Orbital Shipyards

An orbital shipyard is a place where spacecraft and large structures are assembled, serviced and repaired in orbit rather than launched complete. The promise is that the launch vehicle's fairing stops being the ceiling on how large a thing can be. The framing under test is that building spacecraft in orbit removes the launch constraint.

This is the most established slot in its cluster and the page should read that way. Almost everything on it is flown hardware with dates. The International Space Station is on-orbit assembly at scale and it worked. Two commercial servicing vehicles have docked with live geostationary satellites; one delivered its full five-year contract and undocked. A robotic servicing vehicle carrying a DARPA payload launched in July 2026 and is in transit. A Japanese spacecraft flew within fifteen metres of a three-tonne derelict upper stage. A Chinese spacecraft towed a dead satellite three thousand kilometres above the geostationary belt and was tracked doing it by three independent organisations.

And then the framing inverts. On the only occasion large-scale on-orbit assembly has been done, it did not remove the launch constraint — it added a much more expensive one on top, and every kilogram was launched anyway. Both American in-space assembly flight demonstrations of the 2020s were cancelled before launch. The White House strategy of 2022 and the sector's own April 2026 assessment independently name the same barrier, and it is not technology.

The page's dividing lines: Space-Based Manufacturing owns making things in orbit to sell on Earth — its product comes home; this brief's product stays up. Deep Space Infrastructure owns the depots, power and logistics a shipyard would plug into, and cryogenic propellant transfer sits on that seam.

Established The existence proof is the International Space Station, and it comes with its price tag attached. NASA's own facts and figures: mass 925,335 lb, or 419,725 kg; pressurised module length 67 m along the major axis; truss 94 m; solar arrays 73 m across; habitable volume 388 m³; pressurised volume 1,005 m³; eight solar arrays generating 75–90 kW. And the assembly datum that matters here: “The large modules and other pieces of the station were delivered on 42 assembly flights.”

Established The chronology and the robotics are the part a shipyard brief should carry. Assembly began with Zarya on 20 November 1998; the US Orbital Segment was completed in 2011 with the Alpha Magnetic Spectrometer on STS-134. Thirty-six Space Shuttle flights delivered elements, supplemented by Proton, Soyuz-U and later Falcon 9. The robotic infrastructure that made it possible flew separately: Canadarm2 on 19 April 2001 and Dextre on 11 March 2008. The Russian segment went into indefinite hiatus after Rassvet in 2010 and resumed only with Nauka and Prichal in 2021.

Established And the cost, which is the counter-finding in its purest form: about $150 billion, the most expensive item ever built. Dividing that by 419,725 kg gives roughly $357,000 per kilogram of assembled structure. That comparison is crude and should be presented as such — the $150 billion covers thirty years of a programme, not assembly alone — but against any launch price ever charged it inverts the framing. On-orbit assembly as actually practised has cost about two orders of magnitude more per kilogram than getting mass to orbit at all.

Established Servicing, unlike assembly, is operational and commercial and has completed a contract. Northrop Grumman's MEV-1 (2,330 kg) launched on a Proton-M on 9 October 2019 and docked with Intelsat 901 on 25 February 2020, capturing the satellite's liquid apogee engine nozzle — a mechanism “compatible with approximately 80% of geosynchronous satellites.” It delivered five years of life extension and undocked in the geostationary graveyard, announced 9 April 2025 — the first-ever undocking between two commercial spacecraft in geosynchronous orbit — and is now relocating to its next servicing mission.

Established MEV-2 (2,875 kg) launched on Ariane 5 on 15 August 2020 and docked with Intelsat 10-02 on 12 April 2021, in geosynchronous orbit itself rather than in the graveyard. Its initial five-year contract has been extended by four further years, to 2029. These are Northrop Grumman's announcements about Northrop Grumman's missions. The docking dates, target satellites and orbit changes are independently observable from the ground and corroborated by an independent European mission record, which is why they carry the strong flag. The company president's claim that the work is “paving the way for an entire infrastructure of future on-orbit satellite servicing missions” does not.

Established And the newest hardware is in transit as this brief is written. DARPA's release of 21 July 2026 records SpaceLogistics' Mission Robotic Vehicle, carrying the Robotic Servicing of Geosynchronous Satellites payload, launching from Cape Canaveral Space Force Station on a Falcon 9. The payload is two robotic manipulator arms with seven joints each, interchangeable tools, cameras and autonomous close-proximity flight software; the vehicle can install Mission Extension Pods. Transit to geostationary orbit takes about one year, followed by inspection, anomaly resolution, upgrade and relocation over a ten-year-plus operational life. Partners are DARPA, SpaceLogistics, the US Naval Research Laboratory, NASA and SpaceX. An agency describing its own programme on launch day is an interested party: record the launch as fact and the capability claims as claims — and record the schedule too, because RSGS was a programme of the mid-2010s and flew in 2026.

Established The origin of all of it is a 2007 demonstration, and the gap since is the finding. Orbital Express (DARPA with NASA Marshall) launched on an Atlas V on 8 March 2007 under the USAF STP-1 mission: two spacecraft, ASTRO the servicer from Boeing and NEXTSat the serviceable client from Ball Aerospace, with robotics by MacDonald Dettwiler, refuelling hardware by VACCO and docking by Sierra Nevada. It transferred hydrazine and a battery orbital replacement unit, and changed out ASTRO's own flight computer, with final docking on 29 June 2007 across a three-month campaign of increasing autonomy, and was deactivated on 21–22 July 2007. Autonomous servicing was demonstrated in 2007; the first operational commercial servicing docking was in 2020. Thirteen years.

Established The counterweight is DART, and NASA's own report on it is unusually candid. The Demonstration of Autonomous Rendezvous Technology launched on 15 April 2005. A GPS receiver velocity bias of 0.6 m/s caused the navigation filter to diverge and reset “about once every three minutes throughout the mission”; continuous corrective thruster firing exhausted propellant, and the fuel-usage logic over-estimated consumption so that DART declared depletion with about 30% still in the tank. It missed a 6.3 m-radius waypoint by less than 2 m, which prevented handover to the ranging sensor — it could steer toward its target but could not tell how far away it was. Collision avoidance triggered 1 minute 23 seconds before impact. DART struck MUBLCOM at 1.5 m/s while its navigation believed it was 130 m away and retreating at 0.3 m/s.

Established The scorecard and the root causes are the transferable part. Eleven of 27 mission objectives fully or partially met; zero of 14 proximity-operations requirements met. Cost $47 M estimated against $110 M actual. Twelve root causes were named, including a “high-risk, low-budget procurement approach” with insufficient oversight, an inexperienced design team that rejected expert recommendations, untested late changes to a navigation gain setting under schedule pressure, and ITAR restrictions limiting communication with vendors. This is interest against the finding in its cleanest form — an agency publishing the full anatomy of its own failure — and its answer to “how hard is autonomous proximity operations” is that the failure was organisational as much as technical.

Established Non-cooperative rendezvous is now flying. Astroscale's ADRAS-J launched in February 2024 and approached a derelict H-2A upper stage — about 11 m long, 4 m across and 3 tonnes — to within 15 metres, imaging it and the payload adapter fitting that a follow-on would grapple. During one approach the spacecraft initiated an abort manoeuvre and recovered, then made multiple further approaches. Operations ended and controlled deorbit began on 25 March 2026 after 293 days in orbit. ADRAS-J2, which will actually remove the same object, is planned for FY2027. Astroscale reporting on its own mission is an interested party — and the closest-approach distance and target are independently trackable, and the release reports its own abort, which raises rather than lowers its weight.

Established The honest counter-case comes from the same company's own record. ELSA-d launched on a Soyuz-2.1b on 22 March 2021: a 180 kg servicer and a 20 kg client with a magnetic docking plate. On 25 August 2021 the first magnetic capture succeeded. On 25 January 2022 anomalous spacecraft conditions arose during the autonomous capture demonstration; capture attempts were paused and never resumed. By April 2022 four of eight thrusters were non-functional. On 7 April 2022 the servicer still demonstrated absolute-to-relative navigation handover at 159 m, and the mission ended on 5 April 2022. The full autonomous capture of a non-tumbling client — the headline objective — was not achieved.

Established And the capability is not uniquely Western. China's Shijian-21 docked with the defunct Beidou-2 G2 in late December 2021 and, on 22 January 2022, performed a large burn relocating it to about 3,000 km above the geostationary belt — an order of magnitude beyond the standard 300 km graveyard — undocking on 26 January 2022. It was tracked by COMSPOC, ExoAnalytic Solutions and the US Space Force's 18th Space Control Squadron. The dual-use ambiguity was stated directly by an analyst quoted in the trade press: “You could look at China working to develop the capability to remove inactive satellites on orbit as a way in which it is being a responsible space actor… Or… say that this could indicate that China is developing an on-orbit offensive capability.”

Established Now the assembly half, where the record is a cancellation. NASA cancelled OSAM-1 on 1 March 2024, citing “continued technical, cost, and schedule challenges, and a broader community evolution away from refueling unprepared spacecraft, which has led to a lack of a committed partner.” The mission would have refuelled Landsat 7 and assembled a Ka-band antenna using the SPIDER robotic payload. Prime contractor Maxar Technologies, on about $316 M in combined contracts; a NASA spokesperson noted about 450 NASA employees and contractors had worked on it.

Established The Independent Review Board report of 29 February 2024 is the strongest interest-against-finding document in this whole cluster: NASA's own board recommending NASA cancel NASA's flagship. Its numbers: “This effort, originally forecast to cost no more than $750M and launch by 2020, is seven to eight years behind schedule and $1,500M over budget.” The original 2017 baseline was $626–753 M for a 2020 launch; the November 2023 estimate was $2,380 M total, with about $1,400 M already spent and $980 M cost-to-go; most-likely launch readiness at 70% confidence was March 2028. Named technical risks included a Kodiak LiDAR eleven months late, the servicing payload's robotic arm assembly, flight software delivery, and translating test procedures to Goddard's ground system.

Established And the finding inside the cancellation that is almost always missed. The board did not conclude that on-orbit work is a bad idea. It concluded that this mission was aimed at the wrong target: “The community's widely held understanding is that OSAM-1 mission is overly focused on delivering propellant to an aged LandSat-7 spacecraft… not perceived to be of value.” Meanwhile the board judged the assembly demonstration “at least as useful as (if not more useful than) an unprepared servicing test,” and found “general interest” among industry in demonstrated in-space assembly. NASA's own board rated assembly a better bet than refuelling — and NASA cancelled the mission that carried both.

Established The other cancellation completes the picture. Archinaut, later OSAM-2: a NASA $20 M Phase 1 award in 2016 to Made In Space with Northrop Grumman and Oceaneering; a month-long thermal-vacuum manufacturing test at NASA Ames in June 2017; a 37-metre-plus 3D-printed beam in July–August 2017, a Guinness record; a $73.7 M contract in July 2019 for Archinaut One, to print two ten-metre solar arrays on an ESPA-class spacecraft; a 7.0 m flight-like beam printed in orbital-like conditions in 2020; critical design review passed in 2022; launch planned for 2024 — and NASA concluded the project in 2023 without flying it, preserving the data. Redwire had acquired Made In Space in 2020. Both US in-space assembly flight demonstrations of the 2020s were cancelled before launch, which is why this brief's strong flags sit on servicing rather than on shipyards.

Frontier What is actually next on the assembly side is small and its outcome is unknown. DARPA's NOM4D programme — Novel Orbital and Moon Manufacturing, Materials and Mass-efficient Design — moved to orbital demonstrations with two Phase 3 teams. Caltech with Momentus: a 1.4 m diameter circular truss simulating an antenna aperture, assembled in orbit from lightweight composite-fibre longerons by an autonomous robotic gantry, manifested for Falcon 9 Transporter-16 in February 2026. University of Illinois Urbana-Champaign with Voyager Space: composite tube formation from carbon fibre and liquid monomer by frontal polymerisation — “you just ignite one end of the inside of the tube and the reaction self-propagates, stiffening the carbon tubes” — manifested for NASA commercial resupply mission NG-24, tentatively April 2026. The stated ambition is future space-based RF antennas of 100 metres or more. This research pass retrieved no outcome report for either demonstration, both of which were scheduled for the first half of 2026. That is stated here as an unknown rather than assumed to be a success.

Established The institutional frame is a national strategy, and it is candid. The In-Space Servicing, Assembly, and Manufacturing National Strategy, issued by the ISAM Interagency Working Group of the National Science and Technology Council in April 2022, sets six goals verbatim: advance ISAM research and development; prioritise expanding scalable ISAM infrastructure; accelerate the emerging ISAM commercial industry; promote international collaboration and cooperation; prioritise environmental sustainability; inspire the future space workforce. Its definitions are the ones to adopt — servicing is “in-space inspection, life extension, repair, or alteration of a spacecraft”; assembly is “the construction of space systems in space using pre-manufactured components”; manufacturing is “the transformation of raw or recycled materials into components, products, or infrastructure in space.” And its own statement of maturity: “Use cases of ISAM capabilities are not fully characterized and proven due to the nascence of the field,” with three named barriers — insufficient coordination, lack of clear demand signals from government to stimulate private investment, and underdeveloped standards and interfaces.

Established NASA's own capability survey supplies the field's shape in counts. On-orbit Servicing, Assembly, and Manufacturing State of Play, 2021 Edition organises the field into eleven capability areas and counts activity in each: Robotic Manipulation 27 entries, Rendezvous, Proximity Operations, Capture, Docking and Mating 21, Refuelling and Fluid Transfer 13, Structural Manufacturing and Assembly 13, Inspection and Metrology 11, Unplanned and Legacy Repair 5, Recycling, Reuse and Repurposing 5. Its historical spine runs Gemini VIII (first crewed docking, 16 March 1966), Kosmos 186/188 (first autonomous docking, 30 October 1967), ETS-VII (1998), XSS-10 and XSS-11 (2003 and 2005), Orbital Express (2007), MEV-1 and MEV-2, ELSA-d, five Shuttle servicing missions to Hubble, and the ISS. Its assessment of the thin spots is equally direct: “very little has been demonstrated for in-space cryogenic fluid management and transfer”; surface construction plans are “still in very early stages”; the key gap is “standardized, interoperable interfaces for mechanical, fluid, power, data”; and the commercial trend runs “away from expensive, bespoke, human-operated manipulators and toward proliferated robotics in space with low cost, autonomous operations.”

Frontier And the 2026 sector assessment, which is the counter-finding stated by the industry to itself. The State of ISAM 2026, published by a trade outlet on 8 April 2026, finds servicing with early wins and government and defence as primary buyers; manufacturing closest to viability, with a commercial station operator stating that “all of the commercial rack space… has been sold out”; and assembly furthest from commercialisation, with structures still required to fit inside fairings and no regular commercial operations. The sentence that matters: “They're all still pretty much bespoke, one-off contracts… No one is putting in for a five-mission servicing contract to GEO.” Against which one executive's counter-view is recorded in the same piece: “We are either at that tipping point, or we will look back… as having already crossed it in early 2026.” Carry both; the disagreement is the honest state.

Established Position one is not in dispute: large structures can be assembled on orbit. 419,725 kg across 42 assembly flights, with a 94 m truss and 73 m arrays, operating continuously since 1998. Position two, equally solid: it cost about $150 billion, which is roughly $357,000 per assembled kilogram. The two facts belong together and almost never appear together.

Established Position three: commercial geostationary life extension works and has completed a contract. MEV-1 docked in 2020, delivered five years and undocked in April 2025. Position four, the vendor's extrapolation from it: this is “paving the way for an entire infrastructure” of servicing missions. That is an interested party's forecast rather than a result, and the 2026 sector assessment disagrees with it in the same year.

Established Position five: robotic geostationary servicing with dual seven-joint arms is now in transit, launched 21 July 2026 with about a year to reach station. Its capability claims are DARPA's own and remain claims until the vehicle works. Position six: autonomous rendezvous, capture, propellant transfer and orbital-replacement-unit exchange were all demonstrated in 2007. Every operational servicing mission since descends from that flight.

Established Position seven: autonomous proximity operations are hard for organisational as much as technical reasons. DART's twelve root causes name procurement approach, team experience, rejected expert advice and schedule-driven untested changes alongside the GPS bias. Position eight: non-cooperative close approach to a derelict is achievable — 15 metres from a 3-tonne upper stage. Position nine: autonomous capture of a client is not yet achieved, because ELSA-d's headline objective was abandoned after an anomaly and never resumed.

Established Position ten: relocation of a defunct satellite in geostationary orbit by a second spacecraft is operational, and not only in the United States. Shijian-21 towed Beidou-2 G2 three thousand kilometres above the belt, tracked by three independent organisations. Position eleven: the same capabilities are offensive capabilities — stated in those terms by a named analyst, and structurally undeniable, since rendezvous, grapple and relocate do not care whether the target consented.

Established Position twelve: NASA's flagship servicing-and-assembly demonstrator was seven to eight years late and $1,500 M over budget. From its own Independent Review Board. Position thirteen: refuelling unprepared legacy spacecraft is not where the value is, per both the board and NASA's cancellation language. Position fourteen, and the one that reframes the cancellation: in-space assembly is at least as useful a demonstration as unprepared servicing — the judgement of NASA's own board, inside the document recommending cancellation.

Established Position fifteen: 3D printing structural beams in vacuum works on the ground. Over 37 metres printed in 2017 and a 7.0 metre flight-like beam in orbital-like conditions in 2020. Position sixteen: an orbital demonstration of the same is imminent. Held in 2019–2022 for Archinaut One and falsified for that vehicle when NASA concluded the project in 2023 without flying it. Position seventeen: autonomous robotic assembly of a 1.4 m truss and frontal-polymerisation tube formation will fly in 2026. Both manifested; no outcome was retrievable for this brief.

Established Position eighteen, and the one two independent official documents agree on: the binding barrier is a demand signal rather than technology. The 2022 National Strategy names lack of clear demand signals as one of three barriers; the 2026 sector assessment finds servicing contracts still bespoke and one-off. Position nineteen: the sector is at or just past a commercial tipping point. An unnamed executive in the same 2026 piece, and it is worth carrying precisely because it contradicts the assessment it appears in. Position twenty: assembly remains the least commercialised of the three activities, with structures still required to fit inside fairings.

Established Position twenty-one: cryogenic fluid transfer in space is essentially undemonstrated, in NASA's own words. Position twenty-two: interface standardisation is the key gap, named independently by the agency survey and the national strategy. And the genuinely open question the record does not answer: what a shipyard would be for. Nothing retrieved shows anyone waiting on orbital assembly to build something they otherwise could not. Planetary Scale Energy Systems records the one concrete requirement anyone has costed — a 5.9-million-kilogram orbital structure for space-based solar power — which is fourteen times the ISS and is the scale a real shipyard would have to reach.

Established The first bottleneck is demand, and two official documents four years apart name it independently. The 2022 National Strategy lists “lack of clear demand signals from government to stimulate private investment” as one of three barriers, alongside insufficient coordination and underdeveloped standards. The April 2026 sector assessment finds the servicing market still made of “bespoke, one-off contracts… No one is putting in for a five-mission servicing contract to GEO.” Four years of a national strategy later, the constraint named is still demand.

Established The second is interface standardisation, and it is the one that would unlock the rest. NASA's own survey names the key gap as “standardized, interoperable interfaces for mechanical, fluid, power, data”, and the national strategy names it too. MEV's capture mechanism works on about 80% of geostationary satellites precisely because it grapples a feature — the liquid apogee engine nozzle — that everyone happens to have. The whole servicing market currently rests on an accident of common design rather than on a standard.

Established The third is autonomous capture of an uncooperative object, which has not been achieved. ELSA-d's autonomous capture demonstration was abandoned after an anomaly and never resumed; ADRAS-J observed a derelict from fifteen metres and did not grapple it; ADRAS-J2 is planned for FY2027. The gap between approaching and capturing is where the entire debris-removal and salvage case sits, and it is currently unbridged in the flown record.

Established The fourth is cryogenic propellant transfer, and NASA's assessment is blunt: very little has been demonstrated. It sits on the seam with Deep Space Infrastructure, which owns depots, and it is named here rather than claimed — but no shipyard that assembles a departure stage works without it.

Established And the fifth is cost, which is the bottleneck the framing denies exists. The one scale data point says on-orbit assembly cost roughly $357,000 per assembled kilogram against launch prices two orders of magnitude below that. The framing's real promise — that assembly lifts the fairing-diameter ceiling — is true and important. Its implied promise, that this makes large spacecraft cheaper, is contradicted by the only data point in existence.

Established The first adjudicated dependency is Space-Based Manufacturing, and the seam is a single sentence: their product comes home, this brief's product stays up. That brief owns the microgravity physics and the market for goods made in orbit and sold on Earth — crystals, optical fibre, pharmaceutical materials, and the valley-of-death problem that market has. This brief owns structural assembly, joining, robotic manipulation and servicing. Archinaut's printed beams belong here because they are structure; a fibre-drawing furnace belongs there because its output is cargo.

Established The second is Deep Space Infrastructure, which owns the power, docking nodes, logistics and depots a shipyard would plug into. This brief owns the assembly operation itself. Cryogenic propellant transfer sits exactly on that seam — NASA's own survey says it is essentially undemonstrated, and it is a depot capability rather than an assembly one, so it is named here and handed over.

Frontier A third dependency is real, unadjudicated, and explicitly declined rather than assumed. Space Resource Economies and Asteroid Mining own feedstock sourced in space rather than launched, which is what would eventually make orbital assembly economic. Every source retrieved for this brief treats current assembly as working from launched components, so this page states the dependency and refuses to assume it.

Speculative And a dependency that runs the wrong way round, which is the honest description of the sector's problem. O'Neill Cylinders and Space Habitats are usually named as things a shipyard would enable. On the evidence they are better described as the demand this brief lacks — customers that do not yet exist for a capability whose barrier is, by two official accounts, the absence of customers.

Established The experiments in this subject are flights, and the informative ones are the failures. DART's mishap investigation is the single most useful document about autonomous proximity operations in the public record, because it reports a full anatomy: a 0.6 m/s velocity bias, a filter resetting every three minutes, fuel logic that declared depletion with 30% remaining, a 6.3 m waypoint missed by under 2 m, and a collision at 1.5 m/s while the navigation believed the target was 130 m away and receding.

Established ELSA-d is the second, and it is valuable for the same reason. A magnetic capture that worked in August 2021, an anomaly in January 2022 that stopped the autonomous demonstration permanently, four of eight thrusters lost by April 2022, and a successful absolute-to-relative navigation handover at 159 m recorded anyway. A partial failure documented in detail by an independent mission record is worth more to a reader than a success announced by a vendor.

Established The experiments actually manifested are two small orbital assembly demonstrations and their results are unknown to this brief. A 1.4 m composite truss assembled by an autonomous gantry, manifested for February 2026; frontal-polymerisation tube formation, manifested for April 2026. Both were scheduled for the first half of 2026 and no outcome report was obtainable, and an honest unknown is better here than an assumed success.

Established The experiment that would change the field's economics is a standardised servicing interface flown by more than one operator. Both the national strategy and NASA's capability survey name interfaces as the key gap; MEV's success rests on a de facto commonality rather than a standard. A demonstration in which two vendors' servicers dock with the same client fitting would be a smaller flight than any of the above and would matter more than all of them.

Frontier And the experiment nobody has manifested: capture of a tumbling, uncooperative object. ADRAS-J proved the approach; ADRAS-J2 is planned for FY2027 against a known, characterised, non-tumbling target. The flown record contains no autonomous capture of an uncooperative client at all, and the whole debris-removal case depends on it.

Established The engineering requirements for this subject are unusually well characterised, because a national agency wrote them down and counted the activity in each. Eleven capability areas: robotic manipulation, rendezvous and proximity operations and capture and docking and mating, refuelling and fluid transfer, structural manufacturing and assembly, inspection and metrology, unplanned and legacy repair, recycling and reuse and repurposing, among others — with 27, 21, 13, 13, 11, 5 and 5 catalogued entries respectively.

Established The manipulator requirement has flown and has a shape. Canadarm2 and Dextre built and maintain the ISS; RSGS carries two arms with seven joints each plus interchangeable tools and cameras; Orbital Express exchanged an orbital replacement unit and a flight computer robotically in 2007. And the trend is away from that heritage: NASA's own survey records the commercial direction as moving “away from expensive, bespoke, human-operated manipulators and toward proliferated robotics in space with low cost, autonomous operations.”

Established The proximity-operations requirement has a demonstrated floor and a documented failure mode. Sensor handover between absolute and relative navigation is the specific step that killed DART — it missed a waypoint by under two metres and therefore never handed over to its ranging sensor — and it is the step ELSA-d successfully demonstrated at 159 m even after its capture campaign had been abandoned. Handover, not approach, is the hard part.

Established The structural requirement is where the record is thinnest and the numbers are smallest. A 37-metre-plus beam printed on the ground in 2017, a 7.0 metre flight-like beam in 2020, and a 1.4 metre truss manifested for orbit in 2026. DARPA's stated ambition is space-based RF antennas of 100 metres or more, and its release gives no mass-efficiency ratio, so none is quoted here. Against a 5.9-million-kilogram space-solar structure — the one large assembly requirement anyone has costed, and it belongs to Planetary Scale Energy Systems — a 1.4 metre truss is the honest current state.

Established And the interface requirement, which every official document names and no programme has delivered. Mechanical, fluid, power and data interfaces that are standardised and interoperable across vendors. The current servicing market works because roughly 80% of geostationary satellites happen to share a liquid apogee engine nozzle a gripper can grab — an accident of common design, not an engineering standard, and it does not extend to refuelling, to assembly, or to any spacecraft built differently.

Established The two adjudicated neighbours divide the subject cleanly and both divisions are worth stating. Space-Based Manufacturing owns making things in orbit to sell on Earth; this brief owns making and maintaining things in orbit to use in orbit. Deep Space Infrastructure owns the depots, power and logistics nodes; this brief owns the assembly operation. Cryogenic propellant transfer belongs to the second and is named here only because NASA's survey says it is essentially undemonstrated.

Frontier The upstream adjacency is feedstock and this brief explicitly declines to assume it. Space Resource Economies, Asteroid Mining and Moon-Based Manufacturing own material sourced off Earth, which is the condition under which orbital assembly stops competing with launch and starts complementing it. Every source in this brief describes assembly from launched components.

Established The downstream adjacency is where the demand would have to come from, and naming it as demand rather than as application is the point. O'Neill Cylinders, Space Habitats and Space-Based Solar Power are the customers a shipyard needs and does not have. Orbital Rings and Space Elevators are further out still and would need orbital assembly at scales this brief's record does not approach.

Frontier And the enabling adjacencies are robotics and autonomy, which is where the real technical progress in this subject lives. Robotics in Infrastructure and Automated Construction Systems own the manipulator and autonomy state of the art on the ground. DART's mishap report is the strongest available evidence that the binding difficulty in orbital robotics is autonomy under uncertainty rather than mechanism design — the arms work; the navigation is what fails.

Established This is the rare subject on this map where the institutional record is the strongest part of the evidence rather than the weakest. A White House-level national strategy with six named goals and three named barriers. An agency capability survey with eleven areas and counted activity in each. An Independent Review Board report with baselines, spend and cost-to-go. A published mishap investigation with twelve root causes. Two funded DARPA programmes with hardware in orbit. Almost every claim on this page has an institutional document behind it.

Established The OSAM-1 Independent Review Board is the exhibit worth transferring off this page. It is a NASA board, reporting to NASA, recommending cancellation of NASA's flagship, with the numbers written down: $626–753 M baselined for a 2020 launch, $2,380 M projected, $1,400 M spent, $980 M to go, seven to eight years late. And it did not conclude that the underlying capability was wrong. It concluded the mission was aimed at a target the community did not value, while rating the assembly demonstration “at least as useful as (if not more useful than) an unprepared servicing test.” An agency's own board separating a bad mission from a good capability, in the document that kills the mission, is what institutional self-criticism should look like.

Established The interested parties in this subject are unusually checkable, which is why so much of this page carries strong flags. Northrop Grumman's docking dates are independently trackable from the ground. Astroscale's release reports its own abort manoeuvre. DARPA's launch announcement is a launch that happened. The claims that remain claims are the forecasts — “paving the way for an entire infrastructure”, “fundamentally change how we think about… resilience”, “we are either at that tipping point” — and this brief carries them labelled as such.

Established The institutional failure is a demand-side one and it is admitted in writing. A national strategy in 2022 identifies the government's own failure to send a demand signal as a barrier to private investment. A sector assessment in 2026 finds the market still made of one-off contracts. The instrument that would fix it is procurement rather than research — a multi-mission servicing contract of the kind the 2026 piece says nobody is bidding for — and that is an institutional choice, not a technology gap.

Established And a governance gap with hardware behind it. Shijian-21 relocated another state's defunct satellite to an orbit an order of magnitude beyond the standard graveyard, tracked by three organisations, with no regime governing the act. Space Law and Governance owns the regime; what this brief supplies is that the capability arrived first.

Established The dual-use problem here is not hypothetical and it does not need to be argued. Rendezvous, proximity operations, grapple and relocate are the same capability set whether the target consents or not. Shijian-21 towed a satellite to a novel orbit and was tracked doing it. RSGS is a DARPA programme flown with the Naval Research Laboratory. ADRAS-J's fifteen-metre approach to an uncooperative object is, in capability terms, indistinguishable from a fifteen-metre approach to somebody else's working satellite.

Established The analyst's framing quoted in the trade press is the honest statement of the ambiguity, and it should be quoted rather than resolved: the same Chinese demonstration can be read as responsible debris management or as the development of an on-orbit offensive capability, and nothing observable distinguishes them. That is a permanent feature of the technology rather than a stage it passes through, and any governance regime has to be built on intent and transparency because the hardware will not tell you.

Established Debris responsibility runs in both directions and this brief has a case study for each. Servicing removes debris — that is ADRAS-J2's entire purpose. Servicing creates debris when it goes wrong — DART struck its target at 1.5 m/s. The same mission class is both the remedy and the hazard, and DART is the exhibit for both.

Established A fourth consideration is about spending rather than safety. OSAM-1 consumed about $1,400 M and the work of roughly 450 people before its own review board recommended cancellation, on a mission the community did not think was valuable. The ethical content is in the board's finding rather than in the cancellation: the target was chosen without a committed partner, and the resulting write-off is the cost of building a capability nobody had asked for.

Speculative And one forward-looking consideration with no hardware behind it yet. If servicing becomes routine, the operational lifetime of geostationary satellites extends indefinitely and the graveyard-disposal norm weakens, because a satellite with a servicer contract is never quite dead. Nothing retrieved treats that as a policy question, and it follows directly from MEV-1's success.

Established The civilizational finding here is a correction rather than a vision, and it is the sharpest one in the cluster. In-orbit assembly does not remove the launch constraint. On the only occasion it has been done at scale it added a far more expensive constraint on top: 42 assembly flights, 36 of them Shuttle, plus Canadarm2 and Dextre, plus years of crew time, to produce 419,725 kg at a programme cost near $150 billion. Every kilogram was launched anyway. Assembly did not substitute for launch; it added an orbital labour cost to it.

Established The framing's true half is worth stating just as plainly. Assembly lifts the fairing-diameter ceiling, and that is real: the ISS truss is 94 m across and no fairing has ever been. What the record does not support is the implied second half, that this makes large spacecraft cheaper. The one data point says the opposite and nothing retrieved contradicts it.

Established The second finding is a split within the subject that the word “shipyard” hides. Servicing launched, docked, delivered five contracted years, undocked, and launched again. Assembly had two American flight demonstrations in the 2020s and both were cancelled before launch — OSAM-1 in March 2024 after about $1.4 billion spent, and Archinaut One in 2023 after $93.7 M in awards and a passed critical design review. The capability that works services satellites that already exist. The capability the framing depends on — building things — has not flown.

Established And the third reframes what the sector is actually short of. The framing assumes launch is the limit; the record says the limit is a customer. NOM4D's orbital demonstrations are 1.4 metres across. Nothing in the retrieved record shows anyone waiting on assembly to build something they otherwise could not. The one concrete large-assembly requirement anyone has costed is a 5.9-million-kilogram space-solar structure — fourteen ISS masses — and it belongs to Planetary Scale Energy Systems as an economic proposition that brief finds unpersuasive. That is the shape of the problem: the demand that would justify a shipyard is itself waiting on an economic case nobody has closed.

These horizons track flown hardware, funded programmes and contract structures, which is why they are firmer than anywhere else in this cluster:

• 10 yr: Established RSGS reaches geostationary orbit in about a year and begins a ten-year-plus operational life; ADRAS-J2 attempts the first removal of a large derelict in FY2027; the NOM4D orbital demonstrations either flew in the first half of 2026 or slipped, and their outcomes will be public well inside this window. Frontier Expect servicing to consolidate with government and defence as primary buyers and expect the bespoke-contract structure to persist unless a multi-mission procurement appears. Frontier Expect assembly demonstrations to remain metres rather than tens of metres.
• 25 yr: Frontier The determining variable is procurement rather than technology: a standing multi-mission servicing contract, or a standardised interface mandated across a satellite fleet, would change the sector faster than any demonstration. Established Cryogenic propellant transfer is the one capability with a clear demand case attached — departure stages — and NASA's own survey says almost nothing has been demonstrated. Speculative Hundred-metre-class RF apertures are DARPA's stated ambition and would be the first assembly product that could not have been launched whole.
• 50 yr: Speculative A genuine shipyard — a facility that assembles vehicles rather than servicing satellites — requires a customer, and the only candidates on this map are habitats and orbital power, both of which are themselves waiting on economic cases nobody has closed. Frontier The alternative and equally plausible trajectory is that launch keeps getting cheaper and larger, fairings keep growing, and assembly stays a niche for the few structures that genuinely cannot fit.
• 100 / 250+ yr: Speculative Beyond useful forecasting, but the structural statement is unusually clear for this cluster: orbital assembly is a demonstrated capability waiting on demand, not a hypothesis waiting on physics. Handwave Anything at this horizon depends on space-sourced feedstock, which Space Resource Economies owns and which this brief explicitly declines to assume.

• Depends on Two edges, and both are boundaries rather than blockers — this brief is not waiting on a result so much as sharing a field with two neighbours. Space-Based Manufacturing owns making things in orbit to sell on Earth: the microgravity physics, the products, and the market's valley-of-death problem. Its product comes home; this brief's product stays up. Archinaut's printed beams sit here because they are structure. Deep Space Infrastructure owns the power, docking nodes, logistics and depots a shipyard plugs into, and cryogenic propellant transfer sits precisely on that seam — NASA's own capability survey says very little has been demonstrated, it is a depot capability rather than an assembly one, and it is named here and handed over rather than claimed.
• Requires (not on this map) Two constraints that are not briefs on this map, and both are named in official documents rather than inferred. The first is a market structure: the April 2026 sector assessment records that servicing work is “all still pretty much bespoke, one-off contracts… No one is putting in for a five-mission servicing contract to GEO”, and the 2022 National Strategy names lack of a government demand signal as one of three barriers to private investment. The second is institutional rather than technical: NASA's own capability survey names standardised, interoperable mechanical, fluid, power and data interfaces as the key gap, and the whole current servicing market rests instead on the accident that roughly 80% of geostationary satellites share a grabbable engine nozzle.
• Enables No typed enabling edge is claimed, and the reason is the brief's central finding. The capabilities that would consume orbital assembly — O'Neill Cylinders, Space Habitats, Space-Based Solar Power at real scale — are better described as the demand this brief lacks than as the things it enables. Two official documents four years apart name demand as the binding barrier; the sector's own 2026 assessment finds servicing contracts still bespoke and one-off; and both American in-space assembly flight demonstrations of the 2020s were cancelled before launch. An enabling edge would assert a direction of causation the record reverses.
• Adjacent Space Resource Economies, Asteroid Mining and Moon-Based Manufacturing own feedstock sourced off Earth, which is the condition under which assembly stops competing with launch — named here and explicitly not assumed, because every retrieved source describes assembly from launched components. Robotics in Infrastructure and Automated Construction Systems own the manipulator and autonomy state of the art, and DART's mishap report is the strongest evidence that autonomy under uncertainty rather than mechanism design is the binding difficulty. Space Law and Governance owns the regime that Shijian-21's tow of another state's dead satellite arrived ahead of.

Established “Building in orbit removes the launch constraint.” On the only occasion it has been done at scale it did the opposite. The ISS took 42 assembly flights, 36 of them Shuttle, plus two purpose-built robotic systems and years of crew time, to produce 419,725 kg at about $150 billion — roughly $357,000 per assembled kilogram. Established Every kilogram was still launched. Assembly lifted the fairing-diameter ceiling, which is real and valuable; it did not lower the cost of getting mass to orbit, and the implied promise that it would is contradicted by the only data point in existence.

Established “Orbital shipyards exist, or are nearly here.” What exists is geostationary life extension, one robotic servicing vehicle in transit, one demonstrated fifteen-metre approach to a derelict, one foreign satellite relocation, and two cancelled assembly demonstrators. Frontier The largest orbital assembly structure currently manifested anywhere is 1.4 metres across. The word “shipyard” describes an ambition, and the honest current noun is “servicer”.

Established “OSAM-1's cancellation was a verdict on in-space assembly.” It was a verdict on refuelling unprepared legacy spacecraft. NASA's own Independent Review Board found the mission “overly focused on delivering propellant to an aged LandSat-7 spacecraft… not perceived to be of value” — and in the same report rated the assembly demonstration “at least as useful as (if not more useful than) an unprepared servicing test,” noting general industry interest in demonstrated in-space assembly. Established NASA's board rated assembly the better bet and NASA cancelled the mission that carried both.

Established “OSAM-1 cost roughly $2 billion.” The round number is worse than the real ones, which come from NASA's own board: baselined at $626–753 M for a 2020 launch, projected at $2,380 M total in November 2023, with about $1,400 M already spent and $980 M cost-to-go, and a 70%-confidence launch readiness of March 2028. Established Quoting the board is a stronger citation than quoting a rounded press figure, and the four numbers say more than the one.

Established “Autonomous capture of a tumbling object has been demonstrated.” It has not. ELSA-d's autonomous capture demonstration was halted by an anomaly in January 2022 and never resumed, with four of eight thrusters lost by that April; ADRAS-J approached a derelict to fifteen metres and imaged it but did not grapple it; ADRAS-J2 is planned for FY2027. Frontier The flown record contains approach, inspection, cooperative docking and relocation of a docked object — and no autonomous capture of an uncooperative client.

Established “The technology is the bottleneck.” Two official documents four years apart say otherwise and neither is an advocacy publication. The 2022 National Strategy names “lack of clear demand signals from government to stimulate private investment”; the April 2026 sector assessment finds contracts still “bespoke, one-off… No one is putting in for a five-mission servicing contract to GEO.” Frontier Four years of national strategy later, the named constraint has not changed.

Established “Refuelling in space is a solved problem.” Hydrazine transfer was demonstrated by Orbital Express in 2007, which is a real and often-forgotten result. Cryogenic transfer is not: NASA's own capability survey states that “very little has been demonstrated for in-space cryogenic fluid management and transfer.” Frontier The distinction matters because the missions that would justify a shipyard — departure stages, deep-space vehicles — need the cryogenic case, which belongs to Deep Space Infrastructure.

Frontier “Servicing is a peaceful capability and anti-satellite work is something else.” There is no observable difference. Rendezvous, proximity operations, grapple and relocate work identically on a consenting client and a non-consenting one. Established Shijian-21 towed a defunct satellite three thousand kilometres above the geostationary belt and was tracked doing it; RSGS is a DARPA programme flown with the Naval Research Laboratory. The named analyst's formulation is the right one: the same demonstration reads as responsible debris management or as offensive capability development, and the hardware does not distinguish them.

Frontier “The 2026 assembly demonstrations succeeded.” Two were manifested — a 1.4 m autonomously assembled truss for February 2026 and frontal-polymerisation tube formation for April 2026 — and this research pass retrieved no outcome report for either. Speculative An honest unknown is better than a guessed success, and a reader should treat any confident claim about their results with suspicion until a report exists.

Established And the framing itself, which fails in an interesting direction. “Building spacecraft in orbit removes the launch constraint” assumes launch is the limit. Established The record says the limit is a customer. Servicing works, is commercial, has delivered a contract and undocked on schedule; assembly has two cancelled demonstrators and a 1.4-metre truss; and the sector's own documents name demand rather than technology twice, four years apart. Frontier The fairing is not the binding constraint any more, and the constraint that replaced it is not one an engineering programme can lift.