Small ones already do, under supervision from Earth. Uncrewed capsules and satellites have made crystals and run furnaces in orbit. But every one of them used raw material launched from Earth, made small batches, and was designed and checked by people. No factory has yet made useful parts from local material without people stepping in.
What has been made off Earth so far
Manufacturing in space has gone from one-off experiments on the ISS to a handful of companies flying dedicated hardware. Here are the main milestones as of October 2026.
| When | What | Crew needed? |
|---|---|---|
| 2014 | First 3D-printed part made in space: a plastic part made on the ISS with a printer built by Made In Space (now part of Redwire). | Yes, astronauts installed and handled the printer and parts. |
| Feb 2024 | Varda Space Industries' first capsule, W-1, returns to Earth after growing crystals of the HIV drug ritonavir in orbit. | No. Uncrewed capsule, operated from the ground. |
| Aug 2024 | ESA's metal 3D printer, built by an Airbus-led team, finishes the first metal part made on the ISS. It melts stainless-steel wire with a laser, a few hours a day, partly because of noise limits on the station. | Partly. Astronauts installed it and retrieve the samples. Printing is controlled from the ground. |
| 2025–26 | Varda recovers its W-2, W-3 and W-5 capsules in Australia. W-4's reentry was called off and the capsule was never recovered. | No. |
| Dec 2025 – Jan 2026 | UK company Space Forge's ForgeStar-1 satellite switches on its furnace in orbit and then generates plasma, a step toward growing semiconductor crystals in microgravity. | No. Free-flying satellite, operated from the ground. |
The trend is clear: the newest manufacturing hardware flies without a crew. That isn't mainly about autonomy for its own sake. Astronaut time on the ISS is scarce and expensive, a crewed station vibrates and shakes, and a free-flying satellite can offer cleaner microgravity and can run all the time.
Uncrewed isn't the same as autonomous
A capsule with no one aboard is still mostly run from Earth. Engineers write each process step in advance, watch the telemetry and step in when something looks wrong. That works in low Earth orbit, where the signal delay is a fraction of a second and a ground station passes overhead several times a day.
It works less well further out. A command to the Moon and back takes about 2.6 seconds, which is fine for supervising slow work but not for steering a delicate process in real time. Mars is 6 to 45 minutes round trip. (Our explainer on robots and the first Moon base has a demo that shows what that delay feels like.) A factory on Mars would have to handle most faults on its own, because by the time people on Earth see a problem, it's long over.
It helps to think of autonomy as a ladder:
- 1TeleoperatedA person controls each action, live. Only practical close to Earth.
- 2Scripted and supervised TodayThe machine runs a pre-planned process. People on Earth watch and intervene. Varda, Space Forge and the ISS metal printer work roughly this way.
- 3AdaptiveThe machine inspects its own output and adjusts the process: fixing a bad layer, changing a furnace temperature. People set goals, not steps.
- 4Local materialsThe factory takes in raw local material, such as regolith, whose quality varies from scoop to scoop, and still makes parts that meet spec.
- 5Self-maintainingThe factory repairs and replaces its own worn parts. This is the step that leads toward self-replication.
Today's hardware sits at level 2. Level 3 is mostly a software and sensing problem, and it's where better AI should help first. Levels 4 and 5 are mainly hardware problems.
What makes the next levels hard
Feedstock. Every factory in space so far has used carefully prepared material launched from Earth: steel wire, chemical solutions, semiconductor source material. Lunar regolith varies in grain size and mineral content from place to place, and it is abrasive. A process that works with lab-grade input may fail on a bucket of real soil.
Checking quality. On Earth, parts are measured, X-rayed and stress-tested, often by people. A factory far from Earth needs to do that itself and decide what to throw away. A part that fails inspection on the Moon can't be returned to the supplier.
Wear and repair. Furnaces, lasers, nozzles and bearings all wear out. On the ISS, an astronaut can swap a part. In a free-flying satellite or a lunar factory, there's nobody to do that, so the machine has to be built to last, to swap modules robotically, or to accept that it will stop when one part fails.
Power and heat. Melting metal or growing crystals takes a lot of energy, and in a vacuum the only way to get rid of waste heat is by radiating it. On the Moon, most sites go about two weeks without sunlight. That's one reason NASA's LEIA program pairs in-space manufacturing with power topics such as vertical solar arrays and radioisotope generators.
Where better AI comes in
The jump from level 2 to level 3 is the one AI can most directly speed up. Inspecting a part with cameras and sensors, spotting a fault mid-process and adjusting settings is the kind of task where machine learning already helps in factories on Earth. More capable AI could also help design processes that work with variable local materials instead of perfect feedstock.
AI doesn't remove the physical limits, though. A smarter controller can't fix a cracked furnace lining or produce a spare bearing that the factory can't make. For factories off Earth, the hardware is likely to set the pace for a long time.
Uncrewed manufacturing in orbit: working now, for small, high-value batches using material from Earth, supervised from the ground.
Factories that run themselves on local material: not yet demonstrated. Getting there needs reliable processing of raw regolith, built-in quality checking, and machines that can be repaired without people. The next few years should show the first serious attempts, mostly on Earth-based test rigs first.
What to watch
- NASA LEIA awards: which in-space manufacturing and lunar oxygen proposals NASA funds, and how far their prototypes get.
- Space Forge: whether ForgeStar satellites move from generating plasma to growing usable semiconductor material, and whether the satellites can be returned and reused.
- Varda: how often its capsules fly and return. Frequent, reliable return trips are what turn experiments into production.
- Any factory test on the lunar surface. Until then, everything about manufacturing from regolith rests on Earth-based tests with simulated soil.
Sources and further reading
- ESA, "ESA 3D prints first metal part on the International Space Station". esa.int
- Airbus, "Behind the scenes of the first metal part to be 3D-printed aboard the ISS", September 2024. airbus.com
- Space.com, "Space Forge fires up first commercial semiconductor factory in space". space.com
- Wikipedia, "Space Forge" and "Varda Space Industries". wikipedia.org
- Via Satellite, "Varda space capsule successfully returns to Earth after hypersonic reentry test", May 2025. satellitetoday.com
- NASA, "NASA Calls for Proposals to Accelerate Lunar Surface Technologies", September 2026. nasa.gov