In November 2014, aboard the International Space Station, the Made In Space “3D Printing in Zero-G” demonstrator built for NASA printed its first object. It was a faceplate for its own print head. The first thing anyone manufactured in orbit was a repair to the machine doing the manufacturing. Ask who the builder was in that moment, though, and the answer gets less obvious. Was it the printer, the astronaut who loaded the file, or the engineers on the ground who wrote the toolpath? That question is the right way into a concept that architects will meet more and more often: additive manufacturing in space.

The question is also current. On 11 October 2026, VoxelMatters reported that amsight, a quality-software company in Hamburg, has been chosen to lead DIQAM, a 12-month project inside ESA’s FIRST! Initiative. Its aim is to make qualification evidence for 3D-printed space hardware digital and reusable. Today that evidence still depends on computed tomography scans, companion specimens and documents assembled by hand. When a part is printed where no inspector can stand, someone still has to be able to say why it can be trusted. That is a problem of accountability as much as a problem of fabrication.

What it is

What it is: Additive manufacturing in space means building objects layer by layer inside the space environment itself. WIPO’s technology annex gives it a second name, in-situ manufacturing. Where possible, the material is already there: lunar or Martian regolith instead of freight launched from Earth. The concept has two branches. In orbit, it means printing tools, spare parts and eventually structural parts on a station. On a planetary surface, it means robotic construction: pads, berms, roads and habitat shells printed from local soil. As Sacco and Moon set out in their 2019 status review, both branches have to work under two conditions no workshop on Earth deals with, microgravity and vacuum.

Why it works

Why it works: In orbit the standard process is fused filament fabrication (FFF). A drive gear pushes the filament through the nozzle, so extrusion does not depend on gravity at all. What is missing is natural convection. Warm air does not rise, so a closed thermal loop has to keep each layer stable. Tracie Prater and her colleagues at NASA’s Marshall Space Flight Center published the full results of the Zero-G mission in 2019. Their careful comparison of flight prints against ground prints is the evidence the whole field still relies on. FFF remains the most predictable baseline, especially for recycling loops in which a broken part becomes feedstock for the next one.

On the surface there are three main routes. Regolith can be bound (D-Shape uses a magnesium-oxide binder), sintered by laser, microwave or concentrated sunlight, or mixed into a geopolymer. The trade-off is plain: binders need chemistry shipped from Earth, sintering needs power, and geopolymers need activators, so every route spends a resource the site does not have in surplus.

Then comes the inversion that changes how you think about a wall. Lunar gravity is 1.62 m/s². Take regolith at 1,500 kg/m³: a 3 m printed shell presses down with about 7.3 kPa, while a crew habitat at roughly 101 kPa pushes outward. The governing load case is a pressure vessel in tension, not a wall in compression. Printed granular regolith is weak in tension, so the building splits the work. A membrane or tensile liner holds the air. Its stress follows the classic membrane relation σ = P·r / (2·t). The printed shell provides shielding against radiation and micrometeoroids, plus thermal mass against the lunar day-night swing. Its weight enters the calculation only as a small counterweight.

←TODAY: In 2026, ESA is funding a digital qualification framework for printed space parts, because CT scans and hand-assembled dossiers do not scale.
→3012: Every printed wall carries its own evidence: toolpath, recipe and load case, signed at the moment of deposition.
Fulcrum: Once the robot builds where no one can inspect, the model stops being a drawing and becomes the record of who decided what.

Origins

Origins: Architecture worked towards this from the ground before the station had a printer. Behrokh Khoshnevis developed Contour Crafting at the University of Southern California. Enrico Dini developed the D-Shape binder-jet process. Between them they gave the field its first workable ideas for printing walls out of soil. In 2013, Foster + Partners worked with ESA and Monolite on a lunar habitation study: a catenary, cellular regolith shell over an inflatable dome, with a 1.5-tonne demonstration block printed from simulated lunar soil. Cesaretti, Dini, De Kestelier and colleagues documented it in Acta Astronautica in 2014. Its section was already correct: the membrane holds the air and the printed shell shields.

NASA’s 3D-Printed Habitat Challenge (2015–2019) moved the idea from paper to full-scale prototypes. SEArch+ and Clouds Architecture Office won the first phase with the Mars Ice House, a shell of printed water ice. Ice is rich in hydrogen, so it shields against radiation and still lets in daylight. AI SpaceFactory won the final phase with MARSHA, a vertical egg of basalt fibre and PLA biopolymer printed at subscale in Peoria, Illinois. The egg works as a pressure vessel and stacks its floors on a small footprint. In orbit, the commercial Additive Manufacturing Facility reached the ISS in 2016. In 2024, ESA’s metal printer, built by Airbus, made the first metal parts printed in orbit, inside the Columbus module. On the ground, ICON and BIG’s Mars Dune Alpha, a printed analogue habitat of about 158 m² at NASA Johnson Space Center, has hosted the year-long CHAPEA crews since June 2023. Project Olympus, ICON’s lunar construction system developed with BIG and SEArch+, was extended by NASA under an SBIR Phase III contract in 2022. It plans the pad, the berm and the road before the house.

Behind every milestone there is a group whose names rarely reach the headline: the flight engineers who qualified a printer for a crewed module, the technicians who sieved simulant soil by the tonne, and the analysts who compared hundreds of tensile coupons so that a faceplate could be trusted. The concept exists because their checking worked.

Who decides at 384,000 km

The Moon sits about 384,000 km away, so every command arrives more than a second late. A robot printing a berm cannot pause each layer to wait for a human, so it has to finish the layer by itself. Where does human agency go when the operator can only watch? It moves upstream, into the toolpath, the material recipe and the acceptance criteria written before the print starts. The model becomes the site instruction, and whoever writes it carries the accountability. DIQAM is a response to exactly this. It is an attempt to make the evidence travel with the part, so that responsibility does not dissolve into the distance.

In practice

In practice: No Swiss studio is tendering a lunar habitat. The discipline transfers all the same. ETH Zürich’s NCCR Digital Fabrication, with Gramazio Kohler Research and its In situ Fabricator, spent a decade showing that a robot on a real site needs the same things a lunar printer needs: a model it can execute, tolerances it can sense, and a human who has decided in advance what counts as acceptable. A 12-person office in Zürich working in Rhino + Grasshopper and Archicad already writes those decisions down, just not explicitly. Through the PAZ Grasshopper↔Archicad Library, a layer’s job can travel from the Grasshopper definition into the Archicad composite as data, not as a note in a margin. SIA 261 lists the actions on a structure. The space case adds a question SIA 261 does not ask: which layer answers which action?

Atelier: As robots and AI agents take over more of the toolpath, an office’s real authorship moves into the acceptance criteria it writes before anything is built. The lunar section, with liner, gap and shell, each doing one job, is a clean model of that habit. Monday move: in your next wall or roof build-up, add one Grasshopper panel per layer that states its single job (containment, shielding, thermal mass or structure) and the load case that governs it, and refuse to issue the section until every layer has one.

Hack

Hack: Weigh the regolith against the air before you draw a single ring. The few lines below run in a Python 3 script component in Rhino 8’s Grasshopper. They compute t*, the shell thickness at which the shell’s own weight just cancels the cabin pressure on Earth, Mars and the Moon, and then the hoop stress a 4 mm liner of 4 m radius must carry. The calculation follows the one-dimensional check in the PAZ concept panel. Expect about 6.9 m on Earth, 18.2 m on Mars, 41.7 m on the Moon, and about 51 MPa in the liner, which is why the liner, not the printed shell, is the structure.

P, rho = 101325, 1500  # cabin air (Pa), regolith (kg/m3)
for body, g in {'Earth': 9.81, 'Mars': 3.71, 'Moon': 1.62}.items():
    print(body, round(P / (rho * g), 1), 'm of regolith to balance the air')
print('liner stress MPa:', round(P * 4.0 / (2 * 0.004) / 1e6, 1))

Halve the cabin pressure and t* halves with it, because the relation is linear. The design lever is the atmosphere you choose to breathe, not the wall you choose to print.

A note from further down the line

From where this desk writes, the late 2070s, autonomous builders were never the risk in themselves. The risk was the slow habit of accepting their output without being able to reproduce the reasoning. Offices that kept one hand check per project, a load case worked on paper and compared against the machine’s, kept their judgement. Offices that let the check go found out what they had lost only when a result looked plausible and was wrong. The robot at 384,000 km has no choice but to act alone. You still have the choice.

Draw your next section by job, run the four lines above, and check one load case by hand before you trust any toolpath to build it.