space
Industrial resource extraction from the Moon or asteroids operates by 2126.
Last revised Sep 13, 2026.
Current estimate
probability this has happened by each date
The full test below is met by 31 December 2035. Eased slightly because the near-term routes narrowed this cycle: PRIME-1, the only ice-mining experiment ever flown, never penetrated soil; Artemis III was redesigned as a low-Earth-orbit rehearsal; Chang'e-7 slipped out of 2026; and Interlune's first contracted delivery year of 2028 to 2029 leaves no room for twelve continuous months of operation before the deadline.
The full test below is met by 31 December 2050. The lag clock has still not started, because no extraction hardware has operated on the lunar surface for any duration. Set against that, this cycle found that the surface-consumption case is the favourable one in the agency's own model, which is the route most likely to resolve the claim early.
The full test below is met by 31 December 2100. This is where the survivorship correction bites hardest, because this horizon's probability was carried by the argument that a forty-to-seventy-year lag fits comfortably inside it, and that lag was fitted only to frontier industries that arrived.
Extraction of material from the Moon or an asteroid operates for at least twelve continuous months, supplying material that is sold or consumed by an actual operation, documented by the operating agency or company. A single demonstration payload, a sample return, and a plant that produces without its output being used all fail the test.
Resolution criteria
- 2035
- The full test below is met by 31 December 2035. Eased slightly because the near-term routes narrowed this cycle: PRIME-1, the only ice-mining experiment ever flown, never penetrated soil; Artemis III was redesigned as a low-Earth-orbit rehearsal; Chang'e-7 slipped out of 2026; and Interlune's first contracted delivery year of 2028 to 2029 leaves no room for twelve continuous months of operation before the deadline.
- 2050
- The full test below is met by 31 December 2050. The lag clock has still not started, because no extraction hardware has operated on the lunar surface for any duration. Set against that, this cycle found that the surface-consumption case is the favourable one in the agency's own model, which is the route most likely to resolve the claim early.
- 2100
- The full test below is met by 31 December 2100. This is where the survivorship correction bites hardest, because this horizon's probability was carried by the argument that a forty-to-seventy-year lag fits comfortably inside it, and that lag was fitted only to frontier industries that arrived.
- 2126
- Extraction of material from the Moon or an asteroid operates for at least twelve continuous months, supplying material that is sold or consumed by an actual operation, documented by the operating agency or company. A single demonstration payload, a sample return, and a plant that produces without its output being used all fail the test.
Probability history
| Revised | State | p(2035) | p(2050) | p(2100) |
|---|---|---|---|---|
hf-forecast-0207 | researched | 4% -1 pts | 20% -2 pts | 41% -1 pts |
hf-forecast-0197 | researched | 5% no prior revision | 22% no prior revision | 42% no prior revision |
hf-forecast-0019 | researched | — | — | — |
hf-forecast-0012 | initial prior | — | — | — |
Strongest evidence for
- The agency analysis that supplied this seed's strongest counterevidence says the opposite about the surface case. Jones et al. 2020 states that 'delivering propellant to the lunar surface is more expensive than delivering the ISRU systems and using them to produce propellant on the lunar surface' - and local consumption is precisely what this forecast's criterion admits (hf-evidence-0170).
- The prior cycle's headline negative is conditional, not absolute. Breakeven depends on plant lifetime rather than on demand: no breakeven at a one-year lifetime, increasing ranges of breakeven at three and five years (hf-evidence-0170).
- A second product class now has a published favourable case. Metzger and Autry find microwave sintering the economical way to build a lunar landing pad at current transport prices, and importing a finished pad is so unfavourable that it is not among the options modelled; NASA's own figures put the imported binder alone at 70 to 282 tonnes (hf-evidence-0174).
- Demand-side commitment continued to accumulate: a sovereign buyer, two industrial cryogenics firms, and delivery dates inside four years (hf-evidence-0172).
- The counterevidence lane searched specifically for a 2026 cancellation or defunding of lunar in-situ resource work at NASA, ESA, JAXA or any company, and found none. An absence established by looking is evidence.
Strongest evidence against
- The quantity the prior cycle named as decisive is disputed across the literature by roughly sixtyfold, from about 18 to about 1,164 kilograms of plant per tonne per year, while the improvement that would flip the make-or-ship answer is two to three fold. The only value with an engineering point design behind it sits at the pessimistic end (hf-evidence-0171).
- The power ratio the same model names as a lever has moved backwards. Fission surface power specific mass went from 145 kilograms per kilowatt in 2010 to between 171 and 188 in 2025, and the 2025 design states plainly that 'Neither of the two technology options satisfy the SOW goal of 6000 kg' (hf-evidence-0171).
- This seed's forty-to-seventy-year frontier lag was fitted only to industries that arrived. Ocean thermal energy conversion was legislated for 10,000 megawatts by 1999 and reached about 100 kilowatts by 2013; seabed nodule collection was demonstrated by prototypes in the late 1970s and still has no market-ready machines (hf-evidence-0175).
- The independent export path is not independent of transport after all, and at today's prices it is short by about two orders of magnitude (hf-day-0021-takeaway-01, from quantities in hf-evidence-0172 and hf-evidence-0173).
- No extraction hardware has ever operated on the Moon. PRIME-1, the only ice-mining experiment flown, 'was still extended, but did not penetrate lunar soil' after its lander tipped, and the account where extraction research lives is proposed down 32 percent for FY2027 (hf-evidence-0176).
- Still nobody has disclosed a price for off-Earth material. Six years on, the only priced transactions remain four contracts totalling 25,001 dollars, none delivered (hf-evidence-0172, hf-evidence-0072).
Present uncertainties
- Whether ISRU plant specific mass is nearer 18 or nearer 1,164 kilograms per tonne per year, which no amount of further reading can settle because the disagreement is between paper designs with no arbiter; and whether any plant can run unattended for the multi-year lifetime on which both the economics and this forecast's criterion depend.
Next discriminating observation
A measured plant mass and a measured output rate from any operating in-situ extraction hardware, on any surface, which would replace the entire sixtyfold published spread with one number. Second, and available sooner: the demonstrated unattended operating lifetime of any lunar surface system, since the agency model makes that the variable on which breakeven turns and this forecast's criterion requires twelve continuous months of it. A first lunar plant that runs three months and stops would be strong evidence against, and it would look like success in every press release.
Sources
- The ratio the prior cycle said nobody publishes is published, and the two models of the same plant disagree by 8.3 times
Jones et al., 'Cost Breakeven Analysis of Lunar In-Situ Propellant Production for Human Missions to the Moon and Mars' (NTRS 20205007564, dated 14 September 2020, presented at AIAA ASCEND 2020), prints the exact quantity hf-forecast-0197 named as the decisive and unreported one. Its Table 5 gives ISRU Plant Specific Mass as 1.1101 t of plant mass per t/yr produced for the Molten Regolith Electrolysis Equivalent model and 0.1331 for the Duke model, and ISRU Plant Specific Power as 5.6976 and 0.6174 kWe per t/yr produced respectively. The paper states the Duke model 'is approximately 8.3 times more mass efficient'. Power is assumed to come from 'as many copies as needed of the 40 kWe system, with a specific mass of 174 kg/kWe, described by Mason'. The spares rate is 0.1 kg/kg of system/year, and the modelled demands are 17 t/yr on the lunar surface pre-transition and 59 t/yr in cislunar space post-transition. Two further results matter more than the ratios. First, breakeven is conditional and the condition is LIFETIME, not demand: 'at lifetimes of 1 year ISRU does not breakeven in the canonical case, while the range of cases and the magnitude of the breakeven both increase for the Duke model as the lifetime goes from 3 years to 5 years; for the MREE model, there are only a small set of breakeven cases at 5 years ISRU lifetime', and the abstract states that 'without long lifetime ISRU systems, with greater than 5 years of autonomous operation before replacement, the demand in cis-lunar space for a Mars campaign favors propellant delivery from Earth.' Second, and decisive for this seed's resolution criterion, the paper separates the surface case from the export case: 'delivering propellant to the lunar surface is more expensive than delivering the ISRU systems and using them to produce propellant on the lunar surface; however, delivering propellant to cislunar space is less expensive that implementing a full propellant production architecture without infinitely reusable vehicles.'
Limit: The two evidence_refs are the same document, so this record is internally primary but not independently corroborated; no second research group publishes a competing plant-specific-mass figure that this cycle could find, which is itself the finding. Both ratios are paper designs and neither is hardware: the efficient Duke model traces to a 2006 book chapter, 'Development of the Moon' in Reviews in Mineralogy and Geochemistry, and the 174 kg/kWe power figure to a 2010 NASA/DOE Fission Surface Power concept definition. The authors overlap with those of hf-evidence-0070, so this is a successor analysis by the same group rather than a rival. The model assumes oxygen-hydrogen propulsion throughout and lossless propellant storage and transfer, and the paper says explicitly that if propulsion choices go elsewhere 'lunar ISRU may not benefit either campaign'. It is still about propellant only.
The PDF was downloaded directly from NTRS on 2026-09-13 (2,565,932 bytes, 990 lines of extracted text) and every figure and sentence above was read from the extracted text, not from a summary. The NTRS citation page was fetched separately and supplied the abstract, the eight named authors, the 14 September 2020 date and the ASCEND 2020 venue. The rascal.nianet.org host serving the 2019 companion paper that hf-evidence-0070 rests on returned an Incapsula bot-protection stub on both HTTP and HTTPS on 2026-09-13 and could not be re-fetched.
- The decisive ratio is disputed across the literature by a factor of about sixty, and the only value with an engineering point design behind it is the worst one
Four documents, fetched separately, place lunar ISRU plant mass per tonne per year of output across roughly sixty-fold. NASA Glenn's Compass Team pilot-plant design (NTRS 20205008294) states that a 'CLPS-style lander can deliver ISRU system to extract 1125 kg of water in a year to produce 1000 kg of oxygen' and that a '325 kg ISRU system mass estimate leaves 7.5% margin on 350-kg payload lander', with a second Ridge Plant at 839 kg and 4,306 W, assuming 'Regolith in PSR contains 5% water by mass' and 'Assume 75% water extraction/capture'. A JPL cost-symposium presentation (Shishko, Stoica and Mrozinski 2019) characterises two further studies: Kornuta et al.'s Commercial Lunar Propellant Architecture as a '30 mt mine producing 1640 mt/year of propellant', and Charania and DePasquale 2007 as a '20.94 mt mine (including nuclear power plant) produces 49.4 mt/year of propellant'. The power side is firmer and moving the wrong way. Mason and Poston's 2010 NASA/DOE concept states 'The total system mass without margin is 5820 kg' for 40 kWe. Mason et al.'s 2025 NETS paper (NTRS 20250000841) gives a design goal of 40 kWe for 10 years at a total mass of '6000 kg including margin & growth', reports Total FSP with MGA of 7,502 kg for the gas-cooled Brayton option and 6,820 kg for the heat-pipe Stirling option, and states plainly that 'Neither of the two technology options satisfy the SOW goal of 6000 kg.' The same paper projects that for 'industrial scale mining and propellant production... a larger FSP in the 80-100 kWe class may be needed', at about 11.0 t and 12.5 t respectively.
Limit: Every number here is a paper design; none is flight hardware, and the most advanced physical item in the set is a power-conversion risk-reduction testbed. The four are not commensurable as stated: Charania's 424 kg per tonne per year explicitly includes the nuclear plant while Kornuta's 18 kg per tonne per year appears not to, and the pilot plant is deliberately minimal, sized to a 350 kg lander, so it carries no economies of scale and produces oxygen rather than full propellant. The Kornuta figure is a one-line summary on a JPL slide and this cycle did not read the underlying report: the USRA repository copy returned 403 and the ScienceDirect article is paywalled, so it must not be quoted as a CLPA figure. The 109 kilograms per tonne per year baseline the previous cycle used could not be verified at all, because both hosts serving the 2019 paper return an Incapsula bot challenge and the AIAA copy is paywalled.
All four documents were downloaded and read by the master on 2026-09-13. The pilot-plant and 2025 fission papers were re-fetched independently of the lane that first located them, converted with pdftotext, and the quoted strings grepped from the extracted text; the 6,000 kg design goal, the 7,502 and 6,820 kg totals and the 'Neither of the two technology options' sentence were each read in place, as were the 1,125 kg water, 325 kg and 839 kg figures. The 2010 and JPL figures are reported as the lane read them.
- The first disclosed price for off-Earth material is still not disclosed, but the sovereign buyer pegged it to the terrestrial market and two independent routes put that market near nineteen million dollars a kilogram
The observation hf-forecast-0197 named as most discriminating - the first disclosed PRICE for off-Earth material - has not occurred, and the near-miss is more informative than a price would have been. Interlune's release of 7 May 2025 states that 'The U.S. Department of Energy Isotope Program (DOE IP) has agreed to purchase three liters of helium-3 harvested from the Moon for delivery on Earth at approximately today's commercial market price', delivery 'no later than April 2029', and describes it as 'the first DOE Isotope Program purchase of a non-terrestrial natural resource'. No unit price is stated. Bluefors' release of 16 September 2025 states it 'has agreed to purchase up to ten thousand liters of helium-3 annually, for delivery from 2028 to 2037' and discloses neither a total value nor a price. Two independent routes to the reference market price agree: Payload of 8 May 2025 reports helium-3 at 'about $20M per kg, according to Interlune CEO Rob Meyerson', and Gizmodo of 20 September 2025 reports it trades at 'around $2,500 per liter, according to a 2024 estimate from The Edelgas Group'. At the standard molar volume, 2,500 dollars per litre is about 18.7 million dollars per kilogram, which is the master's own arithmetic and not a figure either source prints.
Limit: None of this is a transaction. Every agreement is forward, no material exists, and the quantities are small: three litres for the DOE, and 'up to' ten thousand litres a year for Bluefors is a ceiling whose floor may be zero. The reference clause 'approximately today's commercial market price' points at a market with no published index, so it fixes a price only in the sense that it forbids a premium. Two figures in circulation are irreconcilable with the ones above and were excluded: a Quantum Insider range topping out near 150 million dollars a kilogram, about eight times the converging estimate, and a widely repeated 300 million dollar value for the Bluefors agreement which does not appear in the Bluefors release nor in either secondary article fetched here and should not be used. Interlune's own two 2026 releases state incompatible capital-raised totals. GeekWire and gasworld, the likely origin of several circulating figures, returned 403 to this cycle.
All four URLs were fetched by the master on 2026-09-13, independently of the lane that located them, and every quoted string above was read from the fetched page. The two price routes are genuinely independent: one is a company CEO speaking to a trade outlet, the other a third-party industrial-gas consultancy estimate quoted by a different outlet five months later.
- A NASA review gives the physical scale of helium-3 mining: fifty tonnes of machine moving five million tonnes of regolith a year for thirty-three kilograms of product
Aaron D. S. Olson of NASA Kennedy Space Center, 'Lunar Helium-3: Mining Concepts, Extraction Research, and Potential ISRU Synergies' (NTRS 20210022801), supplies the grade and the machine. On grade: 'Apollo 11 samples (from Neil Armstrong's bulk sample 10084) averaged 11.8 parts per billion (ppb) and ranged from 9.22 to 17.9 ppb in 3He concentration', and 'Helium concentration does not appear to be correlated with depth into the lunar surface'. On the machine, the Mark III design of 2006: 'The M-3 was designed to collect 33 kg of 3He per year. The miner would excavate down to a 3 m depth over a 1 km2 area each year with a 4 m diameter bucket wheel excavator... The 33 kg/yr collection rate assumes a 10 ppb 3He concentration and that the miner only operates during 90% of the lunar daytime (3942 hours/year). The M-3 is designed to excavate 1258 tonnes/hr, heat 556 tonnes/hr, move at 23 m/hr and consume about 350 kW of electrical power.' And on the total: 'an entire mining operation based on the M-3 miner was estimated to have a mass of ~50 tonnes and would require that 5 million tonnes of regolith be excavated.' An alternative in-place heating concept using three 100 m domed enclosures is given as 30 kg of helium-3 a year for a mining operation 'estimated at 174 tonnes'. For comparison, the prototype excavator Interlune unveiled in May 2025 is reported by Payload as 'designed to suck up 100 metric tons of lunar regolith per hour', about one twelfth of the M-3 rate.
Limit: The M-3 is a 2006 paper design and no helium-3 mining hardware has ever operated anywhere. The Apollo 11 grade is from one sample site in a mare region, and the paper itself notes a 20 ppb figure for undisturbed maria regolith which would double the yield. Interlune's own plant mass, power and extraction efficiency are not published, so any comparison against the M-3 is a scaling assumption rather than a measurement. The paper is a review of concepts, and its purpose is to describe mining proposals, not to endorse their economics.
The NASA paper was downloaded by the master on 2026-09-13 (1,465,166 bytes, 15 pages), converted with pdftotext and each quoted sentence grepped from the extracted text. The Payload throughput figure was read from the fetched article on the same day. The two sources are independent: an agency technical review and a trade report of a company announcement.
- For landing pads and shielding, importing the finished product is so absurd that nobody models it - and the only crossover anyone does model runs toward importing MORE from Earth as launch gets cheaper
The prior cycle flagged that its strongest counterevidence was about propellant only, and that shielding, pads and structure were unmodelled. They are partly modelled, and the result does not transfer. Metzger and Autry, 'The Cost of Lunar Landing Pads with a Trade Study of Construction Methods' (New Space, 2023; preprint arXiv 2205.00378), is a full cost model across five transport prices. It finds 'microwave sintering is currently the most favorable method to build the inner, high temperature zone of a lunar landing pad', that the outer zone 'switches to polymer infusion when transportation costs drop below about $110K/kg to the lunar surface', and that a pad costs '$229M assuming that transportation costs will be reduced modestly from their current rate ~$1M/kg to the lunar surface to $300K/kg. It drops to $130M when the transportation cost drops further to $100K/kg, or to $47M if transportation costs fall below $10K/kg.' On the import side, NASA Kennedy's polymer-pad paper (NTRS 20220000334) tabulates the binder alone - with 85 percent of pad mass already local - at 282 tonnes for a 200 m pad, 70 tonnes at 100 m and 3 tonnes at 15 m, and notes that 'SpaceX's Starship is targeting 100-200t payload to the lunar surface. A 200m diameter pad is not feasible with a single Starship payload'. For shielding, Eric He's 2026 review in the Journal of Radiological Protection states that 'NASA notes that estimated delivery to the lunar surface is roughly $1.2 M per kg in 2023. At those rates, even a modest 1 metric ton of shielding material corresponds to roughly $1.2B in delivery cost', and cites 'an early NASA RASSOR concept assumed small robotic precursor deliveries with <500 kg total payload mass, while targeting delivery on the order of 255 500 kg of regolith over a five year campaign.' But the same review sets the requirement low, recommending 'moderate regolith areal densities on the order of 30-40 g cm-2... roughly 15-25 cm of compacted lunar soil', and says further protection should come from 'lightweight, hydrogen-rich materials brought from Earth'.
Limit: Metzger and Autry is a cost model, not hardware; microwave sintering is not flight-demonstrated; and the paper's own sensitivity analysis makes program-delay cost one of its two dominant variables, resting on an assumed outpost valuation that is an assumption rather than a measurement. The NASA pad testing behind the polymer figures used a 444 newton thruster in two-second burns, four orders of magnitude below a real lander. The He review is a review of simulations and beam experiments with no flight hardware, and its dose response is explicitly non-monotonic, rising again near 90 g per square centimetre through secondary neutrons, so a linear sizing model overstates the mass required. A widely circulating figure of 10 to 11 tonnes of regolith per square metre conflicts with this review by about thirtyfold and was excluded.
All three documents were downloaded and read by the master on 2026-09-13, independently of the lane that located them, converted with pdftotext, and each quoted string grepped from the extracted text. Re-reading the Metzger and Autry paper in full corrected the lane's framing of it, which is recorded here because the correction matters: the $110K/kg crossover is between two IN-SITU construction methods, sintering versus polymer infusion, and not between building locally and importing a pad. Importing a finished pad is not among the options the paper models at all.
- The forty-to-seventy-year frontier lag this seed relies on was fitted only to industries that arrived, and at least two comparably confident frontiers are fifty and sixty years past their forecasts with nothing to show
hf-evidence-0078 estimated a forty-to-seventy-year lag from demonstration to material production, from four cases - offshore oil, hydraulic fracturing, oil sands and marine diamonds - every one of which is an industry that arrived. The sample is selected on the outcome, and the missing denominator is populated. Ocean thermal energy conversion: Eos reports that President Carter 'signed a bill calling for 10,000 megawatts of OTEC capacity to be up and running by 1999', against which the operating plants the same article names are Kumejima, Japan in 2013 at approximately 100 kilowatts, and a Hawaii plant 'generating only enough electricity to run about 120 homes'. The realised capacity is roughly one hundred-thousandth of the legislated target, twenty-seven years past its deadline. The article attributes the failure not to physics but to an 'innovation valley of death' in which 'Investment would almost certainly follow a successful OTEC plant' but does not precede one, and to the fact that 'the meteoric rise of wind and solar power has simply overtaken marginal renewables like OTEC'. Seabed manganese nodules: World Ocean Review states that collection 'Prototypes in the late 1970s and early 1980s have shown that this will work', and that 'Manganese nodule mining at an industrial scale is presently not possible because there are no market-ready mining machines', naming an unresolved regulatory framework and that 'the cobalt price has fallen steeply, so that mining of the deposits is not presently economical.' Neither industry exists in 2026.
Limit: This is a hand-picked sample selected FOR failure, and it is therefore no more a base rate than the four arrivals were. The honest conclusion is that the denominator is unknown, not that a pessimistic point estimate should replace an optimistic one. Both cases are also truncated rather than resolved: neither OTEC nor seabed nodule mining has been abandoned, and a permit application for the latter was live in 2026, so they establish a long lag rather than non-arrival. The World Ocean Review text dates from 2014 and its statement about the International Seabed Authority's timetable is superseded. Most importantly, both failed cases had a cheap terrestrial substitute for the same product, which lunar material used in space does not.
Both URLs were fetched by the master on 2026-09-13 and each quoted string was read from the fetched page. The two are independent of one another and of the four cases in hf-evidence-0078. Two further candidate cases raised by the counterevidence lane, commercial fusion and space-based solar power, were excluded from this record because this cycle did not fetch a primary or near-primary source for their original forecast targets.
- No ice-extraction hardware has ever operated on the Moon, and the budget account where extraction research lives is proposed down a third
The Planetary Society's analysis of the FY2027 request, published 23 April 2026, gives a total of 18,829.10 million dollars against 24,438.30 million in FY2026, a 23 percent cut, and a Space Technology line of 624.3 million dollars against 920.5 million, a 32 percent cut. Space Technology is the account in which in-situ resource utilisation research sits, and neither that analysis nor this cycle could locate an explicit ISRU line item in either direction. The same organisation's CLPS tracker records the flight history: Peregrine 'suffered a propellant leak and was directed to reenter Earth's atmosphere on Jan. 18, 2024'; IM-1's Odysseus tipped onto its side but was 'considered a success'; IM-2's Athena 'tilted on its side during landing' and its PRIME-1 drill 'was still extended, but did not penetrate lunar soil'; Blue Ghost Mission 1 'successfully soft landed' and completed its mission. PRIME-1 was the first and only in-situ ice-mining experiment ever flown, so as of September 2026 no extraction hardware has operated on the lunar surface for any length of time. Separately, The Conversation reported in March 2026 that Artemis III has been 'Redesigned as a low Earth orbit rehearsal mission lasting 30 days, testing docking with lunar landers rather than visiting the Moon', and that 'The Exploration Upper Stage has been canceled'.
Limit: A budget request is not an appropriation, and the counterevidence lane established that Congress rejected an identical prior request and appropriated 24.4 billion dollars instead, so the 32 percent cut may never occur. The same request adds 175 million dollars for robotic Lunar Base Camp precursors, which is pro-extraction. The CLPS record is four missions across three vendors of first-generation hardware, which is a weak base rate, and the two failures were landing-attitude failures upstream of any extraction question. Every item here is decade-scale evidence being asked to speak about a century, and the lane that produced it said so.
The two Planetary Society pages were fetched by the master on 2026-09-13 and the budget figures and mission outcome sentences read from them directly. The Artemis III redesign is reported here as the counterevidence lane fetched it from The Conversation and is not master-verified; it is included because it is corroborated in outline by the budget analysis's own framing, and it is marked so a later reader does not treat it as first-hand.