto 𝜁, 𝑅0 is the baseline reliability
corresponding to the baseline cost, 𝑅 is the reliability as built, 𝑅max ≅ 100% is the maximum
achievable reliability. 𝑅0 is the baseline reliability corresponding to the baseline cost model. The
baseline is built using baseline quality components and baseline design resilience (internal
redundancies and robustness of the internal architecture) and is subjected to a baseline amount of
testing, so as-built reliability and cost can go either higher or lower than baseline. 𝐸𝑅 is a
parameter between 0 and 1 that scales the effort of improving reliability. Stancliff, et al. [34,35]
used 𝐸𝑅 = 0.5 and 𝐸𝑅 = 0.95 for lunar rovers and obtained similar results in each case. Here, the
model uses 𝐸𝑅 = 0.5.

Jones et al. [6] assumed 10% of the mass of lunar propellant production hardware must be
replaced each year over a 10-year operational life, which is 𝑅 = 0.50 (half of the hardware that
is produced including spares fails during the baseline period). This can be checked using the
Feasibility of Objective Technique from Military Handbook (MIL-HDBK)-338B [36]
benchmarked by spacecraft data having 𝑅 = 0.92 for 10-year operation provided by Figure 3 of
Yang et al. [37] This estimates that a lunar strip mining technology using nuclear power as
described by Bennett et al. [2] would have mass-weighted 𝑅 = 0.36 for 10-year operation with
an annual spares mass of 28%, much higher than the value of Jones et al. This assumes a failed
element must be replaced entirely, so if we assume most failures occur in modules that are
10%wt apiece while others like RLL engines must be replaced in whole, this improves the mass-
weighted reliability to 𝑅 = 0.72 and the annual spares mass to 3.9%. This reliability value is
dominated by the technological immaturity of the nuclear reactor and chemical processing (water
cleanup and electrolysis) elements and the excavators operating full-time in harsh lunar dust.
Replacing the nuclear reactor with the solar power system described by Sowers [4] improves the
mass-weighted reliability to 𝑅 = 0.76 with annual modularized mass of spares at 3.2%. For the
tent sublimation technology described by Sowers [4], this technique estimates base reliability
𝑅 = 0.80 with annual modularized mass of spares at 2.6%. The model will use 𝑅 = 0.78 for
both methods. Achieving this degree of modularity with robotic replacement is crucial for these
annual spares masses lower than the Jones et al. estimate. Additional cases will be run with 𝑅0 =
0.20 to 𝑅0 = 0.92 to test sensitivity.

The total mass of capital that must be fabricated and transported to the Moon includes spare
modules and spare system elements. This scales as 1/𝑅 in the long-run average, so 𝜁 = 𝜁D +
𝜁F /𝑅. The cost of developing, fabricating, and transporting hardware becomes,

𝜁D 𝑅 − 𝑅0 𝑀K
𝐶K + 𝐶tr = (𝜁D + ) 𝑀K exp [(1 − 𝐸𝑅 ) ]+ 𝑇
𝑅 𝑅max − 𝑅 𝑅 𝐾
(12)

where 𝑇𝐾 = 𝐿K 𝐺. This can be minimized by varying 𝑅. Fig. 1 plots example cases for four
transportation rates. The minima are marked by solid dots, which define the cost-optimized
reliability 𝑅opt . 𝑅opt occurs at lower 𝑅 when 𝑇𝐾 is lower, so the cost of equipment fabrication
drops when the cost of launch drops.

Figure 1. Sum of capital acquisition and transportation costs vs.

reliability for four transportation rates to the lunar surface. Solid
dots are the optimum reliability for each curve. Calculated from
baseline reliability 𝑅0 = 0.78.

Fig. 2 shows 𝑅opt vs. 𝑇K, and fig. 3 shows optimized reliability-cost factor 𝑐R versus 𝑇K. The
parameter choices appear reasonable: if a lunar mining rover with this baseline reliability were
brought up to 96% reliability through improved design and extra testing, it would increase the
development and fabrication cost by a factor of 8. However, lowering the reliability to optimize
for low-cost transportation results in only 12% savings. The main finding is that it is unnecessary
to make costly improvements to the hardware to gain higher reliabilities typical of NASA
exploration missions, so the environmental harshness of mining in lunar dust will not drive costs
exceedingly high.

Figure 2. Cost-optimized reliability 𝑐R vs. transportation cost 𝑇𝐾

for 𝑅0 = 0.78.
 Figure 3. Optimized Reliability Cost Factor vs. Transportation
Cost for 𝑅0 = 0.78.

4.4. Experience Curve

There is a deep economics literature on industrial scaling relationships that can be applied to the
unique conditions of in-space industry, including the experience curve (or learning curve),
economies of scale, and economies of scope. The Supplementary Material provides a more
detailed rationale of the model choices resulting from the literature [38–67].

Wright’s Law for the experience curve is,

Log2 (𝑏)
𝑆(𝑡)
𝑐WL =( )
𝑆(0)

𝑡
𝑆(𝑡) = ∫ 𝑃(𝑡′)d𝑡′ + 𝑆(0)
0
(13)

where 𝑐WL is the factor that is multiplied onto the baseline production cost, 𝑃(𝑡) is the
production rate, 𝑆(𝑡) is the total production up to time 𝑡, 𝑆(0) is the total production up to 𝑡 = 0,
which is the time of the baseline production cost, and 𝑏 (a positive number between 0 and 1) is
the progress ratio. A common value in industry is 𝑏 = 0.80 with broad variation. Here, 𝑏 = 0.75
seems conservative since this is a new industry with more room to learn in the early decades than
the average terrestrial industry, and great advances are expected in robotic autonomy for
reducing teleoperator labor. The value of 𝑏 will be varied to test parameter sensitivity.

4.5. Economies of Scale

Economies of scale (EOS) are distinct from the learning curve [41,54,58]. Haldi and Whitcombe
[48] derived the relationship on industry data,

𝑋 𝑎−1
𝑐𝐸𝑂𝑆 =( )
𝑋0
(14)

where 𝑐EOS = the factor on the unit production cost, 𝑋 = production capacity, 𝑋0 = baseline
production capacity corresponding to baseline unit cost, and 𝑎 is an empirical parameter often
close to 0.6. There is a good argument (see Supplementary Material) that lunar ice mining will
not reach the optimally maximum scale economy within 30 years even assuming the optimistic
market demand model. Nevertheless, for conservatism we can implement a limit to EOS in the
firm-level costs where a maximum is more likely to occur, as opposed to the cost of parts in the
supply chain where market growth will not likely exhaust EOS over the modeled period. Data
from terrestrial mining, biofuels production, farming, and other industries contributed to the
model choices here (see Supplementary Material). The normalized equipment fabrication and
labor costs with EOS become,

(𝑎−1)/2
′
𝑋props (𝑎−1)/2 Min[𝑋props , 𝑋max ]
𝑥 = 𝑥( ) ( )
𝑋0 𝑋0

(𝑎−1)
Min[𝑋props , 𝑋max ]
𝜔′ = 𝜔 ( )
𝑋0
(15)

where 𝑋props = the production rate (kg/y) for propellants, 𝑋0 = the initial propellant market
correlating to baseline costs, 𝑋max = the optimally maximum production rate beyond which
firm-level EOS ends, and the Min function selects the minimum of the two arguments. The EOS
for equipment fabrication has been split equally into two terms: one for supply chain EOS and
the other for firm-level EOS, the latter one limited by 𝑋max . An example of a firm-level limit is
when the payload faring diameter of the rocket that launches chemical processing reactors to the
Moon is not large enough to accommodate larger reactors, and the firm has not yet innovated
ways to assemble larger vats after lunar delivery. We can assume the operations cost shares the
EOS limit with firm-level capital limits if the number of personnel scales with the mass of
equipment. We will investigate cases with the conservative values 𝑎 = 0.6 and 0.8, and with
firm-level EOS limits 𝑋max = 10 and 20 t/day based on analogy to the biofuels industry. Supply-
chain-level EOS will be modified further for economies of scope.

4.6. Economies of Scope

Economies of scope (SOE) is an effect from having overlap with other industries in the supply
chain or other externalities, or from a firm developing multiple products that have commonality
in capital assets and processes. Lunar propellant mining should benefit from supply chain SOE
as other economic activities develop in cislunar space. A conservative estimate is that lunar
metal production compared to lunar ice extraction will have 𝐹overlap = 50% overlap in the
supply chain technologies (rovers, telecommunications, landing pads, regolith excavating and
handling, beneficiation, regolith heating technology, reusable lunar landers to deliver the product
to the point of sale in cislunar space) so economies in the supply chain for ice mining will scale
with 50% of the metal production rate.

Firm-level operations also share common processes (bringing strip-mined material to the
processing plant and heating it for thermal extraction of ice, so it is already delivered and
preheated for subsequent metal extraction), plus management, administration, legal support, and
technical skills, so firm-level SOE should also exist.

In addition to metal production there are many examples of cislunar industries that may start
after launch costs are adequately low and as robotic autonomy becomes more capable, but for
conservatism only lunar metals will be included here. The results may underestimate the
economic competitiveness and value of lunar propellant in the long run. Investors and policy
planners should perform additional modeling that includes the broader SOE.
The normalized hardware and labor costs with both EOS and SOE become,

(𝑎−1)/2
′
𝑋props + 𝐹overlap 𝑋metals (𝑎−1)/2 Min[𝑋props , 𝑋max ] 𝑓firmSOE 𝑋metals
𝑥 = 𝑥( ) ( ) (1 − )
𝑋0 𝑋0 𝑋props

(𝑎−1)
Min[𝑋p , 𝑋max ] 𝑓firmSOE 𝑋metals
𝜔′ = 𝜔 ( ) (1 − )
𝑋0 𝑋props
(16)

The modeling will assume ad hoc that the lunar metals industry begins 10 years after the
propellant industry with a production rate 𝑋metals = 𝛽𝑋p with 𝛽 = 0 for 0 < 𝑡 < 10 years,
linearly increasing to 𝛽 = 0.3 during 10 < 𝑡 < 15 years, and 𝛽 = 0.3 for 𝑡 > 15 years. It will
assume 𝑓firmSOE = 20% matching the mid-value for the terrestrial examples cited above, and
𝐹overlap = 50%. These parameters will be varied to test sensitivity. The Supplementary Material
provides more discussion of these choices.

4.7. Launch Cost

This spherical cow model does not assume a particular launch system. The following discussion
is only to pick a useful range of values. Current launch cost to LEO on Falcon 9 rockets is about
𝐿0 = $2,000/kg. The claim is often made that the SpaceX Starship will achieve very low launch
cost, perhaps as low as $30/kg to LEO, with reduction caused in part by the new technology (the
larger rocket) and in part by the faster launch cadence as the decreasing launch cost reaches
tipping points that expand the scale of in-space activity. Both are examples of economies of scale
and learning curve. Here we focus on the next 30 years and use the economies of scale and
learning curve equations to fit this expected cost reduction. SpaceX had launched 171 Falcon 9
rockets by the end of 2021 with a payload capacity of ~22.8 t each, with 31 of the launches in
2021. For the simple modeling here, assume their annual up-mass 𝑈LEO increases exponentially

𝑈LEO (𝑡) = 𝑈0 Exp(𝑡/𝜏𝐿 )

(17)

where 𝑈0 = 31 ∙ (22.8 tons) per year and 𝜏𝐿 will be determined. Applying EOS and Wright’s
Law to the initial cost,

𝑡 Log2 𝑏
𝑎−1
∫0 𝑈0 Exp(𝑡′/𝜏𝐿 ) d𝑡 ′ + 𝑆0
𝐿p (𝑡) = 𝐿0 [Exp(𝑡/𝜏𝐿 )] [ ]
𝑆0
(18)

where 𝑆0 = 171 ∙ (22.8 tons). With 𝑏 = 0.8 for the typical learning rate and 𝑎 =0.66 for the
industry-average EOS per Haldi and Whitcombe [48], this will achieve 𝐿30 = $30/kg by 𝑡 = 30
years if 𝜏𝐿 = 4.67. But 𝜏𝐿 = 30/Log(𝑈30 /𝑈0 ) so 𝑈30 = 436,000 t/y up-mass. This is equal to
eight launches per day with the Starship 150 t payload capacity to LEO. This would be
aggressive, but it is comparable to Musk’s plan for settling Mars and is comparable to the rate
that is probably needed to achieve $30/kg, so this was a fortuitous result. It suggests that the
optimistically low launch cost is feasible, if the optimistically high launch rate also occurs. Musk
stated there would 1000 Starships building up in LEO during each 26-month planetary alignment
cycle, then departing to Mars together over a 30 day window, and that this process would
continue over a period of 20 years [68]. To refuel 1000 Starships in LEO would require another
~8000 launches per 26 months for an average of ~11.4 launches per day average. The goal here
is not to defend the scenario but to assess whether it could drive launch costs so low that lunar
propellant cannot compete, since this is the extreme case and because it is affecting peoples’
views on lunar resources. Alternative economic parameters that achieve the same outcome are
𝑏 = 0.7 and 𝑎 = 0.85. Either way, the parameters are reasonable to match Musk’s scenario, and
Eq. 18 may be approximated as an exponential decay of launch cost from 𝐿0 to 𝐿30.

4.8. In-Space Propellant Market Model

The goal here is not to predict the propellant market but to create an adequate model for testing
whether lower launch costs can inherently out-compete lunar-derived propellant as the skeptics
of lunar resources claim. A NASA-sponsored 2020 study by the Science and Technology Policy
Institute wrote that lunar propellant will not be economically viable “if Starship meets its cost
and performance targets” [69]. To disprove this, one must consider the range of future Starship
prices as they continue dropping. A fair comparison will use a market size commensurate with
the launch price.

Studies have predicted that the space commerce market should be price elastic [70,71] but the
elasticity would not appear until launch prices have dropped below a threshold. Andrews and
Andrews [72] wrote, “All available market data indicates that the space transportation market has
both an inelastic and elastic portion” and that the elasticity threshold is $1320/kg (converted
from lb to kg) in 2001 dollars, so $2200/kg in 2022 dollars. Ross et al. [18] estimated the
elasticity threshold was $4400/kg (converted from lb to kg) in 2009 dollars, so $6100/kg in 2022
dollars, and used price elasticities of 1.5 (pessimistic), 2.0 (nominal), and 2.5 (optimistic) in their
modeling. Kutter and Sowers [4,28] likewise argued that the market will be highly elastic when
transportation costs are adequately low, and that in-space propellant storage and transfer will
play a key role in these cost reductions.

Launch prices have decreased slowly since the 1990s, but new vehicles (Long March 3B, Dnepr,
Proton-M, and Falcon 9), priced near or below the expected elasticity threshold, began flying in
the late 1990s and early 2000s. As they gained market share, they accelerated the decline in
global-average launch cost, which appears to have gone below the expected elasticity threshold
around 2000 to 2010 [73–75]. The global annual number of orbital launches declined from the
1960s until about 2005 despite the slowly decreasing launch costs, indicating inelastic behavior
as expected above the threshold, but it suddenly began rapid growth in 2005, which has
continued unabated [76]. Before this change, from 1980 to 2005 the cumulative mass of objects
in orbit increased at an average annual rate of ~140 t/y. This doubled to ~280 t/y starting in 2005,
doubling again to ~550 t/y since 2019 [77]. A rough estimate is that the global-average launch
cost decreased ~40% since 2005 when the annual launch mass doubled. This is consistent with
price elasticity 𝐸 ~ 1.9, close to the nominal value of Ross et al. [18], but if the trend since 2019
is real then 𝐸 ~ 3.8. These figures are crude but are the best available.

The market demand is modeled as,

𝐷(𝑡) = 𝐷1 Exp(𝑡/𝜏)

𝜏 = 30/Log(𝐷30 /𝐷1 )
(19)

Where 𝐷1 and 𝐷30 are the demand in years 1 and 30. 𝐷1 is assumed equal to the initial
production capacity of the lunar propellant business, based on the idea that the firm successfully
markets its initial product. Values of 𝐷1 for the various lunar propellant studies are given in
Table A-1, Appendix A.

For 𝐷30 , this study will consider three scenarios: optimistic, moderate, and pessimistic markets.
For the optimistic scenario we consider setting the demand in LEO to the amount of propellant
that, if launched from Earth, would drive the launch cost to the low limit $30/kg per Section 4.7,
which is 𝑈30 = 436 kt/y. This rate is consistent with Musk’s scenario for Mars settlement. For
cases using solar electric propulsion (SEP) from LLO to LEO (discussed below), the propellant
produced at LS must be 𝐷30 = 𝐺LS−LEO 𝑈30 = 2.48 kt/day. This is high but the modeling shows
it is achievable if this size market actually exists.

For the moderate and pessimistic cases with much smaller market and modestly higher terrestrial
launch cost, we consider setting 𝑈30 to 10% and 1% of the optimistic value, respectively, or the
equivalent of one Starship launch every 1.25 days and one every 12.5 days, respectively. For
comparison, the Falcon 9 launch rate in 2022 was once every 11.8 days. In each case, 𝐷30 =
𝐺LS−LEO 𝑈30. If Starship launched at these slower rates, then per Eq. 18 the launch cost at 30
years would be 𝐿p (30) = $119/kg and $436/kg to LEO, respectively.

These choices are checked for reasonableness by the market elasticities they imply. The
optimistic market scenario is consistent with a price elasticity of

ln(𝑈30 /𝑈0)
𝐸=− = 1.58
ln(𝐿p (30)/𝐿0)
(20)

Although this is the “optimistic” case, it is only a conservative elasticity, between the nominal
and pessimistic values used by Ross et al. [18], and much lower than the elasticity currently
indicated by the launch market since 2005. The optimism is only in the low launch cost, not in
the resulting size of the market. This means that lunar propellant will gain smaller EOS and
learning curve improvements in the model than might reasonably be expected. The moderate and
pessimistic scenarios are consistent with 𝐸 = 1.52 and 1.26, respectively. These are all
conservative tests for lunar propellant.
For our purposes it is not necessary to prove that the lowest launch cost and correspondingly
largest market scenario will occur, only that if they do occur then lunar propellant can still
compete, to answer the claims of skeptics. Nevertheless, the reader may wish to consider
possible use-cases for such a large propellant market. The example previously mentioned is
large-scale settlement of Mars, which is the primary goal of SpaceX. This would consume an
annual propellant supply equal to the optimistic scenario. Musk’s plan is to stage 1000 vehicles
at a time in LEO, but an alternative to reduce crowding in LEO is to stage them in Moon Distant
Retrograde Orbit (DRO) [78–80], Near Rectilinear Halo Orbit (NRHO) [81], Lunar Distance
High Earth Orbit (LDHEO) [82–84] or even EML-1 [85]. Conte et al. [79] showed that DRO has
dynamical advantages over LEO, first because it enables monthly Mars departures via Earth
flyby for at least 7 of the months per 26-month synodic period all with reasonable Δ𝑣 and time of
flight (ToF), and second because supplying the propellant from the Moon reduces the total
propellant needed from LEO to Mars and it reduces the effective gear ratio of Mars transfer.
Thus, DRO has net economic benefit over LEO.

Other activities contributing to an optimistic-scale propellant market may include supporting

national exploration goals and defense activities, space tourism, developing multipurpose in-
space infrastructure, providing in-space services, and developing other space resources.
Propellant demand may include activities that are not yet commercially viable because national
governments and some wealthy individuals are investing in the space economy in expectation
that they will become viable. The prospects for these activities have been discussed elsewhere
[28,69–71,87].

4.9. Capital Delivery Gear Ratio

The value of 𝐺 depends on the transportation architecture for delivering capital to LS. If an
architecture like the fully reusable Lunar Starship architecture is used, commonly expected
parameters predict that 15 flights may be needed to include refueling in LEO and LLO so that
150 t of payload can be landed on LS and all 15 Starship-like vehicles returned to land on Earth.
Since the payload capacity to LEO for one such vehicle is also 150 t, the effective 𝐺 = 15.
Alternatively, if the Starship-like vehicle delivers capital only to EML-1 while an RLL
(LOX/LH2, IMF=0.10) takes the capital from EML-1 to LS, then 𝐺 = 8.5. If the RLL acts as the
tug all the way from LEO, 𝐺 = 6, which is the baseline case. This assumes both capital and
terrestrial propellant are launched using the same rocket to LEO, so 𝐿K (𝑡) = 𝐿p (𝑡) per Eq. 18.
(Some of the TEAs used 𝐿K that is much more costly than 𝐿p , giving terrestrial propellant an
unfair advantage, as discussed below.)

4.10. Lunar Propellant Delivery

In this model, delivery of the lunar propellant to the point of sale is performed by the lunar
mining company using the RLL and OTV that are the firm’s capital assets. Its fabrication and
operations costs were included. Many architectures and technology alternatives are available that
can lower the cost of delivery or increase the price of sale [3,88,89]. These include aerobraking
around the Earth, multi-stage ascent and delivery vehicles, propellant tanks that are left in LLO
to reduce the up/down mass to LS, and the use of stoichiometric fuel/oxidizer ratio, which lowers
𝐼sp but increases ∆𝑣/$ if there are no other customers for the excess oxygen. Neglecting them
here sets a higher bar for proving the competitiveness of lunar propellant, and lunar mining firms
will find greater profit margin than predicted here by including these enhancements.

For the RLL, the model uses LOX/LH2 with 𝐼sp = 450 s and IMF=0.10. We will consider some
cases with high efficiency, low thrust SEP using molecular water propellant for transportation
from LLO toward Earth. Such thrusters are being developed [90–93] and may have 𝐼sp = 1,000
to 4,000 s. Here we use 𝐼sp = 2,000. An alternative is to use an existing (Xenon or Mercury-
based) electric thruster and launch the propellant from Earth to keep refilling it for a small cost,
but this is not considered here. If SEP is used, the long transportation time to the point of sale is
accounted as an increase in the cost of the propellant via a discount rate applied to the
transportation time. LLO to LEO is assumed to take 1.5 years based on the SMART-1 transit
time, and the times to intermediate locations are prorated linearly by ∆𝑣. Note that in this
“spherical cow” model, SEP is used only as an example of high specific impulse alternatives to
explore the general effects of varying the in-space transportation. Many other alternatives exist.

The baseline RLL as defined in Table A-1 has adequate payload capacity to deliver the entire
annual product of propellant from LS to cislunar space assuming each roundtrip takes two weeks
plus one week for maintenance between trips. It is also adequate to deliver the capital to LS
during the buildup period.

4.11. Cost of the Competing Terrestrial Propellant

For the cost of terrestrial propellant competing against lunar propellant, we consider two
architectures. The first is like the (expected) fully reusable Lunar Starship architecture; the
second uses a Starship-type architecture only from Earth-launch to LEO then an OTV/RLL
(𝐼𝑆𝑃 = 450 s, IMF=0.10) to location X in cislunar space. The Starship-type architecture for
capital delivery (X=LS) had 𝐺 = 15, but in that case the payload was limited to 150 t whereas if
the payload is the propellant itself and there are crossfeed lines to deliver propellant from the
vehicle’s tanks, extra refueling in LLO enables delivering 714 t to LS so the gear ratio improves
to 𝐺p,LEO−LS = 10. However, the gear ratio with an OTV from LEO to EML1 (reserving
propellant for the OTV to return to LEO) and an RLL from EML1 to LS (reserving propellant to
return to EML1), is better at 7. Although this is better for propellant efficiency, the modeling
here found that the cost to develop the OTV/RLL (using the same cost model as the lunar capital
but with high reliability since it will not operate on the harsh lunar surface) made it non-
competitive against the baseline Starship in most cases. Therefore, the reusable Starship
architecture is chosen as baseline in the model. The cost of terrestrial propellant scales down
over time per the learning curve and economies of scale contained in 𝐿p (𝑡) in Eq. 18, while gear
ratios remain constant per the physics.

4.12. Propellant Use Ratio

The numerator and denominator of ΓX depend on the transportation architectures for delivering
lunar and terrestrial propellant, respectively, as described above. After specifying the two
architectures, ΓX can be mapped to locations in space. Two examples are given in Fig. 4. The
lower the curve, the more competitive lunar propellant is at that location in space.
 Figure 4. Delivery Cost Factor ΓX vs. Δ𝑣 from LEO for two
combinations of transportation architectures. A lower curve means
greater advantage for lunar propellant. OTV = Orbital Transfer
Vehicle between LEO and EML-1, 𝐼𝑆𝑃 =450 s for non-SEP or
𝐼𝑆𝑃 =2000 s for SEP, IMF=0.10. RLL = Reusable Lunar Lander
with hybrid propulsion, 𝐼𝑆𝑃 =2000 s (SEP) between EML1 and
LLO or 𝐼𝑆𝑃 =450 s between LLO and LS, IMF=0.10. SS = fully
reusable Starship-type vehicle 𝐼𝑆𝑃 =375 s, IMF=0.07. SEP = solar
electric propulsion.

4.13. Finance Cost

For baseline, the discount rate starts at 21.7% following the Weighted Average Cost of Capital
deemed minimally viable by Charania and DePascuale [5] considering risk and other factors. It
decreases linearly to 12% at 30 years to represent the steady reduction of risk. The cost of
developing and fabricating the equipment is assumed to be incurred uniformly through the years
of the buildup period and the annual labor charge is also incurred each year during that period.
That period is set to 5 years for the baseline. These expenses are financed since there is no
revenue yet. Uniform payment/future value is used to calculate the accumulated debt by the end
of the buildup period. The launch cost is incurred at the end of the buildup period, increasing the
debt. Then present value/equal payment is used to calculate the necessary annual net income to
retire the debt by the end of the operational life of the capital. Debt payment plus the annual
operations cost is the necessary gross income, which determines the cost of the propellant. The
total interest over both the buildup and operational periods divided by the total mass of product is
the specific finance cost 𝑓. Dividing that by the launch cost gives the launch-normalized finance
cost 𝜉.
5. Modeling Results

Lunar industry was modeled using the technology parameters in Table A-1 (Appendix A). The
modeling is performed by (1) applying parameters for the specific case, (2) calculating 𝑅opt
given those parameters and for the value of 𝐿𝐾 (𝑡) in each year 1 to 30, and (3) calculating the
total cost of lunar propellant at each location in cislunar space for each year using these
parameters and 𝑅opt . For comparison, the cost of Earth-launched water is calculated for each
location in cislunar space for each year 1 to 30. Year 1 refers to the first year that product is
manufactured and delivered, after the buildup period is complete.

The purpose of the baseline case is to investigate the economic relationships in the sector. No
particular mining technology is specified. This case uses technology parameters that are round
numbers and mid-range among the extant TEAs in Table A-1. The transportation parameters for
capital, terrestrial propellant, and lunar propellant were not taken from those studies but were
specified as described above to provide the best possible analysis of the overall sector, since
some of the TEAs had made poor transportation choices. Uniform transportation parameters will
also enable an apples-to-apples comparison between the TEAs. The baseline case will be
considered in three market scenarios.

5.1. Baseline Case, Optimistic Market Scenario

The baseline/optimistic scenario results are shown in Fig. 5. The cost of propellant drops over
time at every location in cislunar space. Lunar propellant drops faster than launch costs, so the
region of space where lunar propellant is cheaper expands toward LEO. Lunar propellant has an
absolute advantage down to DRO and GEO by year 1 and to GTO by year 5, almost reaching to
LEO by year 30. This overturns the claim that decreasing launch cost makes lunar propellant less
competitive.

$/kg
Years from Start

Lunar
Cheaper

Terrestrial
Cheaper
EML1
DRO
GEO
GTO
LEO

LLO

LS

Delta-v from LEO

 Figure 5. Cost of propellant in cislunar space over time for the
baseline case in an optimistic market, either terrestrial-sourced or
lunar-sourced, whichever is cheaper. Dark dashed line separates
region where terrestrial vs. lunar is cheaper. Delta-v values are
approximate.

Fig. 6 shows the results after changing to SEP to deliver the lunar product below LLO. The slope
of prices changes below LLO due to the higher 𝐼𝑆𝑃 despite the increasing finance cost with
transit time. With this change, lunar propellant has an absolute advantage in GTO by year 3 and
to LEO by Year 15.

$/kg
Years from Start

Lunar
Cheaper
Terrestrial
Cheaper
EML1
GEO

LLO
DRO
GTO
LEO

LS

Delta-v from LEO

Figure 6. Same as figure 5 (optimistic market) except it uses SEP
to deliver lunar propellant between LLO and LEO.

Fig. 7 shows the proportions of cost due to labor, capital, and finance. Over time, capital
becomes an increasing share of the costs relative to labor. Finance causes 82% of costs in year 1
dropping to 73% by year 30. This suggests a large role that government could play to improve
the economics of this sector in its startup years, by creating a public-private partnership (PPP) to
lower investor risk, which lowers the acceptable rate of return and thus dramatically reduces
cost. Eventually proof of successful operation will lower the investor risk, so this PPP is needed
only at the start.
 Year 1 Year 30

Figure 7. Proportion of costs due to labor, capital, and financing

for years 1 and 30.

5.2. Baseline Case, Moderate and Pessimistic Market Scenarios

The baseline case was also evaluated with moderate and pessimistic markets by setting 𝐷30 to
10% and 1%, respectively, of the optimistic value, which corresponds to much lower market
elasticities, so these are indeed pessimistic. The launch cost curve was recalculated per Eq. 17-
18, and the market model was recalculated per Eq. 19. Because the lesser market reduces EOS
and learning curve both for terrestrial propellant launch cost and for lunar propellant, the changes
to the competitiveness are small, as shown in Fig. 8. Table 1 shows the years of operation until
lunar achieves absolute advantage at various locations in cislunar space. Absolute advantage
reaches to GTO in almost the same time in each case. Only LEO is noticeably affected: each
order of magnitude reduction in the market delays absolute advantage in LEO by about 2 years.
This demonstrates that the conclusions of this study do not depend strongly on the size of the
market. Taken together, figs. 6 and 8 indicate that the economic viability of lunar resources is not
strongly dependent on launch costs or market size, disproving the claim by skeptics that low
launch cost will drive space resources out of business.
 10% Market Size $/kg 1% Market Size $/kg

Lunar Lunar
Years from Start

Years from Start

Cheaper Cheaper

Terrestrial Terrestrial
Cheaper Cheaper
DRO

DRO
GEO

GEO
EML1
EML1
LEO

LEO
GTO

GTO
LLO

LLO
LS

LS
Delta-v from LEO Delta-v from LEO

Figure 8. Same as figure 6 except for a moderate market 10% of

baseline (Left) and a pessimistic market 1% of baseline (Right).

Table 1. Years of operation until lunar propellant gains

absolute advantage in various locations in cislunar space.
Calculated for baseline technology parameters in three
market size scenarios.
Optimistic Moderate Pessimistic
(100%) (10%) (1%)
LS 1 1 1
LLO 1 1 1
EML1 1 1 1
GEO 2 2 2
DRO 3 3 3
GTO 6 7 7
LEO 19 21 23

5.3. Sensitivity to Other Economics Parameters

The baseline case using SEP in the optimistic market scenario was re-run with different values of
economic parameters to test their sensitivity. The results are shown in Fig. 9. The Supplementary
Material shows the long-run trends in each of the non-dimensional cost parameters of Eqs. 8 and
9. Ending the firm-level EOS (but not the supply chain EOS) at either production rate 𝑋max = 10
t/day or 𝑋max = 20 t/day caused the cost ratio to become flatter, but it happened late enough that
the year 1 cost ratio was already at the threshold for advantage in GTO. Eliminating the SOE in
the supply chain by setting 𝛽 = 0 or increasing SOE by setting 𝛽 = 1 (i.e., other in-space
industries grow to match propellant mining) had negligible effect. Reducing the EOS by setting
𝑎 = 0.8 caused a modest slowing of the curve. Varying the learning curve parameter between
𝑏 = 0.70 and 0.80 had a modest effect. It is difficult to imagine lunar industry having a value
𝑏 > 0.80.

The implication of this plot is that lunar propellant will continuously gain large additional
advantage over terrestrial propellant during the first 10 to 30 years of operation, and possibly
longer. Terrestrial propellant price is exponentially dropping during this period, but lunar
propellant drops faster. The timescale of these curves is short enough that the change is relevant
for planning both national policy and space agency strategy, i.e., it would be a mistake to set
policy based upon the short-run costs predicted by the TEAs, alone. The tightness of the bundle
of curves in Fig. 9 suggests that this is a firm conclusion.

Figure 9. Pre-delivery cost ratio of lunar vs. terrestrial propellant

for varied economic parameters. When the curves cross below the
thresholds 1/ΓDRO, 1/ΓGTO, or 1/ΓLEO, lunar propellant has
achieved an absolute advantage over terrestrial propellant in that
location.

5.4. Sensitivity to Technology Parameters

The same scenario was re-run varying the technology parameters one at a time to identify
sensitivities as shown in Fig. 10. 𝜙 is varied by changing the mass of product 𝑀p,LS in the
numerator while leaving the mass of capital 𝑀K in the denominator unchanged. The effects of
changing 𝜙 and 𝑀K are nearly inverse but not precisely so because changing 𝑀p,LS also affects
the specific operations cost while 𝑀K does not, and changing 𝑀K affects capital transportation
costs while 𝑀p,LS does not. Changing 𝑀K , 𝑀p,LS and 𝜁 (i.e., both 𝜁D and 𝜁F ) have a nearly
identical direct impact on cost in Year 1. Entrepreneurs may lower the cost of lunar propellant
directly by making improvements in any of the three parameters. Over time, 𝜁 has diminishing
effect while 𝑀K has increasing effect due to EOS reducing the cost of capital while 𝐺 stays
unchanged in physics so transportation becomes relatively more important. The Supplementary
Material shows the behavior of the individual non-dimensional cost parameters of Eqs. 8 and 9.

Figure 10. Pre-delivery cost ratio of lunar vs. terrestrial

propellant for varied technology parameters

Maximizing 𝜙 appears to be the dominant strategy for lowering the cost of lunar propellant – i.e.,
maximizing the mass of product per mass of capital. The values of 𝜙 estimated by the extant
TEAs are shown in Table 2 along with the value used in the baseline model, which was
chosen to be mid-range. Because of the role of 𝜙, the TEAs for tent sublimation are more
optimistic than this baseline while the TEAs for strip mining are less optimistic.

Table 2. Values of 𝜙 per the extant studies

Mining Tent Borehole
Strip-Mining Beneficiation
Method Sublimation Sublimation Baseline
Technology Technology
Technology Technology Model
Study K S P CD J B P M
𝜙 442 534 16.1 26.5 22.2 43.4 3.7 36.5 167
Abbreviations for studies: K = Kornuta et al.[1], S = Sowers [4], P = Pelech et al. [7], CD =
Charania and DePascuale [5], J = Jones et al. [6], B = Bennett et al. [2]; M = Metzger et al. [3].

The parameter elasticities on the cost ratio 𝜓0 were tested by running the model with 1%
increments of a parameter and are shown in Table 3. The elasticity for 𝑀p,LS can be interpreted
as the (negative of) elasticity for 𝜙. It is nearly equal to (negative) unity since it divides
directly into cost and has no offsetting effect on the competing terrestrial propellant costs.
𝑀K also affects 𝜙 but it is entangled with hardware fabrication cost and transportation cost
so elasticity for 𝑀K is not identical to elasticity for 𝜙. Elasticity with 𝐼𝑆𝑃 (for capital
delivery) exceeds (negative) unity when transportation costs dominate because of the
exponential dependence in the rocket equation. The cost ratio is nearly inelastic on launch
cost 𝐿0 when 𝐺 ≫ 𝑥, since it affects both terrestrial and lunar propellant costs equally. It is
nearly (negative) unity when capital acquisition costs dominate. Lunar propellant can
therefore be insulated from decreasing launch costs by achieving 𝑥 < 𝐺 as a capital design
goal.

Table 3. Parameter Elasticities on Cost as a Function of the 𝐺/𝑥 Ratio

𝑀p,LS 𝑀K 𝜁 𝐺 IMF 𝐼𝑆𝑃 𝐿0
𝐺/𝑥 = 0.02 -0.990 0.983 0.973 0.011 0.005 -0.023 -0.947
𝐺/𝑥 = 1 -0.990 0.710 0.432 0.277 0.120 -0.614 -0.692
𝐺/𝑥 = 50 -0.990 0.982 0.024 0.958 0.437 -2.116 -.0.040

6. Assessing the TEAs

Four studies were optimistic about the competitiveness of lunar propellant beyond LS or LLO:
Kornuta et al. [1], Bennett et al. [2], Metzger et al. [3], and Sowers [4]. Three studies were
pessimistic: Charania and DePascuale [5], Jones et al. [6], and Pelech et al. [7]. Here we re-
assess these studies considering especially how they handled 𝐺 and 𝜙.

6.1. Charania and DePascuale

Fig. 11 considers five variations of the Charania and DePascuale study [5], hereafter CD. Curve
(1) replicates CD normalized by launch cost, 𝐿p (𝑡). CD only predicted short run costs, so the
curve for subsequent years was calculated using the same economic scaling parameters as the
baseline model. CD assumed high 𝐺 = 64.9 (per current pricing) due to the use of government-
built (non-commercial) rockets for capital transport. Even with EOS/SOE and learning curve,
since 𝐺 is high the reduction in 𝑥 has diminishing returns and the slope in the curve is small.
Because of the technology and transportation assumptions in CD, the curve is two orders of
magnitude too high to be economic.

CD was written before the LCROSS mission that proved lunar ice concentrations at 5% and
higher exist [94], so CD assumed 1% ice concentration. Current thinking in the lunar mining
community is that nobody would pursue a deposit with such low concentration. Curve (2) makes
only one change: it assumes a prospecting campaign has identified a better ore body that obtains
5% yield so 𝜙 is improved by a factor of 5.

Curve (3) makes three more changes. First, it uses commercial launch of the capital assets, so 𝐺
is reduced from 64.9 to 8. This reduces the cost of propellant at LS by 59% in year 1 which
grows to extreme cost savings by year 20 and beyond since 𝐺 is now lower than 𝑥 and the
returns from EOS and learning curve are not diminished. Second, it implemented optimization of
reliability, which provides another 10% reduction in cost at LS. Third, it implements SEP for
delivery of the product from LLO toward LEO. The longer delivery time increased finance costs
but also raised the thresholds 1/ΓX where lunar propellant gains the absolute advantage. The
three thresholds shown on the plot are therefore only applicable to Curves (3) through (5) that
use SEP. This curve gains absolute advantage in DRO by year 15 and in GTO by year 18,
expanding the business case beyond customers near the Moon.
Curve (4) recognizes that NASA is currently working toward a lunar pilot plant for ice mining
and propellant production, so the buildup time of 8 years in CD may be reduced to 4 years and
50% of the development cost 𝜁D (but not the fabrication cost 𝜁F ) of equipment may be covered
by that government-funded effort. Curve (5) now gains absolute advantage for lunar propellant in
GTO by year 13. This is still a long time but the high discount rate of 27.2% is commensurate.

Curve (5) assumes a PPP is established and the discount rate is lowered to 12%. Now, it gains
absolute advantage in GTO by year 8. The CD study becomes optimistic for lunar resources with
these changes.

Figure 11. Pre-delivery cost ratio for CD with four improvements.

The dotted lines are thresholds for cases with SEP (cases 3 to 5).
When crossed, lunar propellant has gained absolute advantage over
terrestrial at the corresponding location (DRO, GTO, or LEO).

6.2. Jones et al. and Bennet et al.

The study by Jones, et al. [6], hereafter J, evaluated a strip mining approach and predicted a low
value of 𝜙 = 22.2. Bennett et al. [2], hereafter B, assessed the modeling by J and showed that
different design choices can increase the economies of scale for the reactors, resulting in 𝜙 =
43.4. J used the Space Launch System (SLS) for capital transport to LS. At the time, SLS was a
good alternative with 𝐺 = 5.4. Since commercial Heavy Lift Launch Vehicle (HLLV) prices
have already dropped but SLS costs have not dropped, SLS now yields 𝐺 = 41.8. This would
dominate the capital costs and ensure future reductions in its fabrication cost would have
negligible return. J neglected finance and operations costs, probably because it was assessing a
government-led effort without full-cost accounting. If they had been included, J would have an
even more pessimistic prediction. Fig. 12 shows the resulting cost curves for J and B using the
same capital transport as the baseline model instead of SLS, with reliability optimized, and
finance and operations cost added using the parameters in Table A-1. The results are not as
optimistic as predicted by B due to the added finance and operations costs. It does not gain
absolute advantage in DRO until year 16 or in GTO until year 20. However, it indicates lunar
resources are competitive for refueling at LS from the start. Sowers [4] explained that the method
J used to approximate 𝜙 (which B followed) was methodologically flawed. It was based on an
inappropriate analogy to Molten Regolith Electrolysis (MRE), which is unlike ice extraction.
Even with the economies of scale corrected by B, this analogy cannot prove the strip mining
technology to be either economic or uneconomic. In contrast, CD was based on a detailed design
study by CSP, Japan, Inc., and Shimizu Corporation. The details of that study are not published
but are described at a top-level by Charania and Kanamori [95]. A detailed, published TEA for
the strip mining technology is still needed.

Figure 12. Pre-delivery cost ratio for J and B using SEP. The
dotted lines are the thresholds for absolute advantage at each
location, 1/ΓX.

6.3. Pelech et al.

Pelech et al. [7], henceforth P, assessed two technologies for lunar ice mining. The first was strip
mining the icy regolith then thermally separating ice from the excavated soil by sublimation,
similar to CD, J, and B. The second was to drill boreholes then sublimate the ice in situ by
injecting energy (e.g., microwave) down the boreholes using battery-stored energy onboard a
rover. P did not assess cost but only the “propellant payback ratio” PPR = G−1 𝜙 ΓX−1, the inverse
of the (launch-normalized, product-specific) capital delivery cost. They argued the PPR is too
low in these two technologies so lunar propellant will not be economic in the near term.

PPR depends on the transportation architecture for G and ΓX and on the design of the capital for
𝜙 = 𝑀p,LS /𝑀K . P assumed capital would be delivered by the Falcon Heavy rocket for G = 7.5,
which is competitive. For lunar propellant delivery P used an RLL with 𝐼𝑀𝐹 = 0.83, which is
unrealistically high. This created a 50% inefficiency in delivering the product to LEO,
contributing somewhat to the low PPR. More importantly, P estimated that 𝜙 will be far too low
in both technologies.

For the strip mining technology P roughly estimated 𝜙~3.7. CD, J, and B predicted 𝜙 an order
of magnitude larger for similar technology. The question becomes which of these studies has the
more reliable estimate. None of the studies published enough detail to answer this. Pelech
(personal communication, T.M. Pelech, Aug. 16, 2022) stated that the technology details in P
were relatively less important because the study’s intent was to demonstrate the usefulness of the
PPR, so significant changes in the estimated 𝑀K are possible. In P, the estimate of the bucket
wheel excavator mass was based upon an analogous terrestrial excavator after scaling the force
equations, but space technology uses materials and designs to achieve lower mass without
compromising the forces, so a terrestrial analogy should produce too high an estimate of 𝑀K .
NASA showed that using carbon fiber and aluminum instead of steel reduces the mass of a large
excavator blade by 70% without sacrificing strength [96], and that reaction forces for digging
buckets can be reduced via percussion in low gravity to avoid using more mass [97,98]. These
technologies are not used in terrestrial applications because transportation of capital is cheap, so
they seek to minimize the cost of capital rather than its mass. Although P did not assess cost,
these mass-savings are typically reflected in the higher fabrication cost of spaceflight hardware.
Since we are already paying for them, it is appropriate to include them in the mass estimate. It is
impossible to explain the full difference in 𝜙 between these studies without more detail on how
the costs were estimated, so the economic viability of strip mining lunar ice is still undecided
although the CD study with changes discussed above favors its viability.

For the borehole sublimation technology, the values in P indicate a much better 𝜙~16.1. If 𝑀K
was overestimated by a factor of 3 then this puts this technology in the economic range. It
appears P determined the mass of the drilling rover by the mass needed for weight on bit in lunar
gravity, rather than the mass needed to fabricate the chassis and subsystems in a bottom-up
assessment. Any shortage in the weight on bit can be made up by filling regolith in a bin on the
rover or by anchoring [99], and drilling methods have been developed to reduce the necessary
weight on bit in low gravity environments [100]. Delivering a more massive rover to provide
weight on bit is therefore not needed and contributes to a lower 𝜙. The estimate of the mass of
the downhole energy rover starts with a rough estimate that batteries constitute 50% of the mass,
followed by a calculation of the mass of batteries. Regenerable fuel cells would require only half
the mass of batteries, and the 50% figure may not be accurate for spaceflight construction.
Without more detail, we cannot say whether the borehole sublimation is economic, but it appears
to be in the range of becoming economic. The focus is clearly on the technology design to
improve 𝜙.

6.4. Kornuta et al. and Sowers

Sowers innovated the tent sublimation technology with the goal of vastly increasing 𝜙. It became
the subject of the United Launch Alliance’s study authored by Kornuta et al. [1], henceforth K.
Another analysis was authored by Sowers [4], henceforth S. K predicted 𝜙 = 441.8 and S
predicted 𝜙 = 534. This will clearly be economic and will outcompete other lunar mining
methods if the technology works as expected. Bennett et al. [13] analyzed a business model
based on K including a propellant delivery architecture showing it to be profitable. Recent
experimental work by Sowers [101] and Purrington et al. [102] indicates that the technology will
work. S based the demand estimate on a portfolio of applications in cislunar space, between 40%
and 100% for use in LEO per the three scenarios.

S estimates an Internal Rate of Return (IRR) of 8.84% in a fully commercial venture, which is
much less than the 21.7% that CD thought necessary to attract investors. S describes how a
public-private partnership can offset development costs to raise the IRR while also making a
lower rate of return acceptable to investors since the government bears a portion of the risk. S did
not model how the costs may evolve in subsequent years and for conservatism ignored
economies of scale (George Sowers, personal communication, May 15, 2022), but the modeling
of cost evolution here shows that even with the 21.7% rate it may achieve absolute advantage to
LEO by Year 5. Based on the results of K and S, there is a clear business case so navigating the
startup financing is the remaining challenge.

6.5. Metzger et al.

The study by Metzger et al. [3], henceforth M, had the goal of solving the startup finance
problem. Two prominent space mining companies (asteroid mining companies) had raised seed
and Series A investment rounds but were unable to secure their Series B rounds [103]. M
approached this by developing a “Minimum Viable Product” technology that should produce net
revenue immediately for the minimum 𝑀K and hence the minimum initial investment, although it
sacrifices 𝜙. The mining company would quickly operate in the lunar environment to mature the
technology and buy-down risk while making revenue to offset costs. The most important product
would not be propellant but investor optimism. Successful Series A and subsequent rounds
would enable scale-up of the operation, including transition to larger-scale mining methods for
greater 𝜙. The technology that was innovated for this [104] was based on a geological argument
that the ice should be in granular form that is highly excavatable, so a low-energy beneficiation
process can reduce the excavated mass by 95% prior to hauling ore into a sunlit (energy-rich)
location for processing. This eliminates the need to beam energy into the permanently shadowed
regions where the ore bodies are found, dramatically reducing the mass of energy infrastructure.
M estimated the resulting 𝑀K to be an order of magnitude lower than other studies, so startup
costs are lowered by an order of magnitude, yet 𝜙 = 36.5 is still high enough to gain absolute
advantage at least to GTO as shown below. It would produce enough propellant to boost four
satellites from GTO to GEO annually. Ground truth of the lunar ice is needed to validate the
geological arguments in M. NASA is currently developing a mission to obtain ground truth on
the ice [104].

6.6. Other Technological Approaches

There are other technologies in development that I did not discuss only because I do not have
access to TEAs on them. Ethridge and Kaukler [105–107] and Ethridge [108] developed down-
borehole heating using microwave or radio frequency energy. Honeybee Robotics developed the
Mobile In-Situ Water Extractor (MISWE) and Planetary Volatile Extractor (PVEx) technologies
[109,110], both of which drill a borehole, the former bringing the cuttings to the surface via an
auger for thermal volatile extraction, the latter heating the icy regolith within the borehole for
thermal extraction. Sercel [111] and Sercel et al. [112] innovated a multi-wavelength, ground
penetrating beam method powered by tall tensegrity towers to capture the solar energy above the
Moon’s permanently shadowed craters. Kuhns et al. [113] innovated “rocket mining” whereby a
rocket thruster under a dome disrupts and heats the soil rapidly to liberate and capture volatiles.
Austin et al. [14] analyzed architectures for robotic lunar ice mining based on three alternative
approaches: mechanical strip mining, pneumatic strip mining with pneumatic beneficiation, and
the Honeybee Robotics MSWE technology. There are several companies working on lunar ice
mining so there may be other technologies or architectures of which I am unaware.

6.7. Summary of Prior Studies

The parameter values in Table A-1 were provided by the different extant studies. Some studies
did not include or report various parameters, so each study was back-filled as explained in the
notes of Table A-1. Optimization of reliability was added onto each case, and the same long-run
economic scaling relationships were applied to each. To compare technologies on an equal basis,
all studies were modified to use the same transportation parameters discussed above. K and S
both considered transportation to be the business of a different firm than the lunar mining
company, so they did not include development of transportation capital but here they have been
included. Fig. 13 shows the long-run average cost for the studies. CD5 is the modified version of
CD from Curve 5 of Fig. 11.

The curves fall into groups based on technology. This justifies the statement in the introduction
that an individual TEA cannot disprove the viability of ice mining as a sector but only of a
particular technology (and then only in a specific form since future innovations may improve the
technology). The clustering also shows the key role of 𝜙 in determining economic viability. All
models except J and B indicate that an absolute advantage is gained in GTO no later than year
12. Both tent sublimation studies predict economic viability in LEO from year 5.

Figure 13. Pre-delivery cost ratios for six studies with SEP using
27.2% discount rate. CD5 = modified CD per curve 5 on Fig. 11.
See text for study identifiers.
 1/ΓDRO, 1/ΓGTO , or 1/ΓLEO, lunar propellant has achieved an
absolute advantage over terrestrial propellant in that orbit.

7. Sensitivity to Discount Rate and Market Size

Figure 15 shows the same six studies as Figure 13 but it assumes they are implemented in a
Public-Private Partnership so the discount rate is 12% (constant over 30 years). Now, all studies
except J and B predict absolute advantage in GTO by no later than year 10.

Figure 15. Pre-delivery cost ratios for six studies using 12%
discount rate. CD5 = modified CD per curve 5 on Fig. 11. See text
for study identifiers.

The two main technological approaches, strip mining and tent sublimation, were tested against
the optimistic, moderate, and pessimistic market cases to see if either technology is more
sensitive to the scale of the market. As shown in Fig. 14, the scale of the market has negligible
effect until about 10 years of operation and only a minor effect after that. Tent sublimation still
has absolute advantage all the way to LEO from year 1. Strip mining is delayed for gaining
absolute advantage in LEO in the more pessimistic market but still has a business case down to
GTO without additional delay.
 Figure 14. Pre-delivery cost ratios for CD6 strip mining (dashed
curves) and S tent sublimation (solid curves) each at 12% discount
rate. Black: baseline market model. Gray: market model reduced to
1% of baseline.

8. Sensitivity to Reliability

To test sensitivity to reliability, additional cases were modeled for both strip mining and tent
sublimation while varying 𝑅0. A brief explanation is needed to explain the difference between
varying 𝑅0 and varying 𝑅. It is a difference between innovating new technology concepts during
Technology Readiness Level (TRL) 1 though 6 versus execution of established technology
concepts during TRL 7 through 9 [114,115]. For the latter, if we build flight hardware using
normative spaceflight parts, materials, design features, fabrication processes, and preflight
testing, then when operating on the Moon the actual reliability of the hardware 𝑅 (or the “as-built
reliability”) will equal the baseline 𝑅0 by definition. Building the flight hardware with better or
worse parts and processes and with more or less testing than the norms for spaceflight will yield
higher or lower 𝑅, respectively, but will not change 𝑅0. These decisions are what drive the
exponential cost of reliability as described by Mettas [33].

The baseline reliability 𝑅0 on the other hand can be improved through innovation and maturation
of new technological concepts along the TRL scale from 1 to 6. To improve 𝑅0 a new
technology concept must address one or more of the following described by MIL-HDBK-338B
[36]: (1) reduce exposure to harsh environmental and operating conditions; (2) reduce operating
time; (3) simplify the technology so it is less “intricate” (e.g., fewer parts); and (4) mature the
technology to identify and address failure modes. For a rover that must operate in harsh dust we
might improve 𝑅 by using more powerful motors and higher-torque gear assemblies that can
continue to turn despite the friction of dust accumulated in rotating wheel joints, or we might
improve 𝑅0 by inventing mobility system that uses reciprocating rather than rotating joints so
they can be booted, preventing dust from entering the joints in the first place.
All other factors being equal, improving 𝑅 increases the cost of the hardware but improving 𝑅0
decreases the cost. After the new technology concept with higher 𝑅0 is matured to TRL-6 and
adopted for in-space use, the team that builds the flight hardware from TRL-6 to 9 can achieve
the targeted value of 𝑅 without using the more expensive and massive motors or other parts, or
more extensive testing, etc. Therefore, varying 𝑅0 in the model represents technological
innovation and maturation of new technologies before incorporation into flight-like builds. Since
so little work has been done in lunar mining, there is room for much innovation to lower 𝑅0, but
since we have no data from actual lunar mining there is also uncertainty in what value of 𝑅0 to
use for the current state of the art, hence the need to vary it here.

The model assumes a value for the baseline 𝑅0 then optimizes 𝑅 for minimum overall cost using
the exponential cost of reliability relationship, Eq. 11. The results are shown in Fig. 16 where
there is a set of curves for CD5 (representing strip mining) and a set of curves for S (representing
tent sublimation). If 𝑅0 = 0.55, 0.70, or 0.85 can be achieved, then strip mining would gain
absolute advantage in GTO by year 9, 7, or 5 respectively and gain it in LEO by year 24, 21, or
17 respectively. For tent sublimation, all cases have absolute advantage in GTO by year 1. Even
if 𝑅0 for tent sublimation were as low as 0.10, it would have absolute advantage in LEO by year
5. 𝑅0 was incremented in uniform Δ𝑅0 = 0.15 steps within each set of curves but the resulting
cost differences between the curves are not uniform increments. This is because the cost of
reliability while optimizing 𝑅 is exponential, not linear. As 𝑅0 achieves increasingly higher
values through technology innovation, the reductions in cost become increasingly greater. This
emphasizes the important role of innovation.

Figure 16. Pre-delivery cost ratios for CD5 strip mining (gray
curves) and S tent sublimation (black curves) assuming baseline
reliabilities 𝑅0 = 0.10 (top curve in each set), 0.25, 0.40, 0.55,
0.70, and 0.85 (bottom curve in each set).

9. Discussion
9.1. Concerns of Inadequate Reliability

Several members of the aerospace community expressed concern that lunar propellant may not
become competitive because of technological and business pragmatics. Several mentioned their
concerns with reliability, either that it will be too hard to achieve adequate reliability in the harsh
lunar environment, or that firms will not be able to lower costs by lowering reliability to some
intermediate level between “too reliable and too costly” and “not reliable at all.” Ultimately, the
only way to remove these concerns is to build and test hardware, advancing it to TRL 6 where
the tests include the relevant environment, and then to operate the system on the Moon. That
work is expensive so decision-makers will want to confidence to proceed. This modeling
suggests we have adequate confidence for that next step. We may note that NASA is already
driving technology maturation for lunar ice mining through contract and grant solicitations and
through the Break the Ice Lunar Challenge, part of the NASA Centennial Challenges program
[116].

The most serious challenge to reliability may be the abrasive lunar dust. Some of the
technologies, like tent sublimation, do not strip mine so they have much less environmental
stress and it may be easier to achieve the reliability targets. System-average reliability can be
improved by making assets more modularized for telerobotic replacement of smaller units. It
should be remembered that most of the mass of assets will have higher reliability than the norm,
because it is usually a small fraction of the mass that fails often. This improves the average
reliability. Open-ended innovation can improve 𝑅0 to reduce costs non-linearly as demonstrated
above. Even if 𝑅0 = 0.10, which is unreasonably low, at least one technology is still economic
per Fig. 16.

9.2. Concerns of Inadequate Robotic Autonomy

Some in the community point to the degree of robotic autonomy as another cause for doubt. The
economic analysis here assumes robotics will operate without humans present. Including humans
greatly increases cost but also greatly increases capability to troubleshoot and repair hardware
and to study improvements in the technology. There may be broader business cases than
discussed here that include humans for greater revenue-generating scope than just propellant
mining, which would offset the costs. For one example, there are non-commercial reasons for
national space agencies to have humans on the Moon doing lunar science and technology
maturation in support of national objectives. As transportation costs to the lunar surface drop, the
cost of humans on the Moon will also drop.

On the other hand, many in the community are convinced we can build a fully robotic operation
with no humans on-site or only minimal human visits. One source of confidence is the
demonstration of progress by the NASA Lunabotics competition [117–120] and the NASA
Break the Ice Challenge [116]. In the first few years of Lunabotics, university student clubs
produced robots that were barely functional in dusty lunar-like regolith. After 10 years of their
learning curve, with knowledge passed down informally within and between the clubs, they were
building highly capable lunar mining robots that were fully autonomous and robust in the harsh
regolith. The competition judges studied the robots and kept records of their improvements over
the years but never shared this data with the students so as not to disturb the competition. In the
first year, almost every robot was immediately stuck in the regolith. After a few years, some
teams had discovered how to make robots nearly immune to getting stuck. Some robots used
vision systems with machine learning to identify and navigate around terrain hazards, automatic
feedback to adjust motor torques and driving speeds, localization and area navigation to find and
then drive to and from the mining zones, feedback controls in the mining systems to adjust for
changes in digging conditions, and docking sensors to align the robot with the simulated
chemical processing plant to offload the ore. Some teams implemented multiple layers of
navigation for redundancy. Robots could drive back and forth in repeated mining cycles with no
human intervention. The teams engaged in continuous innovation in all the robot’s subsystems
year-by-year. The clubs accomplished this using off-the-shelf components such as mobile phones
and gaming consoles, and with budgets on the order of $5,000 to $10,000 USD and a labor force
that worked only part-time in the evenings without pay. It seems unlikely that full-time
professional technologists with corporate support should be unable to make similar progress
using space-qualified components and a much larger budget.

At an early stage, autonomy can manage discrete tasks such as driving around obstacles and
moving the digging bucket to take an optimized bite of soil – tasks that enable teleoperation
despite the 3-second round-trip communications latency between the Earth and Moon. Part of the
expectation for a long-run learning curve in this study is that autonomy will continue to improve
over the next thirty years so greater autonomy can be phased-in smoothly across the various
tasks and integrated operations, asymptotically approaching full autonomy. Autonomy is already
advancing rapidly in terrestrial mining, construction, and manufacturing so this expectation is
well supported. When human intervention is needed, the shortness of the communications
latency allows much faster exchanges than with Mars rovers, making loosely supervised
autonomy feasible with fewer teleoperators per robot standing by for support year-by-year.

9.3. Concerns of Inadequate Cost Reductions

There is also a question whether a learning curve or economies of scale can really reduce costs in
lunar industry since it is unlike terrestrial industry in some ways. Can the processes really be
simplified and improved to reduce costs in every increment? I argue there is no reason to doubt
it, because lunar industry will be identical to terrestrial industry in all the ways that matter.
Capital will still be made on Earth (at least at first), benefitting from improvements in the
terrestrial factories and supply chain just like any other industry. Next generation hardware will
be redesigned for simplicity and efficiency, as usual. Operations on the Moon can be simplified
as we learn how to position equipment around an ore body and schedule tasks efficiently, just
like any terrestrial operation. The payload farings on future rockets like the SpaceX Starship will
be so large that it seems impossible we should need larger diameter capital assets for the
projected throughput of product. Even if we do, that has already been included conservatively in
the model through 𝑋max in Eq. 16. The throughput rate of water cleanup systems and electrolysis
at 𝑋max = 10 t/day allows 2.4 hours for every cubic meter of water that is processed, and that
needs only a tiny pipe, orders of magnitude smaller than a payload faring, so it seems unlikely
that EOS limitations really exist on these scales. The only ways that lunar industry will differ
from terrestrial industry are (1) the transportation costs, which have been included in the model,
(2) the harshness of the environment, which has been modeled in the reliability equations, and
(3) the lack of human access to repair the capital in its operational location. Further technology
maturation is needed to address these last two challenges, but they provide more opportunity for
learning curve, not less.

9.4. Possible Extensions of the Analysis

A similar analysis could be done for transporting capital to Mars to support operations on that
planet. In that case, there are more options to decrease 𝐺 because propellant manufacturing can
take place on Mars, Phobos, and the Moon to create multiple refueling stations in the vicinity of
both Earth and Mars to break up the exponentiality of the rocket equation. A large fraction of the
technologies and infrastructure for lunar propellant mining can be adapted to Phobos and Mars,
so developing lunar propellant mining and manufacturing at the present time on the Moon will
have long-term benefit for Mars and beyond.

Another way to reduce the cost of transporting capital is to get as much of its mass as possible in
situ on the Moon or Mars rather than transporting it from Earth. Here, we considered metal
manufacturing only for its effects lowering supply chain costs via economies of scope. Using the
metals to fabricate structural elements and other parts of the lunar propellant firm’s capital may
bring an additional reduction of costs that bypasses the limitations on 𝐺. Future analyses should
modify the equations to include that dynamic.

10. Summary

This analysis has shown that the profitability of lunar propellant is possible in physics and
plausible in economics as a function of hardware performance parameters that are subject to
open-ended improvement. It depends mostly on reasonable 𝐺 and high 𝜙. The former has a limit
in physics, but the latter does not. The threshold of absolute advantage in this modeling is 𝜙 ≳
35. The tent sublimation method was innovated to obtain higher 𝜙 and its values are predicted to
be an order of magnitude better than the threshold of absolute advantage. The strip mining
method is closer to the threshold. No adequately detailed, publicly available TEA exists to parse
the narrower margin for strip mining. However, when the EOS and learning curve are included,
strip mining gains absolute advantage over terrestrial propellant in GTO within just a few years
of operation. Furthermore, any technology that is close to the threshold like strip mining can
become profitable simply with more innovation. Lowering 𝑀K and increasing 𝑀p,LS are
analogous to the performance improvements that engineers and inventors make in all
technologies routinely. The results of this analysis should reduce doubt that lunar propellant
mining can out-compete terrestrial-launched propellant with absolute advantage both in the near
term and in the long term.

This model assumed launch costs will descend to the lowest values considered possible, and still
lunar propellant gains absolute advantage. The size of the propellant market corresponding to the
launch cost was modeled both pessimistically and optimistically, and in all cases the cost of lunar
propellant dropped faster than launch cost and demonstrated there is a viable business case. The
model made many conservative assumptions, such as neglecting aerobraking, stoichiometric fuel
ratios, lunar sling-launch, and other transportation alternatives that will give lunar resources an
extra advantage, and by assuming that lunar industry’s firm-level economies of scale may “shut
off” at a rather low throughput.

Technological maturation is needed to prove adequate reliability can be achieved for long
operation in lunar conditions. This appears the be the most significant source of doubt within the
aerospace community, so a sufficient demonstration with prototypes in a relevant environment
may be needed. National space agencies can play a strong role in the startup of this industry by
funding this effort.

Mistakes were made in some of the earlier studies. First, point-designs of a single technology
were used to claim that lunar propellant cannot out-compete terrestrial propellant. Since
everything depends on the technology parameter 𝜙, which can vary by orders of magnitude
between different technologies and different designs, it is illogical to use one point-design of one
technology to infer the fate of the entire economic sector. Second, the evolution of cost over time
was found to be significant within timescales relevant to planning space activities, but prior
studies focused on the performance of the first plant in its first year of operation, only. Jones et
al. mentioned that their look at learning curves did not affect their pessimistic prediction, but
they did not document their calculation and the opposite is clearly predicted here. Third, some
prior studies failed to identify 𝜙 as the crucial parameter, so they did not work creatively to
innovate a system that maximizes 𝜙 but instead put together an ad hoc system and whatever it
predicted was the end of the study. Bennett et al. [2] gave an example of taking one of those ad
hoc systems and reengineering it to improve 𝜙 to make it profitable.

11. Conclusions

Lunar-derived rocket propellant can outcompete rocket propellant launched from Earth, no
matter how low launch costs go. The modeling here, which was based on normative industrial
data and conservative assumptions, shows lunar propellant can be cheaper than terrestrial not
only close to the Moon, but also in GTO and in LEO. This can enable a commercial business
case for lunar industry to boost spacecraft, which can support development of space industry
with the scaling of Earth’s commercial markets. The key is to develop a lunar propellant
technology that has a production mass ratio 𝜙 ≳ 35 and to avoid expensive transportation
architectures for delivering the capital to the Moon. The tent sublimation method studied by
Kornuta and by Sowers is estimated to have 𝜙 = 400 to 550, an order of magnitude better than
needed for absolute advantage. Strip mining methods for obtaining lunar ice are probably above
the cost threshold for absolute advantage in year 1 of operation, but technological innovation and
maturation, government-funded development in the early stages, and/or public-private
partnership will lower costs. Furthermore, only 4 to 11 years of operation should be enough for
the learning curve and increased economies of scale to lower the initial cost estimates of strip
mining below the threshold of absolute advantage. It is plausible that additional technologies
besides strip mining and tent sublimation may be profitable. These conclusions hold over a
reasonable range of economic parameters. The remaining key challenge appears to be proof of
operability for years in the lunar environment and especially the dust with a sufficiently low
failure rate. The technological readiness level needs to be advanced to TRL 6 including a focus
on long-run reliability and a demonstration of integrated robotic autonomy with simulated Earth-
Moon communications latency to address remaining doubts.
Funding Information

This work was supported in part by NASA Solar System Exploration Virtual Institute
cooperative agreement award NNA14AB05A, “Center for Lunar and Asteroid Surface Science.”

Acknowledgements

I am grateful to Aeneas Weckenmann (Frankfurt, Germany), who developed the analysis method
for the reusable Lunar Starship including the parameter estimates. I am grateful to Ian Lange,
Division of Economics and Business, Colorado School of Mines, to Akhil Rao, Department of
Economics, Middlebury College, to Pierre Lionnet, Research and Managing Director, ASC
Eurospace, and to Jeff Greason, CTO and Co-Founder of Electric Sky, for reading an earlier
version of the manuscript, providing helpful critiques, and identifying several issues that I had
failed to discuss.

References

[1] D. Kornuta, A. Abbud-Madrid, J. Atkinson, J. Barr, G. Barnhard, D. Bienhoff, B. Blair, V.

Clark, J. Cyrus, B. DeWitt, Commercial lunar propellant architecture: A collaborative
study of lunar propellant production, Reach. 13 (2019) 100026.
[2] N.J. Bennett, D. Ellender, A.G. Dempster, Commercial viability of lunar in-situ resource
utilization (ISRU), Planet Space Sci. 182 (2020) 104842.
[3] P.T. Metzger, D. Sapkota, J. Fox, N. Bennett, Aqua Factorem: Ultra Low Energy Lunar
Water Extraction. Final report, NASA Innovative Advanced Concepts (NIAC) Phase I,
NASA grant no. 80NSSC20K1022, 2021.
[4] G.F. Sowers, The Business Case for Lunar Ice Mining, New Space. 9 (2021) 77–94.
[5] A.C. Charania, D. DePasquale, Economic analysis of a lunar in-situ resource utilization
(ISRU) propellant services market, in: 58th International Astronautical Conference—IAC
Paper, 2007: pp. 07-A5.
[6] C.A. Jones, J. Klovstad, E. Judd, D. Komar, Cost breakeven analysis of cis-lunar ISRU for
propellant, in: AIAA Scitech 2019 Forum, 2019: p. 1372.
[7] T.M. Pelech, G. Roesler, S. Saydam, Technical evaluation of Off-Earth ice mining
scenarios through an opportunity cost approach, Acta Astronaut. 162 (2019) 388–404.
[8] R. Shisko, A comparison of previously published papers on the economics of lunar In Situ
Resource Utliization (ISRU), 2022.
[9] B. Kutter, G. Oneil, B. Pitchford, F. Zegler, A practical, affordable cryogenic propellant
depot based on ULA’s flight experience, in: AIAA Space 2008 Conference & Exposition,
2008: p. 7644.
[10] F. Zegler, B. Kutter, Evolving to a depot-based space transportation architecture, in:
AIAA Space 2010 Conference & Exposition, 2010: p. 8638.
[11] W.J. Larson, T.R. Gill, R.P. Mueller, J.S. Brink, Establishment of a Spaceport Network
Architecture, in: International Astronautics Congress 2012, 2012.
[12] K. Ho, K. Gerhard, A.K. Nicholas, A.J. Buck, J. Hoffman, On-orbit depot architectures
using contingency propellant, Acta Astronaut. 96 (2014) 217–226.
[13] N.J. Bennett, A.G. Dempster, Geosynchronous transfer orbits as a market for impulse
delivered by lunar sourced propellant, Planet Space Sci. 182 (2020) 104843.
[14] A. Austin, B. Sherwood, J. Elliott, A. Colaprete, K. Zacny, P. Metzger, M. Sims, H.
Schmitt, S. Magnus, T. Fong, Robotic lunar surface operations 2, Acta Astronaut. 176
(2020) 424–437.
[15] E. Musk, Elon Musk: A future worth getting excited about | TED | Tesla Texas
Gigafactory interview, YouTube, TED. (2022).
https://www.youtube.com/watch?v=YRvf00NooN8 (accessed September 15, 2022).
[16] A. Glaser, Elon Musk will need one million Mars colonists to get the ticket price down to
$200,000, Vox. (2016). https://www.vox.com/2016/9/27/13081488/elon-musk-spacex-
mars-colony-space-travel-funding-rocket-nasa (accessed September 14, 2022).
[17] E. Musk, Tweet, Twitter. (2019).
https://twitter.com/elonmusk/status/1159964499975135232?s=20&t=InmP6TJwHGneoLI
GcvRFmA (accessed September 14, 2022).
[18] M. Ross, D. Toohey, M. Peinemann, P. Ross, Limits on the space launch market related to
stratospheric ozone depletion, Astropolitics. 7 (2009) 50–82.
[19] T.F.M. Brown, L. Revell, M.T. Bannister, T. Sukhodolov, E. Rozanov, An Inventory of
Global Rocket Launch Emissions and Projected Near-Future Impacts on Stratospheric
Ozone, Authorea Preprints. (2022).
[20] L. Miraux, Environmental limits to the space sector’s growth, Science of The Total
Environment. 806 (2022) 150862.
[21] M.N. Ross, P.M. Sheaffer, Radiative forcing caused by rocket engine emissions, Earths
Future. 2 (2014) 177–196.
[22] E.J.L. Larson, R.W. Portmann, K.H. Rosenlof, D.W. Fahey, J.S. Daniel, M.N. Ross,
Global atmospheric response to emissions from a proposed reusable space launch system,
Earths Future. 5 (2017) 37–48.
[23] J.D. Shutler, X. Yan, I. Cnossen, L. Schulz, A.J. Watson, K.-H. Glaßmeier, N. Hawkins,
H. Nasu, Atmospheric impacts of the space industry require oversight, Nat Geosci. 15
(2022) 598–600.
[24] R.G. Ryan, E.A. Marais, C.J. Balhatchet, S.D. Eastham, Impact of rocket launch and
space debris air pollutant emissions on stratospheric ozone and global climate, Earths
Future. 10 (2022) e2021EF002612.
[25] Mar. Ross, Jam.A. VeddA, The policy and science of rocket emissions, Center for Space
Policy and Strategy, The Aerospace Corporation. (2018) 2–10.
[26] M.N. Ross, K.L. Jones, Implications of a growing spaceflight industry: Climate change,
Journal of Space Safety Engineering. 9 (2022) 469–477.
[27] E. Sirieys, C. Gentgen, A. Jain, J. Milton, O. de Weck, Space sustainability isn’t just about
space debris: On the atmospheric impact of space launches, (2022).
[28] B.F. Kutter, G.F. Sowers, Cislunar-1000: Transportation supporting a self-sustaining
Space Economy, in: AIAA SPACE 2016, 2016: p. 5491.
[29] T. Bruno, E. Otto, W. Clifton, A. Schmitt, E. Lococo, J. Sanders, J. Reed, B. Kutter,
Creation of a U.S. Strategic Propellant Reserve, Medium. (2022).
https://medium.com/@ToryBrunoULA/creation-of-a-u-s-strategic-propellant-reserve-
b111044887e8 (accessed September 13, 2022).
[30] M. Sheetz, Bezos says Blue Origin will one day refuel its lunar lander with ice from the
moon, CNBC. (2019). https://www.cnbc.com/2019/06/19/jeff-bezos-blue-origin-will-
refuel-lunar-lander-with-moon-ice.html (accessed September 14, 2022).
[31] M.J. Listner, Op-ed: A U.S. return to the moon is about preserving the rule of law,
SpaceNews. (2020). https://spacenews.com/op-ed-a-u-s-return-to-the-moon-is-about-
preserving-the-rule-of-law/ (accessed April 14, 2022).
[32] E. Berger, NASA is supporting some seriously risky missions to the Moon—it’s about
time, Ars Technica. (2022). https://arstechnica.com/science/2022/04/nasas-other-moon-
program-is-about-to-take-center-stage/ (accessed January 14, 2023).
[33] A. Mettas, Reliability allocation and optimization for complex systems, in: Annual
Reliability and Maintainability Symposium. 2000 Proceedings. International Symposium
on Product Quality and Integrity (Cat. No. 00CH37055), IEEE, 2000: pp. 216–221.
[34] S.B. Stancliff, J.M. Dolan, A. Trebi-Ollennu, Mission reliability estimation for multirobot
team design, in: 2006 IEEE/RSJ International Conference on Intelligent Robots and
Systems, IEEE, 2006: pp. 2206–2211.
[35] S. Stancliff, J. Dolan, A. Trebi-Ollennu, Planning to fail: reliability as a design parameter
for planetary rover missions, in: Proceedings of the 2007 Workshop on Performance
Metrics for Intelligent Systems, 2007: pp. 204–208.
[36] Military Handbook (MIL-HDBK)-338B, Electronic Reliability Design Handbook. ,
Department of Defense, 1998.
[37] Y.-J. Yang, W. Wang, X.-Y. Zhang, Y.-L. Xiong, G.-H. Wang, Lifetime data modelling
and reliability analysis based on modified weibull extension distribution and Bayesian
approach, Journal of Mechanical Science and Technology. 32 (2018) 5121–5126.
[38] M. Ahmed, Size and efficiency: the case of the Irish plastics industry, Econ Soc Rev
(Irel). 8 (1976).
[39] H. Alem, G. Lien, S. Kumbhakar, J.B. Hardaker, Economies Of Scale And Scope In The
Norwegian Agriculture, 2017.
[40] S.E. Atkinson, R. Halvorsen, Parametric efficiency tests, economies of scale, and input
demand in US electric power generation, Int Econ Rev (Philadelphia). (1984) 647–662.
[41] A.G. Bailey, Q.M. Bui, J.D. Farmer, R.M. Margolis, R. Ramesh, Forecasting
technological innovation, in: ARCS 2012, IEEE, 2012: pp. 1–6.
[42] B. Bayram, G. İnce, Advances in Robotics in the Era of Industry 4.0, in: Industry 4.0:
Managing The Digital Transformation, Springer, 2018: pp. 187–200.
[43] P.M. Berthouex, Evaluating economy of scale, J Water Pollut Control Fed. (1972) 2111–
2119.
[44] H.M. Brown, A.K. Boyd, B.W. Denevi, M.R. Henriksen, M.R. Manheim, M.S. Robinson,
E.J. Speyerer, R. v Wagner, Resource potential of lunar permanently shadowed regions,
Icarus. 377 (2022) 114874.
[45] K.M. Cannon, A.N. Deutsch, J.W. Head, D.T. Britt, Stratigraphy of ice and ejecta deposits
at the lunar poles, Geophys Res Lett. 47 (2020) e2020GL088920.
[46] M. Celli, Determinants of Economies of Scale in Large Businesses—A Survey on UE
Listed Firms, American Journal of Industrial and Business Management. 3 (2013) 255–
261.
[47] T.J. Considine, Economies of scale and asset values in power production, The Electricity
Journal. 12 (1999) 37–42.
[48] J. Haldi, D. Whitcomb, Economies of scale in industrial plants, Journal of Political
Economy. 75 (1967) 373–385.
[49] R.E. Hall, Invariance properties of Solow’s productivity residual. NBER Working Paper
Series., Cambridge, MA, 1989.
[50] T. Jamasb, J. Köhler, Learning Curves for Energy Technology: A Critical Assessment, in:
Delivering a Low Carbon Electricity System: Technologies, Economics and Policy,
Cambridge University Press, 2008: pp. 314–332.
[51] S. Jin, S. Rozelle, J. Alston, J. Huang, Economies of scale and scope and the economic
efficiency of China’s agricultural research system, Int Econ Rev (Philadelphia). 46 (2005)
1033–1057.
[52] K. Junius, Economies of scale: A survey of the empirical literature. Kiel Working Paper
No.813, Kiel, Germany, 1997.
[53] D. Kennedy, Limits to Economies of Scale, in: IIE Annual Conference. Proceedings,
Institute of Industrial and Systems Engineers (IISE), 2011: p. 1.
[54] M.B. Lieberman, Market growth, economies of scale, and plant size in the chemical
processing industries, J Ind Econ. (1987) 175–191.
[55] J.A. Marshall, A. Bonchis, E. Nebot, S. Scheding, Robotics in mining, in: Springer
Handbook of Robotics, Springer, 2016: pp. 1549–1576.
[56] A. McDonald, L. Schrattenholzer, Learning rates for energy technologies, Energy Policy.
29 (2001) 255–261.
[57] S. Nag, J. LeMoigne, O. de Weck, Cost and risk analysis of small satellite constellations
for earth observation, in: 2014 IEEE Aerospace Conference, IEEE, 2014: pp. 1–16.
[58] B. Nagy, J.D. Farmer, Q.M. Bui, J.E. Trancik, Statistical basis for predicting technological
progress, PLoS One. 8 (2013) e52669.
[59] NASA, Learning Curve Calculator, Cost Estimating Website, National Aeronautics and
Space Administration. (2007).
https://web.archive.org/web/20120830021941/http://cost.jsc.nasa.gov/learn.html (accessed
July 3, 2022).
[60] O.H. Poensgen, M. Marx, Coping with or profiting from size, Eur J Oper Res. 22 (1985)
127–147.
[61] H. Rumy, The continuing importance of economies of scale in the automotive industry,
European Business Review. 97 (1997) 38–42.
[62] D. Sahal, A theory of progress functions, AIIE Transactions. 11 (1979) 23–29.
[63] P. Söderholm, T. Sundqvist, Empirical challenges in the use of learning curves for
assessing the economic prospects of renewable energy technologies, Renew Energy. 32
(2007) 2559–2578.
[64] H.K. Taylor, Mine valuation and feasibility studies, Mineral Industry Costs. Northwest
Min Assoc, WA, Spokane. (1977) 1–17.
[65] H.K. Taylor, Rates of working of mines-a simple rule of thumb, Institution of Mining and
Metallurgy Transactions. Section A. Mining Industry. 95 (1986).
[66] M.A. Tribe, R.L.W. Alpine, Scale economies and the “0.6 rule,” Engineering Costs and
Production Economics. 10 (1986) 271–278.
[67] F.-W. Wellmer, R.W. Scholz, What is the optimal and sustainable lifetime of a mine?,
Sustainability. 10 (2018) 480.
[68] E. Musk, Twitter, (2020).
https://twitter.com/elonmusk/status/1217993568482025472?s=20&t=EK0LQ1HiikNpJxfi
nRw2zw (accessed January 15, 2023).
[69] T.J. Colvin, K.W. Crane, R. Lindbergh, B. Lal, Demand Drivers of the Lunar and Cislunar
Economy, IDS Science & Technology Policy Institute, IDA Document D-13219. (2020).
[70] Anonymous, Commercial Space Transportation Study, Final Report, 1994.
[71] Anonymous, Future Spacelift Requirements Study, El Segundo, CA, 1997.
https://ntrs.nasa.gov/api/citations/19990025626/downloads/19990025626.pdf (accessed
January 9, 2023).
[72] D. Andrews, J. Andrews, Designing for the future space transportation missions, in: 37th
Joint Propulsion Conference and Exhibit, 2001: p. 3964.
[73] Anonymous, Launch costs to low Earth orbit, 1980-2100, FutureTimeline.Net. (2018).
https://www.futuretimeline.net/data-trends/6.htm (accessed January 14, 2023).
[74] T.G. Roberts, Space Launch to Low Earth Orbit: How Much Does It Cost?, Aerospace
Security. (2022). Space Launch to Low Earth Orbit: How Much Does It Cost? (accessed
January 14, 2023).
[75] Anonymous, Space Transportation Costs: Trends in Price Per Pound to Orbit 1990-2000,
Bethesda, MD, 2002.
[76] D. Todd, 2021 Launch Year: A new record for both orbital rocket and satellite launch
totals, Seradata Space Intelligence. (2022). https://www.seradata.com/2021-launch-year-a-
new-record-for-both-orbital-rocket-and-satellite-launch-totals-in-a-year/ (accessed January
14, 2023).
[77] Anonymous, ESA’s Annual Space Environment Report, Darmstadt, Germany, 2022.
https://www.sdo.esoc.esa.int/environment_report/Space_Environment_Report_latest.pdf
(accessed January 14, 2023).
[78] J. Williams, D.E. Lee, R.J. Whitley, K.A. Bokelmann, D.C. Davis, C.F. Berry, Targeting
cislunar near rectilinear halo orbits for human space exploration, in: AAS/AIAA Space
Flight Mechanics Meeting, 2017.
[79] D. Conte, M. di Carlo, K. Ho, D.B. Spencer, M. Vasile, Earth-Mars transfers through
Moon distant retrograde orbits, Acta Astronaut. 143 (2018) 372–379.
[80] R. Whitley, R. Martinez, Options for staging orbits in cislunar space, in: 2016 IEEE
Aerospace Conference, 2016: pp. 1–9.
[81] R.C. Woolley, J.D. Baker, D.F. Landau, A.K. Nicholas, Cargo logistics for a notional
mars base using solar electric propulsion, Acta Astronaut. 156 (2019) 51–57.
[82] T.S. Kokan, C.R. Joyner, D.J. Levack, B. Muzek, R.W. Noble, C.B. Reynolds,
Comparison of Human Mars Mission Approaches in the 2030s and Beyond with
Advanced Propulsion Options, in: ASCEND 2020, 2020: p. 4123.
[83] C.B. Reynolds, C.R. Joyner, T.S. Kokan, D.J. Levack, B.J. Muzek, Mars Opposition
Missions Using Nuclear Thermal Propulsion, in: AIAA Propulsion and Energy 2020
Forum, 2020: p. 3850.
[84] M. Qu, R.G. Merrill, P. Chai, End to End Optimization of a Mars Hybrid Transportation
Architecture, in: 2019 AAS/AIAA Astrodynamics Specialist Conference, 2019.
[85] R.C. Oeftering, A Cis-Lunar Propellant Infrastructure for Flexible Path Exploration and
Space Commerce, in: Space 2011 Conference and Exposition, 2012.
[86] T. Bennett, C. Cain, N. Campbell, A.J. Gemer, T. Green, T. Niederweiser, Engineering the
CisLunar Economic System based on ULA’s CisLunar-1000 Vision, in: AIAA SPACE
2016, 2016: p. 5305.
[87] N.J. Bennett, R. Xie, A.G. Dempster, The Moon and NEAs as sources of cislunar
propellant; removing some constraints from a recent paper drives down lunar sourced
propellant cost, Acta Astronaut. 190 (2022) 409–412.
[88] N.J. Bennett, A.G. Dempster, Lowering the barriers to lunar sourced propellant via
competitive parity pricing, Acta Astronaut. 191 (2022) 88–98.
[89] P. Dietz, W. Gärtner, Q. Koch, P.E. Köhler, Y. Teng, P.R. Schreiner, K. Holste, P.J. Klar,
Molecular propellants for ion thrusters, Plasma Sources Sci Technol. 28 (2019) 084001.
[90] K. Holste, P. Dietz, S. Scharmann, K. Keil, T. Henning, D. Zschätzsch, M. Reitemeyer, B.
Nauschütt, F. Kiefer, F. Kunze, Ion thrusters for electric propulsion: Scientific issues
developing a niche technology into a game changer, Review of Scientific Instruments. 91
(2020) 061101.
[91] Y. Matsuura, Y. Yano, A. Kakami, Magneto Plasma Dynamic Thruster Using Water
Propellant, TRANSACTIONS OF THE JAPAN SOCIETY FOR AERONAUTICAL
AND SPACE SCIENCES, AEROSPACE TECHNOLOGY JAPAN. 19 (2021) 523–528.
[92] A. van Paridon, E. Petro, Survey of water powered propulsion systems, in: AIAA
Propulsion and Energy 2021 Forum, 2021: p. 3566.
[93] A. Colaprete, P. Schultz, J. Heldmann, D. Wooden, M. Shirley, K. Ennico, B. Hermalyn,
W. Marshall, A. Ricco, R.C. Elphic, Detection of water in the LCROSS ejecta plume,
Science (1979). 330 (2010) 463–468.
[94] A.C. Charania, H. Kanamori, EXTENSIONS OF NASA’S EXPLORATION
ARCHITECTURE: Performance Capabilities and Market Economics of a Lunar
Propellant Production Facility, in: ISTS-2006-k-13, 25th International Symposium on
Space Technology and Science (ISTS), Kanazawa City, Ishikawa Prefecture, Japan, 2006.
[95] R. Mueller, J. Schuler, A. Nick, A. Wilkinson, C. Gallo, R. King, Lightweight bulldozer
attachment for construction and excavation on the lunar surface, in: AIAA Space 2009
Conference & Exposition, 2009: p. 6466.
[96] K. Zacny, R. Mueller, G. Galloway, J. Craft, G. Mungas, M. Hedlund, P. Chu, J. Wilson,
P. Fink, Novel approaches to drilling and excavation on the moon, in: AIAA SPACE 2009
Conference & Exposition, 2009: p. 6431.
[97] R. Mueller, J. Schuler, J.D. Smith, A. Nick, T. Lippitt, Reducing extra-terrestrial
excavation forces with percussion, in: 2013 IEEE Aerospace Conference, IEEE, 2013: pp.
1–11.
[98] K. Zacny, M.M. Cohen, W.W. James, B. Hilscher, Asteroid mining, in: AIAA Space 2013
Conference and Exposition, 2013: p. 5304.
[99] G. Paulsen, K. Zacny, C. McKay, L. Shiraishi, K. Kriechbaum, B. Glass, M. Szczesiak, C.
Santoro, J. Craft, R.B. Malla, Rotary-percussive deep drill for planetary applications, in:
Earth and Space 2010: Engineering, Science, Construction, and Operations in Challenging
Environments, 2010: pp. 1423–1436.
[100] G. Sowers, Thermal Mining of Ices on Cold Solar System Bodies. NIAC Phase I Final
Report., Golden, CO, 2020.
[101] C. Purrington, G. Sowers, C. Dreyer, Thermal Mining of volatiles in lunar regolith
simulant, Planet Space Sci. 222 (2022) 105550.
[102] Crunchbase, (n.d.). www.crunchbase.com (accessed August 21, 2022).
[103] P. Metzger, System for extracting water from lunar regolith and associated method,
(2021).
[104] A. Colaprete, D. Andrews, W. Bluethmann, R.C. Elphic, B. Bussey, J. Trimble, K. Zacny,
J.E. Captain, An overview of the volatiles investigating polar exploration rover (viper)
mission, in: AGU Fall Meeting Abstracts, 2019: pp. P34B-03.
[105] E. Ethridge, W. Kaukler, Microwave extraction of water from lunar regolith simulant, in:
AIP Conf Proc, American Institute of Physics, 2007: pp. 830–837.
[106] E.C. Ethridge, W.F. Kaukler, System of extraction of volatiles from soil using microwave
processes, (2013).
[107] E. Ethridge, W. Kaukler, Finite element analysis of three methods for microwave heating
of planetary surfaces, in: 50th AIAA Aerospace Sciences Meeting Including the New
Horizons Forum and Aerospace Exposition, 2014: p. 801.
[108] E. Ethridge, System for extraction of volatiles from planetary bodies using microwave and
RF processes, (2017).
[109] K. Zacny, P. Chu, G. Paulsen, A. Avanesyan, J. Craft, L. Osborne, Mobile in-situ water
extractor (MISWE) for Mars, Moon, and Asteroids in situ resource utilization, in: AIAA
SPACE 2012 Conference & Exposition, 2012: p. 5168.
[110] K. Zacny, K. Luczek, A. Paz, M. Hedlund, Planetary Volatiles Extractor (PVEx) for In
Situ Resource Utilization (ISRU), in: Earth and Space 2016: Engineering for Extreme
Environments, American Society of Civil Engineers Reston, VA, 2016: pp. 378–390.
[111] J. Sercel, Lunar-Polar Propellant Mining Outpost (LPMO): Affordable Exploration and
Industrialization, NASA Innovative Advanced Concepts, National Aeronautics and Space
Administration. (2019).
https://www.nasa.gov/directorates/spacetech/niac/2019_Phase_I_Phase_II/Lunar_Polar_P
ropellant_Mining_Outpost/ (accessed September 14, 2022).
[112] J.C. Sercel, A.P. Longman, J.G. Small, Systems and methods for radiant gas dynamic
mining of permafrost for propellant extraction, (2022).
[113] M. Kuhns, R. Kuhns, P. Metzger, K. Zacny, N. Rhys, Practical and Economic Rocket
Mining of Lunar Ice, in: Earth and Space 2021, 2021: pp. 444–457.
[114] J.C. Mankins, Technology readiness levels, A White Paper, Washington, D.C., 1995.
https://aiaa.kavi.com/apps/group_public/download.php/2212/TRLs_MankinsPaper_1995.
pdf (accessed September 14, 2022).
[115] J.C. Mankins, Technology readiness assessments: A retrospective, Acta Astronaut. 65
(2009) 1216–1223.
[116] N. Vetcha, M.C. Roman, M. Fiske, P. Carrato, K. Leucht, T.J. Prater, Overview of
NASA’s Break the Ice Lunar Challenge Phase 1, in: ASCEND 2022, 2022: p. 4214.
[117] R.P. Mueller, G.A. Murphy, 2011 NASA Lunabotics Mining Competition for
Universities: Results and Lessons Learned, in: Space Resources Roundtable/Planetary and
Terrestrial Mining Sciences Symposium, 2011.
[118] R. Mueller, P. van Susante, A review of lunar regolith excavation robotic device
prototypes, in: AIAA Space 2011 Conference & Exposition, 2011: p. 7234.
[119] R.P. Mueller, Lunabotics mining competition: Inspiration through accomplishment, in:
American Society of Civil Engineers, Earth & Space 2012 Conference, 2012.
[120] R.P. Mueller, P. van Susante, E. Reiners, P.T. Metzger, NASA Lunabotics Robotic
Mining Competition 10th Anniversary (2010–2019): Taxonomy and Technology Review,
Earth and Space 2021. (2021) 497–510.
[121] B. Kutter, F. Zegler, T. Bulk, B. Pitchford, J. Barr, Robust lunar exploration using an
efficient lunar lander derived from existing upper stages, in: AIAA SPACE 2009
Conference & Exposition, 2009: p. 6566.
[122] Advanced Missions Cost Model, Global Security. (n.d.).
https://www.globalsecurity.org/military/intro/reference/calc/AMCM.htm (accessed
October 12, 2022).
Appendix A. Table of Data.

Table A-1. Study-Specific Model Parameters

Tent Strip Mine,
Technology: Strip Mine then Sublimation N/A
Sublimation Beneficiate
Parameter Units K S CD J B M Baseline
Mass of LS capital t 30 17.6 20.94 214.8 26.84 2.5 25
1
Mass of space segment (RLL/OTV) t 5.86 5.861 5.0962 22.53 4.5 1.322 5
Payload and propellant capacity of
t 413 413 22 45 40.5 9.3 35
space segment
Inert Mass Fraction 0.26 (RLL)
– 0.1253 0.1253 0.188 0.10 0.124 0.10
0.13 (OTV)
Specific development cost, 𝜁D $K/kg 90.75 50.2 149.64 1226,8 1916 1176,9 120
Specific fabrication cost, 𝜁F $K/kg 22.75,6 41.97 60.84 30.56,8 47.66 29.46,9 40
Buildup Time y 58 5.5 8 58 58 38 5
Annual operations cost $M 23.9 78.6 53.4 158.210 20.910 10.0 50
Operational life y 10 10 10 14 10 5 10
Yearly mass of product at LS t 1640 1100 69.1 393.33 19011 27.9 500
Notes:
All costs have been converted to 2022 US dollars.
(1) DTAL in Fig. 9 of Kutter et al. [121].
(2) Inferred/calculated relying on additional calculations by Shisko [8].
(3) Assumes RLL cost from B
(4) assuming 80% of hardware cost is development and 20% is fabrication
(5) Includes fabrication of RLL prorated from B. Development of RLL will be a sunk cost.
(6) Approximated due to difficulty reconstructing the figures of J.
(7) Assumes $1,684M/kg obtained from Global Security cost estimating tool [122].
(8) Not given. Estimated here.
(9) Not given. Estimated here based on much smaller system than the others.
(10) Not given. Assumes value from K prorated by capital mass.
(11) Capacity of the system, though the demand is given as 166.44 t/y.