The Long Bootstrap: a cislunar roadmap, 2026–2120

A workshop draft, written May 2026, revised June, with the Gateway status corrected in October 2026. About nineteen thousand words. The technical and policy readers it was written for have not seen it yet, so treat it as a position under development rather than a finished paper.

---

1. The argument in one paragraph

The published cislunar roadmaps converge on the same architecture. DARPA’s LunA-10 Field Guide, the National Space Society’s Roadmap to Space Settlement, the Aerospace Corporation’s CSPS analyses, the AIAA ASCEND infrastructure papers, and the Chinese Academy of Space Technology’s Queqiao constellation plan describe substantially the same future: a phased buildout from government-anchored exploration, through propellant-economy bootstrap, to cislunar manufacturing for cislunar customers, with Lagrange-point depots as connective tissue and lunar surface industry as the population and manufacturing anchor. The architecture is settled. What isn’t settled — and what the current commercial discourse is getting wrong — is which layer of the architecture deserves capital now. The visible hardware draws investment. The invisible infrastructure determines whether the architecture works. The bottleneck is inverted from where the capital flows. This document argues the case, names the gating technical and policy milestones, and identifies three structural problems that the current orbital-data-centre boom has not priced in. The piece is a synthesis, not a forecast. The components are in the literature. The integration, and the corrections, are mine.

2. Why this document exists now

Three things changed in the eighteen months ending May 2026. Starcloud raised $170M Series A at a $1.1B valuation, becoming the fastest unicorn in Y Combinator history. NVIDIA announced the Space-1 Vera Rubin module, designed for orbital data centres at twenty-five times H100 compute density. SpaceX, Blue Origin, China, and Starcloud filed for FCC and orbital-slot allocations totalling more than two hundred thousand satellites between them, with Anthropic publicly expressing interest in multi-gigawatt orbital compute as part of a SpaceX deal in early May. Capital is flowing into orbital compute at a velocity the sector has not seen.

The capital is mostly flowing to the wrong layer. The orbital-data-centre thesis as currently pitched — frontier training compute in low Earth orbit, mega-constellations of compute satellites, terrestrial-replacement timelines under ten years — has three structural problems that the published roadmaps already imply but don’t state plainly. The chip-refresh problem, which I’ll develop in Section 8, kills the frontier-training thesis on its own. The bottleneck-inversion problem, in Section 9, explains why the capital is allocated wrong even when the technical case works. The architectural displacement problem, in Section 10, explains why the correct version of the orbital-DC story isn’t in low Earth orbit at all but at the Earth-Moon Lagrange points, twenty years later than the current decks suggest, and supplied by lunar industry that doesn’t yet exist.

This document aims to do three things. First, integrate the published roadmaps into a single layered architecture and show what they agree on. Second, identify the structural problems the commercial discourse is missing. Third, name the actual leverage points for state and private actors who want this future to happen on the faster end of the plausible range. The intended readers are agency program managers, congressional and parliamentary staff, science advisers, capital allocators with multi-decade horizons, and the small number of founders willing to bet on the layers of the architecture that aren’t currently fashionable.

3. The geography of cislunar

Cislunar space is not a region. It’s a graph. Five nodes carry essentially all the traffic that matters for the next century, and the topology of that graph determines what’s economically possible.

Low Earth Orbit (~400–2000 km)

LEO is where today’s space economy lives. Roughly ten thousand active satellites as of mid-2026, dominated by Starlink, Chinese Guowang, OneWeb, Earth-observation constellations, and a long tail of science and government missions. LEO is congested, jurisdictionally contested, and structurally unsuitable for large-scale data centres because of debris density, atmospheric drag (which forces propellant expenditure for station-keeping and limits useful lifetime), and the ground-track problem: a LEO satellite spends most of its orbit out of contact with any given ground station, which forces either a globe-spanning ground network or laser intersatellite links of the sort Starlink and Kepler are deploying.

LEO’s role in the cislunar economy is the staging layer. Earth-launched mass arrives in LEO, gets aggregated, refuelled at LEO depots, and dispatches outward. LEO is the harbour, not the destination. The current orbital-DC thesis treats LEO as the destination, which is the first architectural mistake.

Geostationary Orbit (~36,000 km)

GEO carries about five hundred operational satellites, mostly comms and weather, with a fleet value north of fifteen billion dollars and an aging average vintage. GEO satellites don’t move relative to the ground, which makes them ideal for fixed-pointing services (broadcast comms, weather imaging) and useless for many newer applications that LEO handles better. The interesting GEO question for cislunar economics is the satellite-servicing market — life-extension, refuelling, repositioning, eventually full repair. Northrop Grumman’s Mission Extension Vehicle docked with Intelsat-901 in 2020 and proved the technical case. What’s missing is propellant cheap enough to make servicing routinely profitable, which is what the propellant economy of the 2030s is supposed to deliver.

GEO’s role is the first paying commercial customer for cislunar logistics. Every economic analysis that closes runs through GEO servicing as the bridge between government-anchored Stage 1 and self-sustaining Stage 2. If GEO servicing fails to scale, the bootstrap stalls.

Earth-Moon L1 (~326,000 km)

L1 sits between Earth and the Moon, at the gravitationally balanced point where the Earth’s and Moon’s pulls cancel along the line connecting them. Strictly, stable infrastructure parks in a halo orbit around L1 rather than at the point itself, but the operational logic is the same: a piece of hardware at L1 holds station for roughly ten metres per second of delta-v per year, two orders of magnitude cheaper than LEO station-keeping.

The delta-v geometry around L1 is what makes it the most important node in the entire architecture. From L1 to the lunar surface costs about 2.5 km/s. From L1 to GEO, about 1.7 km/s. From L1 to LEO, about 3.8 km/s — higher than the others because you’re descending into Earth’s gravity well from above. From L1 outbound to L2, deep space, or trans-Mars injection, the delta-v costs are also low. L1 is, in delta-v terms, simultaneously close to everywhere that matters and cheap to maintain. No other point in cislunar shares this property.

L1 has continuous line of sight to Earth and to the lunar near side. It receives essentially uninterrupted solar power. It sits outside the LEO debris environment and outside the radiation belts. For a propellant depot, a satellite-servicing hub, an aggregation node where Earth-launched cargo meets lunar-launched cargo, an assembly yard for structures too large to launch monolithically — L1 is the answer to all of these. The University of Colorado published this architecture in 1989. The National Space Society’s Roadmap Milestone 14 published it again in 2021. Neither received the funding that would have moved an L1 depot from concept to construction.

The Lunar Gateway, the station NASA and its partners were building for a Near-Rectilinear Halo Orbit around the Moon, was paused in March 2026 when NASA redirected the program toward a lunar surface base, with Gateway hardware and partner commitments to be repurposed where they can be. It was never going to be an L1 facility. NRHO has its own advantages — easier access from Earth using Orion, good lunar polar coverage — but it’s not the same node, and a Gateway-class facility at NRHO would not have substituted for an L1 depot. The cancellation didn’t move the money toward L1; it moved it to the surface. As of late 2026 neither orbital node is funded.

Earth-Moon L2 (~449,000 km)

L2 sits beyond the Moon as seen from Earth. China’s Queqiao-1 (2018) and Queqiao-2 (2024) operate near L2 in halo orbits, providing communications relay for Chinese far-side lunar missions including Chang’e-4 and Chang’e-6. The Lunar Gateway, before NASA paused it in March 2026, was planned for a near-L2-equivalent halo orbit. L2 has unique advantages for radio astronomy (the Moon shields it from Earth’s electromagnetic noise, making it the best radio-quiet location in cislunar) and for deep-space mission staging (lower delta-v to outer solar system trajectories than from L1 or LEO).

L2’s economic role is narrower than L1’s. It’s a science and strategic-comms node more than a commercial logistics node. But the shared infrastructure with L1 — comms relays, autonomous logistics, depot operations — means the L1 buildout effectively subsidises L2 operations. Once L1 exists, L2 follows for marginal cost.

Lunar surface

The lunar surface has its own gravity well, but a shallow one. Reaching lunar orbit from the surface costs about 1.6 km/s of delta-v, less than half what’s needed to reach LEO from Earth’s surface. The Moon’s economic role is the population, manufacturing, and ISRU anchor for the entire cislunar system.

The interesting fact about lunar surface delta-v geometry is that the Moon is not a cheap place to launch missions back to Earth orbit. Lunar surface to LEO costs roughly 6 km/s (you have to fight Earth’s gravity well from the wrong direction); lunar surface to GEO is about 2.5 km/s. So the Moon is a good base for serving cislunar and outbound traffic, and a poor base for serving Earth-orbit traffic. This asymmetry is why the right mental model for lunar industry is supplying itself and points outward, not supplying Earth or Earth’s orbits.

Permanent shadow regions at the lunar south pole contain water ice. How much is honestly unknown: mapped onto a terrestrial resource classification, the polar ice is an inferred resource at best — one impact assay (LCROSS, 5.6 ± 2.9 wt% water in the Cabeus ejecta), neutron-spectrometry hydrogen signals at tens-of-kilometres resolution, and aggregate estimates that span orders of magnitude depending on assumptions about the ice’s physical form. The widely cited figures in the hundreds of millions of tonnes are remote-sensing extrapolations, not measurements; the first in-situ assays arrive with Chang’e-7 (2026) and the revived VIPER (late 2027), and extraction at industrial rates remains undemonstrated. Polar peaks of near-eternal sunlight provide continuous solar power for surface operations. The combination — accessible water plus continuous power plus the gravity-well asymmetry favouring outbound supply — makes the lunar south pole the unique location for cislunar industrial bootstrap. Every serious roadmap converges on this. So does every serious national program currently funded.

What the geography implies

Three things follow from the topology, and they structure everything in Sections 4 through 7.

First, the cislunar economy is not lunar. It’s a five-node network in which the lunar surface is one node among several. Roadmaps that focus only on lunar surface infrastructure miss the connecting structure that makes lunar industry economically meaningful. Roadmaps that focus only on Earth-orbit infrastructure miss the resource base that makes the connecting structure affordable. Both failures are common in the published literature. The integration in this document is the correction.

Second, L1 is the keystone. Without an L1 depot, the cislunar economy is a collection of point-to-point missions, each carrying its own propellant from Earth, each amortising its launch cost over a single trajectory. With an L1 depot, missions become routes, propellant becomes a commodity, and the network effect compounds. The single highest-leverage piece of public infrastructure available in the 2026–2035 window is an L1 depot that doesn’t currently exist on any funded program of any space agency. This is the central anomaly the roadmap analysis exposes.

Third, the geography is fixed but the timing is not. The five-node topology, the delta-v costs, the location of lunar polar ice — none of this is going to change. What changes is when each node comes online, who funds it, and which side of the emerging Artemis-Accords-vs-Sino-Russian governance split captures the standards and the precedents. The geography is destiny; the politics is contingent. The window for shaping the contingent part is roughly the next decade.

The remainder of this document works through the layered buildout the geography implies, the gating questions that determine timing, the structural problems the current commercial discourse is missing, and the leverage points for actors who want the architecture built on the faster timeline rather than the slower one.

4. Layer 0 — Foundations, 2026–2032

The decisions made in this window lock in the next fifty years. None of the visible 2026 hardware matters as much as the four invisible foundations that have to exist for the architecture to work: launch cost on a sustained downward curve, an L1 depot, autonomous ISRU at meaningful scale, and federated cislunar situational awareness. Three of the four are underfunded relative to their option value. The fourth — launch cost — is on track but more fragile than the public discourse acknowledges. This is the layer where state actors have the most leverage and the least time. It’s also the layer where commercial capital is most badly misallocated, because the foundations are infrastructure and infrastructure doesn’t pitch well to venture timelines.

The Starship cost-curve dependency

Everything downstream of Layer 0 assumes Starship reaches sustained reliable cadence at roughly $500–1500/kg to LEO by 2030, and continues falling toward $200–500/kg through the early 2030s. This is the enabling assumption for the entire architecture. If it holds, every other timeline in this document is in range. If it slips by a factor of two on cost, or three years on schedule, the whole sequence pushes out by roughly a decade per factor.

The public reporting on Starship costs is mostly stated prices rather than achieved prices, which is the wrong number. The right metric is cadence-times-reuse-turn-time. A Starship that flies once a month at theoretical $200/kg is not the same engineering achievement as one that flies twice a week at $1000/kg; the second is more economically transformative even at five times the unit cost, because the network effects of high cadence dominate the marginal-mass economics. As of mid-2026, Starship is closer to the first than the second. Whether it transitions to the second within the Layer 0 window is the single largest variance in the entire forecast.

There is no current second source. Blue Origin’s New Glenn is operational but its reuse and cadence profiles are well behind Starship; China’s Long March 9 is paper; nobody else is building anything in the same payload class. So the architecture is single-vendor-dependent on a vehicle that has not yet completed its development arc. This is a real structural risk that the cislunar roadmaps generally underweight, because every roadmap was written assuming Starship works.

The state-actor response should be programmatic redundancy — funded development of a second heavy-lift class vehicle on a different architectural path, accepting that the second vehicle will cost two to three times the SpaceX number and ship four to six years later. Blue Origin is the natural counterparty in the US context; ESA’s Themis and any successor are the natural counterparty in Europe. China and India will develop their own regardless. The redundancy argument is straightforward — single-vendor dependency on critical infrastructure is the kind of thing the cislunar economy specifically cannot afford by 2035 — but it is currently not funded at the level the risk justifies.

The L1-depot funding anomaly

This is the central anomaly the roadmap analysis exposes. Every serious published cislunar architecture — Aerospace Corporation, NSS Milestone 14, the AIAA ASCEND infrastructure papers, the LunA-10 derivative work, and the Chinese Queqiao operational pattern — converges on L1 (or operationally near-equivalent halo orbits) as the depot, aggregation, and assembly node that makes cislunar logistics economic. No funded program of any space agency is currently building one.

The reasons are not technical. Propellant depots in stable halo orbits are not a research problem; they are an engineering integration problem on technologies that already exist. Cryogenic propellant transfer is being demonstrated under NASA Tipping Point programs. Long-duration zero-boiloff storage is hard but not novel; the constituent technologies are mature. Station-keeping at L1 is roughly ten metres per second of delta-v per year, which makes the lifecycle economics of a depot favourable compared to almost any other orbital infrastructure. The technical readiness exists. The funded program does not.

The reasons are institutional. NASA’s near-term funding is consumed by Artemis crew architecture (SLS, Orion, HLS and, since March 2026, a lunar surface base) plus science missions. For a decade the orbital budget line belonged to Gateway, which was planned for NRHO, not L1, partly because Orion’s delta-v capability is better matched to NRHO and partly because Gateway was designed as a crewed waypoint rather than a logistics depot. The Gateway-vs-L1-depot conflict was never a technical conflict; both could have existed, serving different functions. But within the NASA program structure, Gateway captured the budget line that would otherwise have funded L1 work, and when NASA paused Gateway in March 2026 the line went to the surface base, announced at about $20 billion over seven years, not to L1. The cancellation removed the competitor and left the anomaly intact. ESA is in a similar position: its contributions were committed to Gateway and are now to be repurposed, not redirected into parallel L1 infrastructure. China is building its own Lagrange-point infrastructure (the Queqiao series) but in the L2 region, optimised for far-side comms and lunar operations rather than for cislunar logistics generally. India’s lunar program is at an earlier stage and has not committed architecture this far out.

So the situation is that every published architecture identifies the same keystone, and no national program is funded to build it. This is the kind of coordination failure that defines pre-mature markets. The standard solution is either a multi-agency consortium with dedicated funding (the model that built CERN, ITER, the SKA) or commercial provision with anchor government contracts (the model that’s now building Starlink and crewed Earth-orbit capacity). Neither is currently underway for L1. The window for choosing the model is closing — by roughly 2030 the architectural decisions made on Artemis Phase III and the Chinese International Lunar Research Station will have hardened into expensive sunk costs, and an L1 depot will have to graft onto existing infrastructure rather than serving as the original aggregation point.

The single highest-leverage state action available in the 2026–2030 window is the announcement and funding of an L1 depot demonstrator program at agency budget scale, ideally as a multi-national consortium. Probability this happens on the timeline that matters: under thirty percent. The default outcome is that L1 emerges later, more expensively, and via commercial buildout once Layer 2 propellant economics force the question. That is workable; it is also a decade slower than the architecture would otherwise allow.

Autonomous ISRU as the gating milestone

The economic case for everything from Layer 2 onward rests on industrial-scale autonomous in-situ resource utilisation. Specifically: extraction, beneficiation, and processing of lunar regolith and polar ice at rates of at least hundreds of kilograms per day per system, sustained for periods of months without significant human supervision. As of mid-2026, the demonstrated state of the art is several orders of magnitude below this — sample-collection-scale operations, single-digit-kilogram throughputs, mission durations of days to weeks, with continuous teleoperation from Earth at 2.6 seconds round-trip latency.

The gap between current demonstrated capability and industrial-scale autonomous ISRU is roughly the same gap as exists between Boston Dynamics’ best 2024 humanoid demonstrations and a robot that could meaningfully replace a human warehouse worker. It is not the gap between current capability and a sci-fi hand-wave; it is the gap between research and industrial deployment, which historically takes ten to twenty years even when the funding is aggressive. For cislunar timelines that matter, the autonomy work has to start now and run continuously.

The technologies required are not exotic. Excavation in low gravity and high-abrasion dust environments; thermal management of extraction processes in vacuum; closed-loop chemical processing without Earth-shipped consumables; robotic maintenance and repair of the autonomy systems themselves. Each is an active research area. None has been integrated into a sustained autonomous operation at relevant scale. The integration is the problem, more than any single technology.

The funding pattern here is the inverse of the L1 problem. The L1 depot is technically ready and institutionally blocked. Autonomous ISRU is technically immature and institutionally well-funded relative to its readiness — NASA’s Lunar Surface Innovation Initiative, ESA’s PROSPECT, JAXA’s lunar resource work, the Chinese sample-return progression, plus commercial efforts from Astrobotic, Intuitive Machines, ispace, and a long tail of smaller actors. The funding is there. What’s missing is the integration milestone — the equivalent of NASA’s COTS program for ISRU, in which agency anchor contracts buy actual sustained autonomous extraction at meaningful scale from commercial providers, on a timeline aggressive enough to force the integration work to happen.

The right gating milestone, stated plainly: by 2032, at least one autonomous ISRU system operating on the lunar surface, producing usable propellant or oxygen or structural feedstock at rates above 100 kg/day, sustained for at least 90 consecutive days with no more than supervisory ground intervention. This is achievable. It is not currently a stated program goal of any national agency or commercial provider. It should be.

Cislunar SSA as the invisible backbone

Cislunar situational awareness — the ability to track, identify, and characterise every object operating between LEO and the lunar surface in real time — is currently fragmented across US Space Command, ESA Space Surveillance and Tracking, the commercial sector (LeoLabs, Slingshot, Numerica), and Chinese systems that don’t share data with the others. The fragmentation is acceptable while traffic is light. By the early 2030s, with mega-constellations deploying, lunar surface operations multiplying, and at least three sovereign actors operating cislunar hardware, it becomes a systemic risk.

A Kessler-cascade event in LEO — a debris collision generating enough secondary debris to trigger further collisions in a self-sustaining chain — is the canonical worst case. The probability is non-zero, the consequence is the loss of useful LEO for decades to centuries, and the second-order effect is that insurance markets close, financing dries up, and every cislunar program slips. The risk is currently underpriced because the insurance markets have not been forced to price it accurately, because the SSA data isn’t good enough to price it accurately. This is a recursion: the data quality determines the insurability, which determines the financing, which determines the buildout, which determines the traffic, which determines the data requirements. The recursion currently runs slowly. Once traffic crosses certain thresholds, it runs fast, and the system either resolves into a stable equilibrium with high-quality federated SSA or it cascades into a high-friction equilibrium with high insurance costs and slow buildout.

The action available now is the standards work — agreement on data formats, attribution rules, and federation protocols for cislunar SSA across the major operators. This is unglamorous, it is cheap, and it has the longest payoff window of any Layer 0 investment. Whoever drives the standards captures structural advantage for decades, in the same way that the actors who drove TCP/IP and SSL captured advantage in the internet era. The current state of cislunar SSA standards work is roughly equivalent to the state of internet protocol standards in 1980 — pre-foundational, contested, not yet locked in. The window for influence is open and closing. ISO, ITU, IADC, and the UN COPUOS Long-Term Sustainability working group are the venues. Almost nobody outside a small specialist community is paying attention. This is exactly the kind of leverage point that rewards early sustained engagement and is invisible to capital markets until after the standards have locked in.

Governance and the bifurcation question

The cislunar economy will either operate under a single international regime or split into two regimes along Artemis-Accords-versus-Sino-Russian lines. The choice is being made now, by which actors sign which agreements, which standards bodies they participate in, and which interoperability decisions they make on the first generation of hardware. The Artemis Accords have fifty-plus signatories as of mid-2026. China, Russia, and the International Lunar Research Station partner states have not signed and are not expected to. The two blocs are building parallel infrastructure with currently incompatible standards and divergent positions on the most consequential question — property rights in extracted resources.

The US 2015 SPACE Act, Luxembourg’s 2017 law, Japan’s 2021 law, and the UAE’s framework all assert that extracted off-Earth resources are private property of the extractor. Russia and China have not endorsed this view and have at various points argued that it conflicts with the 1967 Outer Space Treaty’s prohibition on national appropriation. Whether extraction is “appropriation” is the legal question on which the entire commercial cislunar economy rests. If extraction is appropriation, the Outer Space Treaty prohibits it, and the entire propellant-economy thesis loses its legal foundation. If extraction is not appropriation, the existing national frameworks suffice and the commercial buildout proceeds.

This is not going to be resolved by either bloc capitulating. The likely outcome is a parallel governance structure in which each bloc operates under its own resource-property framework, with the two systems mostly non-interoperable and occasionally in conflict. The historical analogue is the early Cold War period of competing technical standards — different broadcast standards, different rail gauges, different aviation regulations — eventually rationalised over decades into mostly-interoperable systems but with persistent legacy friction. Cislunar in 2050 is likely to look more like that than like a single unified regime. The implication for actors building toward this future is that they will need to choose which side of the bifurcation to participate in, and that choice carries durable consequences. European actors specifically have leverage here because they sit at the boundary and could plausibly maintain interoperability with both sides at the cost of slower commitment to either.

The state-level action available now is participation in the standards bodies and treaty negotiations at a level commensurate with the long-term stakes. The current level of engagement by most actors is well below that level. The space-policy community is small relative to the consequence of the decisions being made. This is another invisible-infrastructure leverage point.

What Layer 0 success looks like

By 2032, on the timeline that allows the rest of the architecture to proceed: Starship operating at sub-$1000/kg to LEO with weekly cadence, and at least one fully independent heavy-lift redundancy program funded; an L1 depot demonstrator under construction or operational, ideally as a multi-national consortium; at least one autonomous ISRU system on the lunar surface producing useful product at >100 kg/day with 90+ days of sustained operation; cislunar SSA federated across the major operators with agreed data standards and attribution protocols; a clear governance pathway, even if bifurcated, with property-rights frameworks operational on at least one side and being respected in practice. None of these are currently on track to the level Layer 0 success requires. All of them are technically achievable on the timeline. The gap is institutional, not technical, which means the leverage points are political and procedural rather than scientific.

The remainder of Layer 0 is the buildout of human and robotic presence on the lunar surface — Artemis crew landings, CLPS robotic missions, Chinese ILRS construction, ESA contributions, the long tail of commercial lunar landers. This is the visible work and it is mostly funded and mostly on track. The invisible foundations under it are not. The asymmetry between visible-and-funded versus invisible-and-underfunded is the central diagnostic finding of this section, and it generalises: the same pattern recurs in Layers 1 through 4, with the leverage always sitting in the infrastructure that doesn’t pitch well to whichever capital pool is in fashion at that layer.

5. Layer 1 — Propellant economy, 2030–2040

Layer 1 is where cislunar industry first produces something a non-government customer will pay for. The product is propellant. The customer is anything that needs to move in cislunar space and would rather buy fuel at a depot than carry it from Earth. The architecture is straightforward in outline and difficult in execution, and the single most important framing — usually missed in popular accounts — is that the depot is the market, not the regolith. Lunar water in a permanently shadowed crater is worth nothing. Lunar water electrolysed, cryogenically stored, and delivered to an L1 depot is worth whatever the next customer pays not to launch it from Earth. The economics of Layer 1 are the economics of that translation.

Lunar polar water as first export

The lunar polar ice deposits are the only resource within reasonable reach that simultaneously satisfies the three conditions Layer 1 requires: extractable with foreseeable technology, convertible to a product with established cislunar demand, and located at the bottom of a delta-v well shallow enough that the export economics close. Carbonaceous asteroids satisfy condition one and have superior chemistry; they fail condition three on operational cadence. Earth-launched water satisfies all three conditions but loses on cost once Layer 1 infrastructure exists. Mars satisfies none of them on the relevant timeline. Lunar polar ice is the answer by elimination, and every published architecture converges on this.

The conversion chain is electrolysis of water to hydrogen and oxygen, cryogenic liquefaction of both, and storage with minimal boil-off until delivered. The chemistry is undergraduate. The integration at industrial scale in the lunar polar environment is hard. Cryogenic systems in vacuum lose heat by radiation only, which is in principle helpful, but the polar environment introduces complications — periodic illumination of the polar peaks, deep cold of the permanently shadowed regions where the ice actually is, the tens-of-kilometres distance between the cold-storage regions and the power-generation peaks, dust contamination of every moving part. None of these is novel research. All of them are engineering problems that have not yet been solved together at production scale.

The throughput needed for the first paying market is modest. A single MEV-class satellite-servicing tug operating out of L1 needs on the order of two tonnes of propellant per servicing run. A handful of such tugs serving the GEO fleet at current servicing demand profiles needs perhaps fifty to a hundred tonnes per year of propellant at L1. The lunar surface throughput required is somewhat higher because of conversion and transport losses, but still well under a thousand tonnes per year of raw water extracted. This is a number that’s plausibly achievable with first-generation industrial ISRU once the autonomy and integration milestones from Section 4 are met. It is not a number that requires the kilometre-scale mining operations sometimes depicted in lunar-economy renderings.

The L1-customer-not-lunar-surface point

The mental model most popular accounts use — “lunar propellant for lunar missions” — is structurally wrong, and getting this right reorganises the rest of Layer 1.

A lander on the lunar surface that needs propellant to leave the surface mostly does not save by buying lunar propellant rather than bringing it from Earth, because the propellant savings are eaten by the integration overhead of the surface refuelling operation, the additional landed mass for fuel-transfer hardware, and the constraint that the lander has to choose a landing site near the propellant production facility rather than near its science or operational target. Lunar propellant sold to lunar surface customers is real but narrow.

The customer that actually justifies the propellant economy is the cislunar tug operating between L1 (or NRHO, or other halo regimes) and GEO, between L1 and lunar orbit, or between L1 and outbound trajectories to Mars, asteroids, and outer-system targets. Each of these customers wants propellant at L1, not on the lunar surface. So the lunar propellant operation is fundamentally an export operation, and the delivery cost from lunar surface to L1 is part of the all-in price.

That delta-v is roughly 2.5 km/s from lunar surface to L1, which is shallow enough to allow reusable shuttles. A purpose-built lunar-surface-to-L1 tanker, operating on lunar-derived propellant, paying back its dry mass over many cycles, can deliver propellant to L1 at a price determined mostly by the cost of producing propellant on the surface plus the amortised tanker capex. Detailed mass-balance studies — Aerospace Corporation, Charania and Olds 2006, the more recent NSS analyses — converge on figures somewhere between $1000/kg and $10,000/kg at the L1 depot, depending on assumptions. The range is wide because the assumptions are speculative; the central case is around $2000–4000/kg at L1.

This range matters because it sets the Layer 1 break-even against Earth-launched propellant. If Starship delivers $200/kg to LEO and $1000–2000/kg to L1 via existing chemical transfer stages, the lunar propellant case is marginal at best. If Starship is closer to $1000/kg to LEO and the L1 transfer is $3000–5000/kg, lunar propellant wins clearly. The entire Layer 1 economics hinges on this comparison, and the comparison hinges on Starship cost progression — which is exactly the dependency Section 4 identified as the single largest variance in the forecast. This is not a coincidence. The whole architecture is sensitive to the same underlying parameter, and serious Layer 1 planning has to model both numbers together rather than treating them as independent.

GEO satellite servicing as the commercial bridge

GEO servicing is the bridge between government-anchored Stage 0 and self-sustaining Stage 2 because it is the first cislunar activity with an unambiguously commercial customer, an unambiguously commercial revenue model, and a market large enough to support a self-sustaining service industry once unit economics close.

The market exists today. Northrop Grumman’s MEV-1 and MEV-2 have been operating attached to Intelsat satellites since 2020 and 2021 respectively. Astroscale, Northrop’s SpaceLogistics business unit, Orbit Fab, Starfish Space, and several others are at various stages of fielding refuelling, life-extension, and inspection services. The constraint on the market is propellant cost. Every existing servicing vehicle has to launch with its full propellant load from Earth, which limits how many servicing operations it can perform before its own dry mass becomes too high relative to remaining fuel to be worth continuing. The MEV class can extend a customer satellite’s life by five to seven years, after which the MEV itself becomes the limiting factor.

A servicing tug refuelled at L1 from lunar propellant has roughly an order of magnitude better economics, because its own dry mass amortises over many more operations and because the customer satellite can be moved to GEO graveyard and replaced with a new one rather than stretched at decreasing margins. The market this enables is plausibly $5–15B per year by 2040 in the central case — large enough to support several servicing providers, large enough to absorb significant Layer 1 propellant production, and small enough to fail entirely if the propellant unit economics don’t close. The market is also, importantly, the first cislunar activity for which the customer is a commercial entity rather than a government. This is the structural transition Layer 1 has to deliver.

The implementation gating technologies are in-space cryogenic propellant transfer (NASA Tipping Point programs working on this since 2020, not yet demonstrated at scale), standardised refuelling interfaces (partial standardisation under IDSS for crewed vehicles, almost no progress for cryogenic propellant interfaces), and long-duration zero-boil-off storage (active research, no commercial demonstrators). Each is solvable. None is solved. The pacing item is probably cryogenic transfer; a successful end-to-end demonstration of LH2/LOX transfer between two cooperative spacecraft, sustained for the operationally relevant duration, would change the timeline for Layer 1 meaningfully and is the kind of milestone that should be aggressively funded.

Asteroid water as alternative supply

The asteroid alternative to lunar polar water is real and improving, and the right framing is complement not substitute — different supply sources with different cost-versus-cadence profiles serving the same L1 customer.

A carbonaceous near-Earth asteroid in a favourable orbit can deliver water to L1 at delta-v costs comparable to lunar surface departure, sometimes lower. The water content of a C-type asteroid is often higher per unit mass than lunar regolith, the extraction chemistry is in some respects simpler (water can be driven off by heat alone for hydrated minerals, without the cryogenic-cold-trap operation lunar ice extraction requires), and there is no gravity well to climb out of. These advantages are real and persistent.

The disadvantages are operational cadence, autonomy duration, and uncertainty in the resource estimate per target. Mission durations of two to seven years for a single round trip; launch windows every several years per target; no ability to course-correct between missions on a tight feedback loop; resource composition estimates with order-of-magnitude uncertainty until you arrive. The Planetary Resources and Deep Space Industries generation of asteroid-mining ventures (2010–2018) all failed largely on the cadence-and-uncertainty problem rather than on technical infeasibility. The current generation — AstroForge, TransAstra, several smaller actors — has better launch costs and better autonomy than the 2010s generation but has not yet resolved the structural cadence problem.

The plausible Layer 1 trajectory is lunar propellant as the high-cadence steady supply, with asteroid water as supplemental supply once the operational pattern is established and once at least one or two C-type NEAs have been characterised well enough to support repeat extraction missions. The first asteroid-derived propellant delivery to an L1 depot is plausibly in the 2035–2042 window if a serious mission is in development by 2028 and adequately funded; later than the lunar-propellant first delivery, but in time to matter for Layer 1 maturity. The economic role of asteroid water is to provide supply elasticity — additional capacity when lunar propellant demand exceeds lunar surface production scaling, plus geographic diversification of supply (asteroid sources are less vulnerable to single-point failures on the lunar surface). It does not replace lunar propellant. It complements it.

A note on AstroForge specifically and the wider asteroid-mining commercial generation: the funding and engineering are real, the technical bets are reasonable, and the failure rate is going to be high because microgravity industrial processing remains an unsolved problem and sample-return is not the same problem. The right institutional posture is patience with multiple parallel attempts and acceptance that the first generation of asteroid-resource operations will mostly fail before the third or fourth generation finds a working pattern, in the same way the first three or four launch-vehicle startup generations mostly failed before SpaceX. This is normal for frontier industry and the funding model needs to reflect it.

The unresolved technical gates

Three gates separate the current state from operational Layer 1, in roughly the order they have to be cleared.

The first is sustained autonomous ISRU at the production scale identified in Section 4. Without this, there is no lunar propellant supply at all, regardless of customer demand. This is the long-pole item and the main reason Layer 1 maturity is 2035 at the earliest rather than 2030.

The second is in-space cryogenic propellant transfer at the operationally required duration and reliability. Without this, you can produce propellant on the lunar surface and you can produce customers at L1 but you cannot connect them economically. This is the gate that the current NASA Tipping Point work is meant to address, and it is unclear whether the pace of that work matches the pace Layer 1 requires.

The third is the standards-and-interfaces work that lets multiple producers, multiple operators, and multiple customers transact propellant as a commodity rather than as a series of bespoke bilateral deals. Propellant standards are well downstream of the technical gates but they determine whether Layer 1 looks like an open market or a vertically integrated company-store arrangement. The historical analogue is petroleum in the 1860s–1890s — a vertically integrated phase under Standard Oil before the antitrust and standards regime broke the integration and created an open market. Cislunar propellant will likely follow the same arc compressed into a decade, with the early entrants vertically integrating because there is no other way to bootstrap and the open-market structure emerging later under regulatory and antitrust pressure once the technical gates are cleared. The advocacy for early open standards is structurally similar to the standards arguments in Layer 0 — same kind of leverage, same kind of underfunding, same kind of payoff window.

The Starship-versus-lunar-propellant break-even

The bear case for Layer 1, rarely stated out loud by lunar-economy advocates because the same people are often Starship advocates, is straightforward arithmetic. If Starship reaches its most aggressive cost targets — sub-$200/kg to LEO with weekly cadence and reliable transfer-stage operations — the break-even price for lunar propellant at L1 falls to roughly $1000/kg or below, and at that price the lunar-propellant case becomes marginal against Earth-launched propellant. Lunar propellant still wins on supply diversification and on independence from Earth atmospheric constraints, but the unit-economics advantage that drives the rest of the architecture compresses substantially. The propellant economy still exists; it just delivers less of the cislunar bootstrap energy than the architecture assumes.

The honest framing is that Layer 1 is robust against a moderately good Starship outcome ($500–1500/kg to LEO) and fragile against an aggressively good Starship outcome (sub-$200/kg). The cislunar architecture loses some of its self-justifying economic logic if Earth-to-orbit launch is too cheap, because the case for any off-Earth production rests on launch costs being expensive enough to justify the alternative. Anyone forecasting Layer 1 has to forecast both numbers and be honest about the sensitivity. Most published analyses are not.

The more interesting case, and the one that becomes important in Layer 2 and beyond, is that propellant is the easy product to displace with cheap launch, and lunar industry’s resilience depends on moving up the value chain quickly — to oxygen for life support, to structural materials, to manufactured components — where the launch-cost-versus-production-cost arithmetic is more favourable to off-Earth supply. Layer 1 propellant is the door-opener for Layer 2 materials, not the endpoint. Operations and capital that close on propellant alone are vulnerable; operations and capital that use propellant to seed downstream production lines are robust.

Exporting momentum: the mass-driver complement

The propellant export carries two stacked uncertainties — the reserve, which Section 3 classifies honestly as inferred rather than proven, and the Starship break-even just described. There is a second lunar export architecture that carries neither, it is fifty years old, and its absence from the current discourse is another instance of the visibility bias Section 9 develops. O’Neill’s lunar mass driver (NASA SP-428, 1977; bench hardware demonstrated above a thousand g with Kolm at MIT by 1980) throws sintered regolith off the Moon electromagnetically. Nothing is heated or expelled; the delivery velocity is set by track length and current rather than exhaust chemistry; the energy is solar electricity, at roughly 2.9 MJ/kg for a slug thrown to lunar escape — about a fifth of the energy required merely to crack the same kilogram of silica into heavy, recondensing exhaust products. A dumb slug tolerates thousands of g, so the track is hundreds of metres rather than kilometres.

The deeper argument for the mass driver is the stock-versus-flow distinction the propellant thesis blurs. Polar water is a stock: finite, unmeasured, and simultaneously the future settlement’s drinking water, breathable oxygen, radiation shielding, and agriculture. Heinlein named the failure mode in 1966 — a colony that exports its own hydrosphere down a one-way gravity well is eating capital and booking the drawdown as income, and the same gravity asymmetry that makes the export profitable makes the refill impossible. Regolith is a flow surplus: effectively infinite, available equatorially with no prospecting, and carrying zero life-support option value. An export economy built on thrown regolith has no depletion date; one built on exported water has a depletion date whose distance depends entirely on a reserve we have not measured.

What kept the mass driver on paper for fifty years was never the launcher — it was the catch. O’Neill’s architecture required open-loop ballistic delivery to a passive catcher near L2, sixty-four thousand kilometres away, because 1977 had no guidance package that could ride a thousand-g projectile at any mass or cost that closed. Launch dispersion integrated unbounded over the transit, and the catcher had to be enormous and lucky. That single constraint is now the most thoroughly retired risk in spaceflight: cubesat-class autonomous GNC, optical and radar terminal tracking, propulsive terminal guidance, and routine autonomous rendezvous have all flown operationally. The modern architecture packages N tonnes of sintered regolith into a freighter round carrying one minimal guidance unit — a trim thruster, a transponder, a retroreflector — throws sub-escape (roughly 1.7–1.8 km/s) into a low lunar catch orbit rather than ballistic to L2, and catches with an active, manoeuvring catcher, with a patient solar-electric tug ferrying accumulated mass from lunar orbit to L1. That splits O’Neill’s one never-solved problem into two already-solved ones: a sub-escape throw to orbit, and routine autonomous rendezvous.

What this changes for Layer 1 is the exposure profile. Bulk lunar mass delivered to L1 — radiation shielding, structural feedstock, and oxygen via molten-regolith electrolysis at the node (oxygen is roughly 85 percent of a LOX/LH2 load by mass, and regolith is roughly 43 percent oxygen) — decouples from both of Layer 1’s stacked uncertainties at once. It does not depend on the polar reserve, because regolith is everywhere; and it is far less sensitive to the Starship break-even, because thrown mass at 2.9 MJ/kg of solar electricity competes on different arithmetic than electrolysed cryogens. Propellant remains the residual market for what genuinely needs stored chemical energy — crew transfers and high-thrust burns — which is a smaller market still riding the unmeasured reserve, but the architecture’s bulk-logistics layer no longer rides it with them.

The capability that gates this is the same capability that gates beamed power delivery into the permanently shadowed regions and depot rendezvous generally: deterministic delivery of mass or energy across cislunar distance. These are one problem in three costumes, and the problem is demonstrable now, at cubesat scale, in LEO — a guided freighter and an autonomous catcher proving closed-loop mass transfer across a meaningful gap, for the budget of a smallsat mission. Section 12 places this demonstrator on the leverage list, because it is cheap, currently unowned by any program, and retires the one risk that has kept the mass driver on paper since 1977.

One governance note belongs here rather than in a footnote. A system that can deterministically place tonnes of mass at a chosen cislunar point is, by construction, a precision kinetic delivery system; the catch problem and the targeting problem are the same problem, and adding guidance welds the export catapult and the bombardment catapult more tightly together, not less. Heinlein understood this too. A real lunar mass driver is a treaty question alongside an engineering one, and the governance work belongs in the same Layer 0 forums Section 4 identifies.

What Layer 1 success looks like

By 2040, on the timeline that allows Layer 2 to follow on schedule: an L1 depot operational and routinely supplied with lunar-derived propellant; at least one and ideally two or three competing servicing-tug operators using L1 propellant for commercial GEO servicing at scale; first asteroid-derived propellant delivery to L1, even if at demonstrator scale; in-space cryogenic propellant transfer demonstrated at operational reliability; cislunar propellant trading as a commodity with at least nascent open standards and multi-party interfaces; commercial revenue exceeding government anchor revenue for at least one cislunar operator. Probability all of these land by 2040: probably 20–30 percent. Probability at least the L1 depot, the first lunar propellant delivery, and the first commercial GEO servicing using L1 propellant land by 2040: closer to 50 percent, conditional on Layer 0 milestones being met. The whole sequence telescopes; if Layer 0 slips, Layer 1 slips by at least as much.

The transition from Layer 1 to Layer 2 is structurally different from the transition from Layer 0 to Layer 1. Layer 0 to Layer 1 requires new commercial revenue streams to emerge where none existed. Layer 1 to Layer 2 requires existing infrastructure to absorb additional product lines once the marginal cost of new lunar-derived production drops. The hard work is in the Layer 0 to Layer 1 transition; Layer 1 to Layer 2 is mostly a matter of capacity expansion and product diversification, which is the kind of growth pattern that commercial capital is much better at funding than the upfront infrastructure work. This is why Layer 1 is the structural pivot of the entire roadmap — get to Layer 1 cleanly and the rest of the buildout proceeds on its own momentum. Fail to reach Layer 1 and the whole architecture stalls at the government-anchored Stage 0 plateau indefinitely.

6. Layer 2 — Bulk materials and structural manufacturing, 2038–2050

Layer 2 is where the cislunar economy stops being a propellant business and becomes an industrial economy. The structural change is that the marginal cost of producing new lunar-derived products drops sharply once the propellant infrastructure exists — the power generation, the autonomous extraction systems, the surface-to-L1 transport, the depot operations, the workforce of robotic and human operators — these are all sunk costs of Layer 1 that Layer 2 amortises across additional product lines. The economic effect is a cascade rather than a series of independent industries. The same logic that put a propellant plant on the lunar surface puts an oxygen plant next to it, then an aluminium refinery, then a radiation-shielding sintering line, with each addition charging only the incremental capex against existing infrastructure. This is the layer where lunar industry first looks like industry rather than like a series of one-off government contracts.

It is also the layer where the orbital data centre thesis, in a form quite different from the current commercial pitch, finally becomes a real possibility — not in low Earth orbit, not in the 2030s, not with the supply chain currently being assembled by the Starcloud generation. The version that works comes online here, twenty years late by the standards of the current decks, and supplied by lunar industry that takes the propellant economy as its foundation. The chip-refresh problem governs why this is the case, and I develop it in detail in Section 8. For Layer 2 itself, what matters is that the physical infrastructure for large-scale orbital construction at the Lagrange points becomes plausible during this window, and the question of what to construct there becomes one of the central strategic questions of the 2040s.

The marginal-cost cascade

The standard mistake in lunar-economy modelling is to price each new product line as if it had to fund its own infrastructure. This is the framing used in most pitch decks, in part because each individual venture has to justify its own capital raise. But once the Layer 1 propellant operation exists, the relevant marginal costs for any next product are only the costs specific to that product — the additional extraction equipment, the additional processing chemistry, the additional product-specific quality control. Power, transport, communications, autonomy supervision, human supervision, life support overhead, surface-to-L1 logistics — all of these are shared with the propellant operation. The unit economics of Layer 2 products are dramatically better than Layer 1 economics would suggest, provided the Layer 1 operation exists at adequate scale to absorb the sharing.

This is the same pattern that played out in terrestrial industrialisation. The first railway in a country is uneconomic on its own; the second line that connects to it is dramatically cheaper per mile because the locomotive works, the signalling systems, the workforce, and the financing structure already exist. The first oil well requires a vertically integrated company; the second well in the same basin requires only a contract with the existing pipeline operator. The pattern in cislunar will be similar — Layer 1 is structurally hard because it has to bring all its own infrastructure; Layer 2 is structurally easier because it inherits Layer 1’s infrastructure as fixed cost. This is why the transition from Layer 1 to Layer 2 is mostly a question of capacity expansion and capital availability rather than a question of fundamental new technology, and why commercial capital is much better suited to funding Layer 2 expansion than to funding Layer 0 or Layer 1 foundations.

The implication for present-day investors is non-obvious. The right investment now, if you want to capture Layer 2 returns, is not in Layer 2 products — which won’t exist as fundable propositions until Layer 1 is operational — but in the Layer 1 operators who will own the shared infrastructure that Layer 2 products will run on. The actor that owns the lunar power grid, the surface-to-L1 transport, the autonomy supervision platform, captures the rent on every subsequent product line. This is the cislunar equivalent of owning the railroads in the 1850s rather than owning the manufacturers who used the railroads. Most commercial capital is currently looking at the wrong layer, and the venture-capital structure of seven-to-ten-year exit horizons makes the right bet structurally hard to fund.

Oxygen and structural aluminium as second-product lines

Two products follow propellant most naturally because they share the most infrastructure with the polar water operation. Oxygen for life support is essentially free — it’s the oxygen half of the water electrolysis that produces propellant, separated and stored rather than combined back for combustion. The production scale is small relative to propellant; a sustained crewed lunar presence of a hundred people consumes perhaps a few tonnes of oxygen per year for breathing, an order of magnitude less than the propellant operation produces. Oxygen as life support is more accurately framed as waste-stream value capture than as a separate product line, and its economic significance is mostly that it eliminates a cost-of-living item for lunar surface operations rather than that it generates external revenue.

Structural aluminium is different. Anorthite, the most abundant feldspar in lunar highland regolith, is roughly 19 percent aluminium by mass. The reduction chemistry is well understood on Earth — electrolytic reduction in molten salts is the same family of processes that produces terrestrial aluminium, with modifications for the lunar feedstock and the vacuum environment. The product is structural-grade aluminium suitable for radiation shielding, habitat construction, satellite buses, and the structural elements of orbital assemblies. The production volume that closes economically is significantly larger than the propellant operation — tens of tonnes per year at least, plausibly hundreds — and the customer base is mixed: surface infrastructure consumes some of it, but the bulk of the demand is at L1, for orbital construction that uses lunar aluminium rather than Earth-launched aluminium because the launch cost differential is large at structural mass volumes.

The chemistry has unresolved issues. Electrolytic reduction requires electrolyte management — the molten salts are reactive, the system has to maintain composition and temperature under lunar surface conditions, the byproducts have to be managed. Pilot-scale demonstrations exist on Earth (NASA Glenn, MIT, several university programs), but no industrial-scale lunar demonstration. The pacing item is roughly the same as for ISRU autonomy in Section 4 — the integration of multiple individually understood unit operations into a sustained autonomous production system in the lunar environment. The technology is ready in principle. The integration is not.

Once aluminium production is operational, the second-order products follow with marginal cost only — titanium from ilmenite, iron from regolith iron-bearing minerals, silicon for solar cells, glass and ceramics from sintered regolith for habitat shells and radiation shielding. None of these is technically novel; all of them depend on the autonomy and infrastructure foundations Layer 1 puts in place. The Layer 2 product portfolio is broad and economically diverse, which is part of why this layer is structurally more stable than Layer 1 — a single product failure in Layer 2 does not collapse the operation, because the shared infrastructure supports multiple parallel revenue streams.

Regolith-derived radiation shielding for orbital construction

The single Layer 2 product most underweighted in the popular discourse is bulk radiation shielding from sintered regolith. Radiation is the unsolved structural problem for any sustained human or sensitive-electronics presence beyond LEO. The current solution is to bring shielding mass from Earth, which is prohibitively expensive — galactic cosmic ray protection requires roughly twenty grams per square centimetre of shielding, which for a 50 cubic metre crewed module means several tonnes of shielding mass that has to either fly with the module from Earth or be added in orbit.

Sintered regolith is excellent shielding material. The aggregate density and atomic composition give good attenuation characteristics for both galactic cosmic rays and solar particle events; the sintering process is straightforward — concentrated solar or microwave heating of regolith bound with small amounts of binder produces structural ceramic suitable for shielding panels, habitat shells, or bulk fill. The energy budget is favourable because solar power on the lunar surface is essentially free; the autonomy budget is favourable because sintering operations are mechanically simpler than electrolytic reduction. The production rate that closes economically is in the hundreds of tonnes per year range, which is achievable with first-generation Layer 2 production capability.

The customer is orbital construction. Every L1-based station, every cislunar tug shielded for long-duration operation, every science platform at L2 or in deep space, every crewed module beyond the Van Allen belts, needs more shielding mass than is economic to launch from Earth. The market is not large in headcount terms — the customers are space agencies, defence customers, eventually commercial orbital construction operators — but the per-unit mass is high and the lunar-aluminium-and-regolith industrial base supports the production at low marginal cost. Radiation shielding is plausibly the second-largest Layer 2 product line by mass shipped after propellant, and its absence from most pitch decks reflects the same bias that underweights the L1 depot: it is infrastructure, not endpoint, and the value flows to whoever owns the production capacity rather than to a single visible end-user product.

The first asteroid metals to L1

The asteroid leg of Layer 2 follows the same complement-not-substitute pattern as the asteroid water in Layer 1, with one important difference: the relevant asteroid resource is not concentrated in C-type carbonaceous bodies but in M-type metallic bodies, which have different orbital populations, different access economics, and different processing chemistry.

M-type asteroids are dominated by iron-nickel metal with platinum group metal concentrations one to two orders of magnitude higher than terrestrial ores. The total resource available in the near-Earth M-type population is enormous — even a single hundred-metre M-type body has millions of tonnes of recoverable metal — but the technical challenge of extracting and processing that resource at industrial scale in microgravity remains unsolved. The Planetary Resources generation’s failure was in part a failure to recognise how hard microgravity industrial processing is; the current generation’s bets on this problem are reasonable but unproven.

The architecturally interesting question is not whether to process asteroid metal but where. Two schools: in-situ processing on the asteroid, shipping refined product back to L1; and capture-and-process at L1, redirecting the raw body or large chunks of it to a processing facility near the depot. In-situ processing minimises mass transported back but requires solving microgravity industrial processing at distance, with multi-year feedback loops between problem and solution. Capture-and-process at L1 moves the hard problem to “redirect a multi-thousand-tonne object precisely without it fragmenting in transit” — also unsolved, but unsolved in a different way that has tighter feedback loops and benefits from the L1 infrastructure.

My read is that capture-and-process at L1 wins for the first generation of asteroid metal operations because the processing problem is hard enough that operational cadence matters more than fuel cost, and L1 gives you ten thousand cadence improvements per year compared to direct in-situ work on the asteroid. The fuel cost of redirecting is real but not decisive; the fuel comes from the L1 depot, where lunar propellant is already amortised across other operations. The processing facility at L1 develops over multiple iterations rather than being committed up front, which is a better fit for a technology that has not yet been demonstrated at industrial scale. The first asteroid metal delivery to a paying customer at L1 is plausibly 2045–2050 on this architecture, with single-tonne demonstrator quantities earlier and meaningful production volumes later.

The customer base is the same as for lunar aluminium: orbital construction, satellite buses, radiation shielding, with additional niche markets in platinum-group catalysts for chemistry operations at L1 and on the lunar surface. The Earth-export market for asteroid PGMs is conceptually attractive but probably economically marginal even at very large scale, because the volumes that would justify the production would also collapse the terrestrial market price for the metals being produced. Asteroid metals in Layer 2 are best understood as inputs to cislunar industrial activity rather than as exports back to Earth, in the same pattern as lunar materials — the cislunar economy serves itself first and Earth only as a far-downstream possibility.

Where the orbital data centre thesis actually becomes plausible

The orbital data centre story currently being pitched is wrong about location, wrong about timing, and wrong about supply chain. It is correct that large-scale compute infrastructure eventually moves off Earth. It is wrong about almost every concrete detail of how this happens.

The correct version, in outline: orbital data centres of strategic significance — multi-gigawatt class, sustained over decades — are built at L1 in the 2040s and 2050s, using lunar-derived structure and radiation shielding, with chips shipped from Earth in routine cislunar logistics traffic, and refresh handled by scheduled module replacement using the same shipping infrastructure rather than by abandonment-and-relaunch. The energy comes from large-scale solar arrays built using lunar-derived photovoltaics. The cooling comes from large-scale radiators that take advantage of the L1 thermal environment and use lunar-derived bulk metal for radiator mass. The chip-refresh problem, which kills the current LEO orbital-DC thesis, is solved structurally at L1 because the logistics infrastructure for routine cislunar traffic exists by Layer 2 — chips can be refreshed every three years if needed, in the same logistics cadence that supplies propellant and structural materials.

This is a very different industry from what Starcloud and the current generation are building. The current generation is building LEO mega-constellations of compute satellites, which fail on the chip-refresh problem and on the thermal problem and on the bandwidth problem and on the structural problem. The correct version is building L1 stations that look more like ISS-scale infrastructure than like satellite constellations, supplied by lunar industry, refreshed on logistics cadence, scaled by adding modules rather than by launching constellations. None of this is happening now. The infrastructure required for it is exactly the Layer 0 through Layer 2 buildout this document describes. The orbital-DC thesis is correct about the eventual destination and wrong about the route. Nothing in current Earth-orbit hardware contributes meaningfully to the L1-based version that actually works.

The implication for Anthropic’s expressed interest in multi-gigawatt orbital compute via SpaceX is that the deal is honestly characterised as a long-horizon option on Layer 2 infrastructure that doesn’t yet exist, rather than as a near-term compute procurement. The 300 megawatts at Colossus 1 in Memphis is the real transaction. The orbital component is best understood as a thirty-year strategic position on whether the cislunar architecture this document describes actually materialises. That’s a reasonable position to take given Anthropic’s likely planning horizons, but it is structurally different from the way the deal is currently framed in public discourse, and the public framing is misleading about the actual timeline and the actual technology pathway.

What Layer 2 success looks like

By 2050, on the timeline that allows Layer 3 manufacturing for cislunar customers to follow: lunar industrial base producing oxygen, propellant, structural aluminium, sintered regolith shielding, and at least preliminary titanium and silicon at meaningful scale; L1 hosting a permanent multi-hundred-tonne depot-fab-staging complex with construction operations using lunar-derived structural materials; first asteroid-metal delivery to L1 in operational quantities (not just demonstrator scale); first L1 station built primarily from lunar materials rather than Earth-launched mass; cislunar industrial GDP — measured as value of product produced using lunar inputs and consumed at cislunar customers — crossing some defensible threshold like $10 billion per year. Probability all of these land by 2050: probably 15–25 percent. Probability the lunar industrial base and the L1 depot expansion are operational at meaningful scale by 2050: closer to 40–50 percent, conditional on Layer 1 success.

The transition from Layer 2 to Layer 3 is again structurally different from the prior transitions. Layer 2 is about bulk materials; Layer 3 is about manufactured components — satellite buses, orbital construction elements, eventually full satellites. The hard work in this transition is not the production technology, which mostly already exists in terrestrial form and adapts to cislunar conditions with engineering rather than research effort. The hard work is the autonomy of integrated manufacturing — the ability to assemble complex multi-component products in cislunar conditions without continuous human supervision. This is the gating problem for Layer 3 and the place where Layer 2 capacity becomes Layer 3 capability. The buildout of this autonomy starts in Layer 2 because the lunar industrial base provides the test environment; the deployment of it in operational Layer 3 production is mostly 2050+.

The structural-problems trilogy in Sections 8 through 10 — the chip-refresh problem, the bottleneck-inversion problem, and the architectural-displacement problem — is best read against the Layer 2 picture, because Layer 2 is where the structural problems either resolve or compound. The chip-refresh problem resolves at Layer 2 because cislunar logistics infrastructure exists to handle routine chip shipments. The bottleneck-inversion problem resolves at Layer 2 because the invisible-infrastructure leverage points from Layer 0 either paid off and Layer 2 functions, or didn’t and Layer 2 stalls. The architectural-displacement problem — the question of where in cislunar the high-value activity actually lands — resolves at Layer 2 because that’s the layer at which the geographic patterns become clear. Whether the cislunar economy becomes a network of distributed nodes or concentrates at one or two locations is decided in the Layer 2 buildout, and the choices made then shape the rest of the century.

7. Layer 3 — Cislunar manufacturing for cislunar customers, 2045–2060

Layer 3 is the first layer where the cislunar economy is structurally circular — where products manufactured in cislunar space are consumed by cislunar customers, with Earth’s role reduced to supplying the components that cislunar industry cannot yet make for itself, and to receiving services (orbital DC compute, satellite manufacturing capacity, deep-space mission staging) rather than supplying mass. The transition from Layer 2 to Layer 3 is the structural transition from industrial base to manufacturing economy. It is the layer at which cislunar industry stops being an extension of terrestrial industry and starts being its own thing.

The products are satellite buses, orbital construction elements, propellant tanks, life-support modules, scientific instruments, and increasingly the high-value manufactured goods that previously had to be launched from Earth at structural-mass premium. The customers are cislunar tugs, lunar surface bases, L1 and L2 stations, deep-space missions staging out of cislunar, and eventually Mars expeditions. The supply chain is mostly cislunar except for chips, specialty polymers, advanced organic chemistry, biological consumables, and a long tail of items that depend on terrestrial industrial complexity that cislunar cannot reproduce on a fifty-year horizon. Some of those items will eventually be made in cislunar; most will continue to be imported from Earth indefinitely. The boundary between “made here” and “imported” is not where the popular discourse expects it to be, and the implications are non-obvious.

The satellite-bus inflection point

The clearest Layer 3 commercial signal is the first satellite bus manufactured primarily from lunar-derived structural materials, with avionics and payload electronics shipped from Earth, integrated and launched from L1 to its operational orbit. This crosses a structural threshold the cislunar economy needs: it demonstrates that the lunar industrial base can produce a complex multi-component product to the precision and reliability that customers will accept.

The economics work because satellite buses are mass-dominated by structure, power, propulsion, and thermal — roughly 60–70 percent of dry mass for a typical comms or science satellite — and structural mass is exactly what lunar industry can produce most cheaply once Layer 2 is operational. The avionics, payload electronics, sensors, and antennas remain the high-value low-mass components that Earth still produces. The integration happens at L1, where lunar structural elements meet Earth-shipped electronics in a depot-fab facility staffed by humans and autonomous systems in roughly equal measure.

The first satellite bus of this kind plausibly flies in the late 2040s, with the integration capability building through the 2040s as Layer 2 capacity grows. The customer pattern is initially government — agencies will buy lunar-manufactured buses for cislunar science missions before commercial operators trust them for GEO comms — and shifts to commercial through the 2050s once the manufacturing record is established. The market opportunity is large but slow to mature: hundreds of millions to low single-digit billions per year by 2055, growing toward $10–20 billion per year by 2060 as commercial adoption expands and as the lunar-manufactured product range broadens to include life-support modules, propellant tanks, and orbital construction elements.

The strategic significance of the satellite-bus inflection is that it changes the global structure of the satellite industry. Once cislunar manufacturing exists at scale, the traditional satellite-manufacturing primes — Lockheed, Northrop, Airbus Defence and Space, the European and Japanese majors — face a structural choice between integrating cislunar capacity into their own supply chains or being eventually disintermediated by cislunar-based competitors. The same choice will face Chinese state primes against Chinese cislunar manufacturing. This is not a 2030 question. It is a 2050 question. But the strategic positioning for it is happening now in the form of which primes are participating in lunar industrial buildouts and which are not.

Orbital construction at scale

The second Layer 3 product line is large-scale orbital construction — assemblies too large to launch monolithically from Earth, built up from lunar-derived structural elements at L1 and other halo regions. The customers are operators of large orbital platforms — multi-gigawatt orbital data centres, kilometre-scale solar collectors, large astronomical telescopes (the natural successor to JWST is a 50-metre-class instrument that cannot launch as a single payload), large crewed habitats, and the infrastructure for sustained interplanetary mission staging.

The technical capability builds through Layer 2 — autonomous assembly of complex multi-component structures in space, robotic and human-supervised welding and joining operations, large-scale truss and panel assembly using lunar-derived aluminium and titanium. The pacing item is the autonomy stack rather than the production technology, because the production technology is largely terrestrial techniques adapted to cislunar conditions while the autonomy is novel. The first kilometre-class structure assembled at L1 from lunar-derived materials is plausibly mid-2050s. The economics of this are favourable specifically because the structures are too large to be Earth-launched competitively — there is no terrestrial alternative, so the cislunar price competes against not-doing-it rather than against an Earth-based supplier.

The most economically significant orbital-construction product is probably radiator surfaces for orbital data centres, large-scale solar collectors for power generation at L1 or L2, and habitat shells for crewed cislunar presence. These are mass-dominated products with low avionics content, exactly the profile that suits lunar industrial supply. By 2060, the cumulative mass of lunar-manufactured orbital construction at the Lagrange points plausibly exceeds the cumulative mass of Earth-launched structural mass in cislunar space — the moment at which cislunar industry produces more structural mass than it imports.

What stays on Earth indefinitely

The category of products that will continue to come from Earth indefinitely is larger than space-economy optimists like to admit and smaller than terrestrial-industrial protectionists assume. The category includes:

Frontier-generation semiconductors. Logic chips at the current node and one or two generations behind. The fab-supply-chain problem (Section 8) makes lunar logic fabs implausible on any timeline that matters; the relevant question is not when lunar fabs exist but how cheaply Earth chips can be shipped to L1 for integration with locally-manufactured hardware. Once Layer 2 logistics infrastructure exists, the answer is “cheaply enough that it doesn’t matter.”

Specialty polymers and organic chemistry. Kapton, Mylar, the family of films used in multi-layer insulation; specialty adhesives, sealants, lubricants; photoresists; the entire family of materials produced by terrestrial petrochemistry that depends on a deep supply chain of feedstocks and intermediates. Cislunar can produce many simple polymers from regolith-derived feedstocks eventually, but the high-performance specialty polymers will continue to be Earth-supplied because the economic value of replicating that supply chain off-Earth is low relative to the cost.

Biological consumables. Food, pharmaceuticals, medical supplies, life-science research materials, anything that depends on biological production processes. Cislunar will produce some food and possibly some pharmaceuticals on long timelines, but the bulk of biological consumables will come from Earth indefinitely. The economic argument here is not even close — terrestrial biological production capacity is so vast and so cheap that any cislunar replication is structurally uneconomic.

Specialty alloys and metals. The exotic alloys used in particular high-stress aerospace applications, certain rare-earth applications, the specialty steels used in cryogenic and high-temperature applications. Cislunar will produce many of these eventually, but the specialty metallurgy supply chain is deep and the volumes consumed in space are small enough that Earth supply continues to dominate.

Humans. The most economically significant Earth export to cislunar is people. A cislunar population of several thousand by 2060 is plausible; one of millions is not, on any reasonable timeline. Most of the humans in cislunar will continue to be Earth-natives spending time off-Earth and then returning, with a slowly growing population of long-residents and eventually a small population of cislunar-born. The biological and cultural dependence of cislunar on Earth remains durable.

The boundary between cislunar-produced and Earth-imported settles by Layer 3 into a stable pattern that holds through the rest of the century. Cislunar produces structural mass, propellant, oxygen, bulk metals, ceramics, basic chemicals, eventually basic polymers and basic agricultural products. Earth produces high-value low-mass items — chips, specialty chemicals, pharmaceuticals, advanced electronics — and continues to be the source of new humans. The trade is structural and durable, in the same way Iceland imports fruit and exports aluminium and that pattern is unlikely to invert.

The Mars complement

Mars enters the cislunar picture in Layer 3 not as a competitor but as a customer. Whatever Mars settlement looks like — whether it grows on the SpaceX-projected timeline of crewed missions in the 2030s, the more conservative agency timeline of crewed missions in the 2040s and sustained presence in the 2050s, or somewhere in between — it is supplied from cislunar more economically than from Earth surface for most product categories.

The delta-v from L1 to trans-Mars injection is roughly 0.8 km/s, dramatically lower than from Earth surface. The propellant for outbound Mars missions comes from the L1 depot. The structural mass for Mars-bound vehicles comes from lunar industry. The life-support consumables for the trip out come from a combination of Earth-supplied and L1-staged sources. Mars itself produces propellant from local resources once Mars-ISRU is operational, which is a parallel buildout on a slower timeline than the lunar buildout but with the same structural logic.

The interesting question is not whether Mars is supplied from cislunar but how much of the Mars supply chain becomes cislunar-resident. The plausible answer by 2060 is “most of the bulk supply chain, with chips and specialty materials still flying from Earth via cislunar staging.” Mars settlement of any meaningful scale rests on the cislunar infrastructure existing first. The popular framing in which Mars is the next destination after Earth-orbit, skipping cislunar, is structurally wrong. The right framing is that Mars is the second destination served by cislunar industry, with the lunar surface and L1 being the first.

Governance under load

Layer 3 is also when the governance arrangements established in Layer 0 face their first real load test. Cislunar property rights, propellant trading standards, debris attribution rules, satellite licensing across jurisdictions, taxation of cislunar industrial activity, dispute resolution between cislunar operators of different national affiliations — all of these become live operational issues in the 2050s as cislunar manufacturing matures.

The bifurcation question from Section 4 will have either resolved into a stable parallel-regime structure by Layer 3 or will be actively breaking down. The good outcome is roughly the Antarctic Treaty model — competing claims preserved in abeyance, scientific cooperation across blocs, commercial activity governed under each operator’s home jurisdiction with mutual recognition for cross-bloc transactions, durable peace. The bad outcome is more like the South China Sea — overlapping claims actively asserted, military presence shadowing commercial activity, repeated near-miss incidents that occasionally escalate. The cislunar version of either outcome is not predetermined; the choices made in Layer 0 and Layer 1 set the conditions but Layer 3 is where those conditions are tested.

The actor that emerges strongest from this period is the one that maintained interoperability and credibility with both sides during the bifurcation. European actors are well positioned for this role by geography and historical disposition; whether they take it depends on whether they fund the infrastructure to remain credibly capable at both sides of the boundary. Currently the European cislunar program is funded at a level well below that role. This is the single largest underweighted strategic position in cislunar economics, and it generalises the European-leverage point I made in Section 4.

What Layer 3 success looks like

By 2060, on the timeline that allows the rest of the century to proceed: cislunar manufacturing producing satellite buses, orbital construction elements, propellant tanks, and life-support modules at industrial scale; first kilometre-class structure assembled at L1 from lunar-derived materials; first multi-gigawatt orbital data centre operational at L1, with chips from Earth and structure from the Moon; cislunar industrial GDP measured in tens of billions per year; cislunar population in the low thousands across lunar surface, L1, L2, and lunar orbit; Mars supply chain operating primarily through cislunar staging; governance arrangements stable enough that cross-bloc commercial transactions are routine. Probability all of these by 2060: 10–20 percent. Probability cislunar manufacturing operational at meaningful scale and the L1 facility expanded toward Layer 3 capability: 30–40 percent.

The transition from Layer 3 to Layer 4 — the speculative pre-Layer-4 territory of helium-3, asteroid platinum-group metals for Earth markets, and the genuine industrial-scale exports back to Earth — is not predictable from the Layer 3 vantage. Some Layer 4 activities will be technologically viable by 2060 and economically marginal; others will be technologically viable and economically interesting; others will turn out to be permanent boondoggles. The discriminating variable is which Earth-market needs become acute enough to justify the cislunar supply cost. Frontier fusion fuel demand is the most-discussed but probably least-likely; mass-market Earth applications of platinum-group metals (catalytic converter equivalents for any future application) is more plausible; novelty exports for the high-end consumer market are real but small. The honest forecast for Layer 4 is “unclear and contingent on terrestrial market evolution that’s hard to predict thirty years out.”

What is clear is that Layer 4 is downstream of Layer 3, which is downstream of Layer 2, which is downstream of Layer 1, which is downstream of Layer 0. The buildout sequence is invariant; only the timing varies. Skipping a layer is not an option, because each layer’s economics depend on the prior layer’s infrastructure existing. The single largest mistake the current commercial discourse makes is to advertise Layer 3 or Layer 4 returns from Layer 0 investments. The capital allocated this way will mostly be lost, and the lessons learned will mostly be relearnings of lessons from the 2010s asteroid-mining wave. The actors who succeed will be the ones who invested in Layer 0 and Layer 1 with realistic expectations about when Layer 2 and Layer 3 returns become available.

The structural problems trilogy in Sections 8 through 10 develops the diagnostic frame for why current capital is allocated wrong, why the chip-refresh problem kills the LEO orbital-DC thesis on its own terms, and why the architectural displacement to L1 changes which actors and which infrastructure choices matter. Read against the Layer 0 through Layer 3 buildout described in Sections 4 through 7, the structural problems become the explanation for the gap between where the architecture says capital should flow and where it currently does.

8. The chip-refresh problem

The chip-refresh problem is the single structural argument that kills the current low-Earth-orbit data-centre thesis on its own terms, before any of the other structural problems are even considered. It is straightforward arithmetic, has no plausible workaround within the current architecture, and is essentially absent from the published commercial pitches. Stating it plainly is the most important diagnostic move this document makes.

The arithmetic

Frontier semiconductor performance per watt roughly doubles every two to three years on the current trajectory. The most aggressive industry projections assume Moore’s-law-equivalent performance scaling continues through at least 2035 and probably through 2040; the most pessimistic assume saturation by the early 2030s. The midpoint assumption — performance per watt doubles every three years — is conservative relative to current NVIDIA, AMD, and Intel trajectories and is the assumption this section uses.

Terrestrial data centres refresh their compute hardware on roughly a three-year cycle. The economic logic is straightforward: a chip that’s three years old delivers half the performance per watt of a current-generation chip, which means the operating cost per unit of useful computation roughly doubles compared to the new chip. At terrestrial energy costs and current chip prices, three years is roughly the point at which the operational savings of new hardware exceed the capital cost of replacement. Hyperscale operators (Google, Microsoft, Meta, Amazon) all run on this cycle. The economic logic is so robust that even sharp variations in chip prices, energy costs, or interest rates don’t move the optimal refresh interval by more than a year either way.

For LEO data centres, refresh requires either bringing the satellite back to Earth (impossible for any non-trivial mass at current launch economics), servicing it in orbit (technically possible but operationally expensive and dependent on infrastructure that doesn’t yet exist), or replacing the satellite entirely (which means the old satellite becomes debris or is deorbited, and a new satellite is launched). The third option is what the current orbital-DC pitches assume, often without stating it explicitly.

The arithmetic of replacement-as-refresh is brutal. A LEO data centre satellite at, say, 5 tonnes dry mass plus 5 tonnes of propellant for station-keeping over a three-year life — 10 tonnes total — costs roughly $10–30 million to launch at current Starship-projected costs and $50–100 million at current Falcon 9 costs. The chips inside the satellite cost on the order of $10–50 million depending on chip generation. The structural and avionics costs are perhaps $5–15 million more. Total per-satellite cost, $30–100 million, of which a substantial fraction has to be paid every three years for refresh. Compared to a terrestrial data centre that refreshes only the chips ($10–50 million) and keeps the building and infrastructure for thirty years, the LEO version pays its structural mass three to ten times over the same operational period.

The break-even comparison gets worse when energy and cooling are factored in. The terrestrial data centre pays for energy (significant but bounded) and cooling (significant in aggregate, often water-cooled at large scale). The LEO data centre gets energy free from solar collectors and rejects heat through radiative cooling — both of which scale poorly. Solar collectors big enough to power a multi-megawatt data centre are enormous; radiators big enough to reject the corresponding heat are kilometre-scale at multi-gigawatt levels. The thermal problem is solvable but the structural mass required is significant, which compounds the refresh cost.

Why workarounds don’t work

Three workarounds get proposed and none of them survive examination.

Workaround one: longer refresh cycles in orbit. The argument is that LEO chips can run for longer than terrestrial chips because the operational environment is more predictable. This is true and irrelevant. The constraint on refresh cycle is not chip reliability but performance economics. A six-year-old chip in orbit delivers a quarter of a current-generation chip’s performance per watt; the operational savings of new hardware are larger than ever. Longer refresh cycles don’t help; they make the problem worse because the gap between orbital and terrestrial performance widens.

Workaround two: orbital servicing for chip refresh. The argument is that a servicing tug delivers new chip modules to the LEO satellite, swaps them out, and returns. This is technically possible and economically interesting at GEO (where the satellite is high-value enough to justify it) but very hard at LEO because of the orbital geometry. LEO satellites are in inclined orbits that the tug has to match; the propellant cost of multi-satellite servicing missions at LEO is high; the operational tempo required (every three years, for hundreds or thousands of satellites in a mega-constellation) is well beyond current servicing demonstrations. Workable in principle, but the infrastructure required is itself a multi-decade buildout that doesn’t yet exist.

Workaround three: tolerate the performance gap because the LEO advantages outweigh it. The argument is that LEO data centres have other advantages — free cooling, free energy, no terrestrial siting constraints, no terrestrial energy policy risk — that justify operating older-generation hardware. This is the strongest of the three workarounds and it still doesn’t survive scrutiny. The performance gap between current-generation and three-generations-old hardware is a factor of eight on performance per watt; the LEO advantages are real but bounded, and they shrink rapidly as performance per watt becomes the dominant cost driver for frontier compute. At three generations behind, the LEO data centre is competing against terrestrial data centres that are eight times more energy-efficient per unit of computation, and the terrestrial energy savings dwarf any LEO operational savings.

There is no workaround that survives examination on LEO architecture. The problem is structural: LEO is the wrong place for compute hardware that depreciates rapidly relative to chip-generation cadence.

What the problem reveals about the current pitches

The Starcloud business model, the Axiom ODC nodes, the NVIDIA Space-1 Vera Rubin module, the SpaceX cislunar-DC concepts as currently publicly discussed — all of these are built around an architecture that the chip-refresh problem rules out. The companies’ technical staff almost certainly understand the arithmetic. The fact that the public pitches don’t address it directly is evidence about what kind of capital they are raising rather than evidence about what they actually believe the long-term architecture will be.

Some of these companies are positioning for a longer-term play that does survive the chip-refresh problem, and the current LEO hardware is a demonstrator and a market-positioning move rather than the actual long-term business. Kepler’s edge-inference architecture, Sophia’s distributed inference plan, Starcloud’s longer-term L1 ambitions (if they exist), and NVIDIA’s broader cislunar partnership strategy could all be read this way. The companies that fold in the early 2030s will be the ones that took the LEO-DC pitch literally. The companies that succeed will be the ones that used LEO hardware as a stepping stone toward Layer 2 and Layer 3 architectures.

This is consistent with the earlier point about Starcloud’s announcement of Bitcoin mining ASICs on Starcloud-2. Bitcoin mining has different chip-refresh economics than frontier AI training — Bitcoin ASICs lose performance per watt more slowly than AI accelerators, and the operational cost of mining is more sensitive to energy cost than to chip generation. A LEO Bitcoin operation is closer to making economic sense than a LEO AI training operation, because the chip-refresh problem is less severe. The fact that Starcloud is hedging its second satellite toward Bitcoin is consistent with the company’s technical staff understanding the chip-refresh problem perfectly well and looking for workloads that survive it.

Why the L1 architecture survives the chip-refresh problem

The orbital-data-centre thesis as it actually works in the 2040s and 2050s, at L1 rather than LEO, survives the chip-refresh problem for structural reasons specific to the Layer 2 buildout.

First, cislunar logistics infrastructure exists by Layer 2. Routine transit between Earth and L1 happens on a frequency comparable to commercial aviation or maritime shipping by the late 2040s. Sending a module of new chips to L1 is not a bespoke mission; it is a manifest line item on a regular cargo flight. The marginal cost of chip refresh at L1 is the cost of the chips plus a small share of the routine logistics flight, comparable in scale to terrestrial chip-delivery costs.

Second, the data centre at L1 is built as modular components, not as a monolithic satellite. Refresh means swapping modules, not replacing the entire facility. The structural mass of the data centre — radiators, solar collectors, habitat space for human operators, basic power distribution — is lunar-derived and has a thirty-to-fifty-year operational life. Only the compute modules refresh on three-year cycles, and those are designed for routine swap-in-swap-out operations.

Third, the energy economics are durable. L1 has effectively constant solar illumination, scaled by lunar-manufactured solar arrays that can be expanded indefinitely. The terrestrial energy-supply concerns that motivate orbital data centres in the first place (grid constraints, siting limits, geopolitical risk to energy supplies) are persistent enough that the L1 alternative becomes structurally compelling once the supporting infrastructure exists, even in scenarios where Earth-launched LEO data centres remain uneconomic.

So the orbital data centre thesis is correct in outcome and wrong in implementation. The thesis says compute moves off Earth. The current implementation puts compute in LEO where it can’t be refreshed. The correct implementation puts compute at L1 where it can be refreshed economically, but only after the rest of the Layer 2 infrastructure exists. The current capital flowing into LEO orbital data centres is, in the optimistic interpretation, paying for stepping-stone hardware that develops the engineering capability to eventually build the L1 version. In the pessimistic interpretation, it’s paying for hardware that the chip-refresh problem destroys as a business within five to ten years.

Both interpretations agree on one thing: the LEO orbital-DC industry of the late 2020s and early 2030s does not become the orbital-DC industry of the 2050s. The eventual L1 industry is supplied by lunar industrial infrastructure that the current generation is not building, and the current generation’s hardware does not transfer to that architecture in any meaningful way. The investors who are positioning for the long-term orbital-compute future via the current LEO investments are essentially betting that the LEO generation provides intangible learnings (operational experience, regulatory positioning, brand recognition) that translate to the L1 generation. That bet is not obviously wrong, but it is also not the way the investments are pitched, and the gap between the pitch and the actual long-term thesis is large enough to matter for capital allocation decisions.

9. The bottleneck-inversion problem

The bottleneck-inversion problem is the diagnostic explanation for why current capital flows are misallocated even when the technical case is sound. It generalises the diagnostic finding from Section 4 — that visible-and-funded versus invisible-and-underfunded is the central asymmetry of Layer 0 — into a general claim about how frontier industries get capitalised and where the structural mismatches occur.

The pattern

In every layer of the cislunar buildout, capital flows toward the visible hardware (rockets, satellites, landers, the next dramatic launch event) and away from the invisible infrastructure (depots, autonomy stacks, situational-awareness federation, standards work, governance institutions). The visible hardware is funded too generously relative to its option value; the invisible infrastructure is funded too sparsely relative to its option value. The bottleneck on the architecture’s progress is in the invisible layer, but the capital flows mostly to the visible layer.

This is not a mistake by individual capital allocators. It is a structural property of how the industry pitches itself, how the press covers it, how the public engages with it, and how funding decisions are made within agencies and venture firms. The visible hardware has a single-sentence story: “we launched a thing.” The invisible infrastructure has a longer story that doesn’t fit the same coverage formats. The hardware has discrete events that align with funding milestones; the infrastructure is continuous work that doesn’t generate the same press flow. The hardware can be photographed; the infrastructure cannot. The hardware fails dramatically and visibly when it fails; the infrastructure fails silently, in the form of futures that don’t happen because the infrastructure wasn’t there when needed.

This pattern is not new. The same dynamic shaped the buildout of every major infrastructure category in industrial history. The railroad boom of the 1870s overfunded locomotives and rails and underfunded signalling and dispatch coordination; the resulting accident rates were brutal until standards and federations emerged. The early aviation industry overfunded aircraft and underfunded navigation, traffic control, and weather services; the standardisation of those services was decades behind the aircraft. The early internet overfunded routers and switches and underfunded the standards and governance that eventually made the network universal; ICANN’s contested institutional arrangements still reflect that gap.

Cislunar is repeating the same pattern at a larger scale and with longer feedback loops. The over-funding of LEO mega-constellations and the under-funding of cislunar SSA federation are the most obvious examples. The over-funding of lunar landers and the under-funding of autonomous ISRU integration is another. The over-funding of GEO satellite replacement and the under-funding of in-space cryogenic propellant transfer is a third. The pattern recurs at every layer.

Why state actors are the only realistic counter

The pattern is hard to correct from the venture-capital side because the invisible infrastructure has the wrong return profile for VC. VC wants seven-to-ten-year exits with skewed return distributions; the infrastructure that gates Layer 0 through Layer 2 buildouts pays off in the 2040s and 2050s and has more even return distributions because it captures economic rent rather than generating breakthrough valuations. The infrastructure is excellent business but bad VC, in the sense that the unit economics will make money but the math doesn’t work for a fund with a ten-year timeline.

This means state actors are the only realistic counter to the bottleneck inversion. State actors have the time horizons, the public-goods rationale, and the capital to fund infrastructure that pays off over thirty-year periods. The historical precedent is clear — interstate highways, the air traffic control system, the underlying internet infrastructure, the deep-water port system that handles modern maritime trade. All of these were state investments in infrastructure that markets couldn’t fund on the relevant timeline. Cislunar will need the same kind of state investment, and the actors who provide it will capture structural advantage for decades.

The state actors that emerge strongest from the Layer 0 through Layer 2 period will be the ones that understood the bottleneck inversion early and funded the invisible infrastructure aggressively. The actors that funded only the visible hardware will end up with rockets and landers but no economic ecosystem to use them, in roughly the same way that Soviet-era Antarctic infrastructure ended up with stations and equipment but no integrated economic function. Visible hardware without infrastructure is theatre. Infrastructure without visible hardware is invisible. The combination is what builds frontier economies. Currently no state actor is doing the combination well; the actors that come closest are doing one half or the other.

The US is heavy on visible hardware (Artemis, SLS, Orion, Starship contracts, the CLPS commercial lander programme) and light on invisible infrastructure (no L1 depot, fragmented SSA, limited standards leadership, low autonomy-integration funding). China is heavy on integrated state planning, including some invisible infrastructure (Queqiao constellation), but with less commercial integration and less depot-style logistics. Europe is light on both, with strong scientific contributions but limited capacity in either category. India is at an earlier stage. Japan is concentrated on autonomy and contributing components. The geopolitical winner is whichever bloc figures out the combination first; currently no bloc is, which is part of why the cislunar architecture is slipping behind the timeline the geography would allow.

The role of cislunar policy entrepreneurship

Within the state-actor space, there is meaningful leverage for individuals and small institutions that understand the bottleneck inversion and advocate for invisible-infrastructure funding inside agency and legislative processes. The cislunar-policy community is small relative to the consequence of its decisions; a handful of well-placed analysts and advocates can shape the trajectory of agency program plans and congressional appropriations to a degree that is genuinely consequential.

This is not unique to cislunar; the same dynamic shapes most areas of state-funded infrastructure. The Interstate Highway System exists because of a coalition of policy entrepreneurs who understood the strategic case (Eisenhower, the trucking industry, the petroleum industry, the Defence Department, urban planners) and assembled the political coalition that funded it. The internet exists because of a much smaller coalition of researchers and DARPA program managers who understood the strategic case for packet switching and persuaded the agencies to fund it. The deep-water container ports exist because of an even smaller coalition of maritime industry analysts and port authorities who understood the standardisation case and pushed the necessary investments. In each case, the consequential policy work was done by a few dozen to a few hundred people.

The cislunar equivalent of those coalitions is being assembled now, mostly in the form of standards-body participation, agency working groups, and academic-policy interactions that happen below the threshold of public attention. The actors in those forums shape the cislunar architecture more than any individual launch event. The leverage is asymmetric — a small number of well-placed individuals can move the trajectory by years on critical infrastructure decisions — and the leverage is currently underweighted by capital and attention. This is the deepest version of the bottleneck-inversion finding: the leverage on the future is concentrated in places where almost no public attention or capital allocation is happening.

For the readers of this document who are in a position to influence cislunar policy — agency staff, congressional staff, science advisers, philanthropic funders of policy infrastructure — the highest-leverage action is participation in the standards and governance work that the rest of the world is currently underweighting. ISO TC20/SC14 on space systems and operations, the Inter-Agency Space Debris Coordination Committee, UN COPUOS’s Long-Term Sustainability working group, the various national bodies setting cislunar SSA standards, and the bilateral and multilateral agency forums where Artemis Accords and ILRS interoperability questions are being decided — these are where the cislunar architecture is actually being shaped. They are profoundly under-attended relative to their consequence. This is exactly the kind of asymmetry that rewards informed and sustained engagement.

10. The architectural displacement problem

The architectural displacement problem is the geographic version of the chip-refresh problem and the bottleneck-inversion problem. It is the structural claim that the high-value cislunar activity does not happen where the current commercial and policy attention is focused — not in low Earth orbit, not on the lunar surface as a destination unto itself, not in geostationary orbit — but at the Earth-Moon Lagrange points, particularly L1, in a buildout that is twenty years behind the timeline the current discourse implies.

Where the value actually lives

The economically interesting cislunar activity of the 2040s through the 2070s — the activity that generates the largest revenue, that creates the most durable strategic positions, and that absorbs the most structural mass and capital — happens at L1. The lunar surface is the population, manufacturing, and resource base; L1 is the market, the assembly yard, and the gateway to the rest of cislunar space. The relationship between them is analogous to the relationship between a mining region and the port city that serves it. The mining region produces the goods; the port city captures the value of moving the goods to market, financing the trade, and providing the services that the mining region eventually depends on. Historically, port cities have often outgrown the mining regions that initially justified their existence. Cislunar may follow the same pattern, with L1-based industry eventually becoming larger and more economically diverse than lunar surface industry.

This is the architectural displacement: the high-value activity displaces from the obvious destination (the lunar surface, which is where the resources are) to the connecting node (L1, which is where the resources meet the customers). The displacement is not a 2026 phenomenon — currently L1 has no infrastructure to speak of — but it is implied by the architecture and it becomes visible in the 2040s as Layer 2 matures.

The implication for present-day actors is that capital and attention currently allocated toward the lunar surface as the primary destination are partly misallocated. The lunar surface is important; it is necessary; it is unavoidable. But it is not where the value will concentrate. The Layer 1 and Layer 2 actors who own L1 infrastructure capture more economic value over the long term than the actors who own only lunar surface infrastructure, in the same way that owning a major port has historically been more lucrative than owning the mines that the port serves.

Why the displacement is not currently visible

The architectural displacement is not visible in current discourse for several reasons that are worth naming.

First, L1 does not yet have anything to look at. The Lunar Gateway, paused in March 2026, was bound for NRHO, not L1. No spacecraft is currently operating at Earth-Moon L1 (the Earth-Sun L1 is occupied by science missions like SOHO, but that is a different node). The popular and policy discourse uses lunar surface and lunar orbit images that are easy to render and easy to photograph; L1 is invisible and abstract. This is the same image-vs-infrastructure dynamic that drives the bottleneck inversion.

Second, the Artemis program is structured around lunar surface as the destination, and Artemis is the largest funded cislunar program. Artemis defines the surrounding discourse in the United States and among Artemis Accords signatories. The program’s framing — humans on the lunar surface, sustained presence at the south pole, eventual Mars missions — is correct as a program but incomplete as an architecture. The L1 layer is missing from the Artemis narrative because it is not in the Artemis program plan.

Third, the Chinese program is structured around lunar surface and L2-vicinity infrastructure, not L1. The Queqiao constellation, the planned International Lunar Research Station, and the Chang’e sample-return progression all happen at the lunar surface or in lunar/L2 orbit. China’s L1 strategy is essentially absent from public announcements, which may reflect strategic non-disclosure rather than absence of planning, but in either case L1 is not the visible focus.

Fourth, commercial actors pitch the visible destinations — landers, lunar surface operations, satellite servicing in GEO — because those have current customers and current revenue paths. L1 has no customers now because there is nothing at L1 to be a customer of. The infrastructure precedes the customer, which is a structural problem for commercial pitching. The actors who eventually capture L1 value are positioning for it now, but they are not pitching it publicly because the pitch does not close on a venture timeline.

The combination of these factors means L1 is structurally underweighted in every dimension — agency program planning, commercial capital allocation, press coverage, public engagement, academic research, policy advocacy. The underweighting is the architectural-displacement problem in its operational form.

What displacement implies for strategic positioning

For state actors, the architectural-displacement implication is that L1 infrastructure investment now captures structural advantage for decades. A serious L1 depot demonstrator program funded in the 2026–2030 window, even if it does not complete until the mid-2030s, positions its sponsoring agency or consortium as the cislunar logistics anchor for the rest of the century. The same investment made in 2040, after lunar industrial activity has forced the question, costs more, delivers later, and captures less strategic position because the architecture is already partly built and the entrant has to graft onto existing infrastructure.

For commercial actors, the implication is that L1 services — depot operations, propellant trading, assembly services, eventually orbital construction and orbital data centre operations — are the long-term high-value businesses. The companies positioning for them now via Layer 1 propellant infrastructure investments are making the right bet even if their current LEO operations don’t survive the chip-refresh problem. The companies whose bet is purely LEO without any L1 strategy are missing the architecture.

For European actors specifically, L1 is the leverage point that aligns most cleanly with European strategic preferences. L1 infrastructure is dual-use (commercial and scientific), it requires multi-national consortium funding (which Europe excels at organising), it benefits from interoperability across the Artemis-versus-Sino-Russian bifurcation (which Europe is positioned to maintain), and it has long time horizons that suit European industrial-policy frameworks better than they suit American or Chinese state planning. The European cislunar strategy that would actually have leverage is heavy on L1 and light on competing with the United States or China on Artemis-style lunar surface programs. The current European program is not structured this way, and the gap is the single largest underweighted strategic position in global cislunar policy.

For deep-space mission planners, the implication is that L1 is the natural staging point for Mars and outer-system missions, with consequent cost reductions and operational flexibility for any program that uses cislunar staging rather than direct Earth-departure trajectories. NASA’s deep-space architectures should integrate L1 staging assumptions in the mid-2040s and beyond; ESA’s analogous frameworks should integrate it earlier and more aggressively because L1 sits at the European strategic sweet spot.

The synthesis of the three structural problems

The three structural problems integrate cleanly. The chip-refresh problem says LEO is the wrong place for orbital data centres. The bottleneck-inversion problem says invisible infrastructure is where the leverage is. The architectural-displacement problem says L1 is where the high-value activity actually lives.

Read together, the three problems imply a specific strategic position: the actor that funds invisible Layer 0 infrastructure now, particularly L1 depot and cislunar SSA federation, captures structural advantage in the eventual L1-based industry of the 2040s and beyond, including the L1 orbital data centre industry that the LEO Starcloud generation is currently failing to build correctly. The architecture says capital should flow toward L1 infrastructure with a Layer 0 horizon. The current discourse says capital should flow toward LEO satellites with a 2030s horizon. The architecture is right and the discourse is wrong, by a margin large enough that the consequences will be visible in the next decade as the LEO investments fail to deliver on their pitches and the L1 architecture either emerges from quiet state and consortium funding or fails to emerge and the whole cislunar timeline slips by another decade.

The diagnostic finding is sharp. The corrective action is clear. The question is whether any actor — state or commercial — moves first to fund the L1 infrastructure that the architecture requires. The probability of this happening on the timeline that matters is the single most important variable in the entire cislunar forecast, and it is currently tracking at below thirty percent.

11. The gating questions

The forecast in this document depends on the resolution of a small number of gating questions whose answers will be visible by the early 2030s. Each is binary or near-binary in its consequence — a yes answer puts the architecture on the timeline this document describes; a no answer pushes the timeline out by a decade or more per failure. The questions are not equally weighted; the first two carry most of the variance.

Question 1: Does Starship reach sustained $500–1500/kg with weekly cadence by 2030?

The single largest variance in the forecast. A yes answer puts every other timeline in range. A no answer — meaning Starship stalls at higher costs, lower cadence, or both — pushes Layer 0 through Layer 2 out by ten to fifteen years and changes which actors can participate. The signal to watch is not stated prices but achieved cadence-times-reuse-turn-time, which becomes legible from launch data in the 2027–2029 window.

Question 2: Does autonomous ISRU work at >100 kg/day, 90 days, by 2032?

The technical gate for everything from Layer 2 onward. A yes answer means Layer 1 propellant economy is on track for 2035–2040 maturity. A no answer means Layer 1 slips into the 2040s and Layer 2 into the 2050s, with cascading consequences for the rest of the architecture. The signal to watch is integration-scale demonstrations rather than component-scale research progress; the gap between current state and the milestone is closeable but not on autopilot.

Question 3: Does an L1 depot get funded and built by the mid-2030s?

The keystone infrastructure question. A yes answer means the cislunar economy operates as a network with a hub from the start. A no answer means cislunar emerges later as a collection of point-to-point operations that eventually retrofit to an L1 architecture at higher cost. The signal to watch is multi-agency consortium announcements or commercial anchor-tenant contracts in the 2027–2030 window. Currently no signal is visible; default outcome is a no answer.

Question 4: Do property rights in extracted materials get internationalised, or bifurcated?

The governance question that determines whether cislunar is one economy or two from 2040 onward. A yes-internationalisation outcome is unlikely; the realistic best case is stable bifurcation with mutual recognition for cross-bloc transactions. The signal to watch is whether Artemis Accords expand to include Chinese or Russian participation (essentially impossible on current trajectories) or whether parallel ILRS-side frameworks emerge that interoperate with Accords-side frameworks at the technical and commercial levels (more plausible). The signal becomes legible by the late 2020s.

Question 5: Does microgravity industrial processing get demonstrated at meaningful scale?

The gate for whether asteroids participate in Layer 2 or only in Layer 3 and beyond. A yes answer expands the resource base and the supply elasticity of the Layer 2 economy. A no answer keeps asteroids as a Layer 3-plus speculative resource and concentrates Layer 2 production on the lunar surface. The signal is AstroForge or comparable mission outcomes in the 2028–2032 window; the failure rate will be high and the discriminating evidence will require several missions.

Question 6: Can the chip-refresh problem be solved without lunar fabs?

The question that determines whether orbital data centres at L1 in the 2040s and 2050s rely on lunar chip manufacturing (which the supply-chain ecosystem problem makes implausible on that timeline) or on Earth-shipped chips via routine cislunar logistics (which is the working assumption of this document). The signal is the cost and cadence of Earth-to-L1 logistics by the late 2030s. If routine cargo transit costs and timelines stabilise at the low end of plausible ranges, the answer is yes and the orbital-DC architecture works. If cislunar logistics remains expensive and slow, the chip-refresh problem partly recurs at L1 and the orbital-DC industry develops more slowly than this document forecasts.

12. Where to put the leverage

If the diagnostic analysis is right, the prescriptive implications are clear. For state actors, commercial actors, and policy entrepreneurs operating in the cislunar space, the leverage is concentrated in places that do not match the current distribution of capital and attention.

For state actors, in priority order:

First, fund the L1 depot. The single highest-leverage state investment available in the 2026–2030 window. Multi-agency consortium model preferred; bilateral or trilateral fallback acceptable. Budget scale comparable to a moderate science mission, payoff window thirty years, strategic position for the rest of the century. No current program. The first space agency or consortium to commit captures structural advantage for decades.

Second, fund autonomous ISRU integration to the 100 kg/day, 90 days milestone. COTS-style anchor contracts with commercial providers, on aggressive timelines, with explicit integration milestones rather than research-component milestones. NASA is well positioned for this; ESA and JAXA can participate; China is on a parallel track via its own programs. The technology is ready in principle. The integration funding is what’s missing.

Third, fund cislunar SSA federation and standards work. The unglamorous longest-payoff leverage point. Money goes to standards bodies, federated data infrastructure, and the policy-staff capacity to drive these forums. Single-digit millions per year per major agency, payoff window fifty years.

Fourth, fund a closed-loop cislunar delivery-and-catch demonstrator. A guided freighter and an autonomous catcher at cubesat scale, in LEO and then lunar orbit, demonstrating deterministic mass transfer across a meaningful gap. Smallsat-mission budget, no current owner, and it retires the single risk that has kept the lunar mass driver on paper since 1977 — the same capability that gates beamed power delivery and depot logistics (Section 5).

Fifth, fund a second heavy-lift redundancy program. Programmatic insurance against single-vendor Starship dependence. Higher per-flight cost than Starship, but the redundancy value is real enough to justify the program even if the second vehicle never reaches cost parity. Blue Origin is the natural counterparty in the US context; Themis-successor in Europe; nationally-funded programs in China and India will emerge regardless.

Sixth and least visible — participation in international governance forums. ISO, ITU, IADC, UN COPUOS, bilateral agency working groups. Headcount and attention rather than capital. The actor that staffs these forums credibly shapes the architecture; the actor that does not, doesn’t.

For commercial actors, in priority order:

First, invest in Layer 1 infrastructure operators rather than Layer 2 or Layer 3 products. The actor that owns the lunar power grid, the surface-to-L1 transport, or the L1 depot operations captures rent on every subsequent product. The right comparison is owning railroads in the 1850s rather than the manufacturers who used them.

Second, structure capital for thirty-year payoff horizons, accepting that this rules out conventional VC and requires patient capital from sovereign wealth, family offices, very long-horizon corporate investors, or government anchor contracts. The infrastructure required for Layer 0 and Layer 1 is excellent business and bad VC.

Third, position for L1 services even when L1 has no current customers. The companies that have actively-stated L1 strategies in 2026 are the companies most likely to dominate L1 services in 2040. Currently almost none do.

For policy entrepreneurs and analysts, the leverage is concentrated in shaping the agency-level program decisions, the standards-body participation, the academic-policy interactions, and the public-discourse framing that influence where state capital eventually flows. The cislunar-policy community is small relative to the consequence of its decisions; meaningful leverage is available to informed and sustained engagement.

13. Conclusion

The cislunar economy is going to happen. The geography makes it inevitable, the technology readiness makes it timely, the political and economic pressures that drive it (terrestrial energy constraints, strategic competition, scientific opportunity, the persistent human impulse outward) are not going to weaken. The question is not whether but on what timeline, and on whose terms.

The published roadmaps converge on a layered architecture: government-anchored foundations in the late 2020s and early 2030s; propellant economy in the 2030s; bulk materials and structural manufacturing in the 2040s; cislunar manufacturing for cislunar customers in the 2050s; a stable five-node network spanning LEO, GEO, L1, L2, and the lunar surface by 2060. The architecture is robust to most variations in assumption. What varies is the speed.

The fastest plausible trajectory delivers a self-sustaining cislunar economy by the late 2050s with low-thousands of humans living and working off-Earth. The slowest plausible trajectory delivers the same architecture by the 2080s. The gap is forty years. Most of the gap is determined by decisions being made now, in the 2026–2032 window, by actors who do not currently understand the bottleneck-inversion problem and are funding visible hardware at the expense of invisible infrastructure.

The three structural problems — chip-refresh, bottleneck-inversion, architectural displacement — explain why the current orbital-data-centre boom is structurally wrong, why the capital flowing into LEO compute satellites will mostly be lost, why the correct version of orbital compute happens at L1 in the 2050s with lunar-supplied infrastructure, and why no current program is on track to build that infrastructure. The capital allocated to LEO data centres in 2026 is paying for a transitional industry that does not become the industry of the 2050s; the actors who will own the 2050s industry are mostly not the actors who are publicly visible now.

The single largest underweighted strategic position in global cislunar policy is L1 infrastructure investment by a state actor or multi-state consortium in the 2026–2030 window. The single largest underweighted commercial position is Layer 1 infrastructure ownership rather than Layer 2 product investment. The single largest underweighted public-goods position is participation in standards and governance forums.

The argument is not that cislunar is overhyped. It is the opposite — that cislunar is correctly hyped in destination but profoundly underhyped in timing and architecture. The buildout that the published roadmaps describe is real, is achievable, and is forty years away from delivering on its potential. The actors who position for that timeline now will capture structural advantage for decades; the actors who pitch shorter timelines will mostly disappoint their investors and their political sponsors.

The geography is destiny. The politics is contingent. The window for shaping the contingent part is the next decade. The architecture is settled, the bottleneck is invisible, and the leverage is concentrated where almost no one is currently looking.

---

Todd McCaffrey · May 2026, revised June and October 2026. Workshop draft, Sections 1–13, about 19,200 words.