There are more than 13,000 tonnes of human-made material in Earth orbit right now, and the majority of it no longer does anything useful. Dead satellites, spent upper stages, shed panels and fragments — mass that cost, at historical launch prices, tens of thousands of dollars per kilogram to place there, and that even today costs roughly $2,700 per kilogram on a dedicated Falcon 9 and around $7,000 per kilogram on a rideshare to replace. In any other industry, a stockpile like that would be called an ore body. In orbit, we call it debris and plan missions to drag it into the atmosphere to burn.
That is starting to change. A serious research community — spanning NASA, DARPA, ESA, universities, and a wave of startups — is asking a different question: what if the most valuable resource in low Earth orbit is the material that is already there? Analyses of a circular economy for space debris, including work led by researchers at Durham University, have estimated that the recoverable scrap metal in orbit could ultimately be worth hundreds of billions of dollars as feedstock for in-space construction. The numbers are speculative, but the physics behind them is not: mass in orbit has already paid the brutal energy toll of launch. Reusing it is the single largest cost lever in the space economy that nobody has pulled yet.
Space Dust Industries is a Colorado AI company working at the foundation of that future. We have not flown hardware, and we will not pretend otherwise. What we are building — on the ground, today — is the machine intelligence that orbital recycling will demand: perception systems that can identify unknown materials, autonomy that can run industrial processes without a human in the loop, and the simulation and robotics research to tie them together.
The state of the art: a field being born
In-Space Servicing, Assembly, and Manufacturing — ISAM — is the umbrella under which orbital recycling is taking shape, and its recent history is a lesson in how hard this is. NASA’s flagship OSAM-1 mission, which was to robotically refuel a satellite never designed for servicing, was cancelled in 2024 after roughly $1.5 billion in expenditure, undone by cost growth and the lack of a committed partner. But the agency did not abandon the field: it stood up the COSMIC industry consortium to coordinate ISAM development and is contributing its robotic-servicing expertise to DARPA’s program to service satellites in geosynchronous orbit.
Meanwhile, smaller players have been quietly demonstrating the individual pieces. In May 2024, Colorado-based ThinkOrbital operated the first autonomous electron-beam welding system in space, returning the welded samples to Earth for NASA and ESA analysis — proof that real metalworking, not just assembly of prepared parts, is possible in orbit. Australia’s Neumann Space has fired a pulsed cathodic arc thruster on orbit that runs on solid metal propellant, and has shown on the ground that it can burn metal recycled from debris. And CisLunar Industries — headquartered, like us, in Colorado — is developing a Modular Space Foundry to melt and reform scrap metal in microgravity, working with Neumann Space, Astroscale U.S., and Colorado State University under a U.S. Space Force-funded project to close the loop: capture debris, process it into metal rod, and feed it back to a thruster as fuel.
That last idea deserves a pause, because it reframes everything. A dead upper stage is not just scrap — it is propellant. A recycling spacecraft that can eat what it catches can refuel itself from its own targets. The economics of debris removal, which have never closed on their own, start to look very different when every capture is also a fill-up.
Melting metal where there is no “down”
Our concept for solar-thermal debris processing: focused sunlight does the melting, so the scarcest orbital resource — power — is drawn straight from the Sun.
Reprocessing metal in microgravity is genuinely strange. There is no gravity-driven convection to stir a melt, no buoyancy to separate slag from metal, no “pouring” into a mold. Surface tension dominates: molten metal wants to become a floating sphere. Processes that are trivial in a terrestrial foundry — degassing, casting, separating mixed alloys — have to be redesigned around electromagnetic confinement, centrifugal force, or wire- and rod-based forming that never lets the melt float free. The compensation is that microgravity also removes problems: containerless processing avoids crucible contamination, and vacuum is free.
Energy is the other constraint. Melting aluminum takes on the order of a megajoule per kilogram once you account for heating and latent heat, and orbital power is precious. That is why our concept work centers on concentrated solar thermal processing — using lightweight focusing optics to reach melt temperatures directly from sunlight rather than routing everything through photovoltaics and resistive heaters. It is the same logic the legacy of terrestrial solar furnaces suggests: when the process is heat, collect heat.
You cannot recycle what you cannot identify
Spectroscopy plus machine learning: the perception layer that turns anonymous scrap into sorted, certified feedstock.
Here is the problem that gets skipped in most orbital-recycling artwork: a captured piece of debris is an unknown. Is that panel 6061 aluminum or 7075? Is the structure titanium or stainless? Which fragments carry composite overwrap, cadmium plating, or beryllium components that would poison a melt? Terrestrial scrap yards solve this with instruments — X-ray fluorescence guns and laser-induced breakdown spectroscopy (LIBS), increasingly paired with machine-learning classifiers that sort mixed metal streams on fast-moving belts. NASA has flown LIBS twice to Mars on the ChemCam and SuperCam instruments, zapping rocks and reading their composition from the flash. The physics is proven in space; the automation is proven on Earth. What does not yet exist is the marriage of the two: an autonomous system that can survey a tumbling, sun-blasted, degraded object, fuse spectroscopic readings with vision and thermal data, and output a materials manifest a foundry can trust.
That perception problem is squarely an AI problem, and it is where SDI’s research is focused. On our on-premise multi-GPU cluster, we train and evaluate classification models of exactly this kind — learning from spectral and imaging data, quantifying their own uncertainty, and running on power-constrained edge hardware, because a recycling spacecraft will not have a data center and may not have a reliable link to one. Offline, self-sufficient inference is not a compromise in orbit; it is the requirement.
Water: the other recyclable
Metal is only half of the circular economy. The other half is water — and here, uniquely, orbital recycling is not a vision but twenty years of operational fact. The International Space Station’s Environmental Control and Life Support System recovers water from urine, sweat, and cabin humidity, and in 2023 NASA reached a long-sought milestone: 98 percent water recovery. The station’s oxygen comes from solar-powered electrolysis of that recycled water, and a Sabatier reactor closes part of the carbon loop by combining exhaled CO2 with the electrolyzer’s hydrogen byproduct to make still more water. Every future deep-space habitat will inherit this architecture.
Electrolysis also happens to produce the classic high-performance rocket propellant pair — hydrogen and oxygen. Water sourced from recycled stores, and eventually from lunar or asteroid ice, could be split by solar power into propellant wherever it is needed, turning depots and habitats into refineries. Our interest is in the layer that makes such systems trustworthy: predictive maintenance. Regenerative life support is a web of pumps, membranes, sorbent beds, and sensors in which small degradations cascade, and the ISS’s own systems have needed repeated intervention. Machine-learning models that watch thousands of telemetry channels, learn each system’s healthy signature, and flag drift weeks before failure are a proven discipline in terrestrial industry — and a natural extension of the anomaly-detection and time-series work we do on the ground today.
Closing the loop
The end state we are working toward: every output of one process becomes an input to another.
Pull the threads together and a picture emerges. Debris capture feeds a foundry. The foundry feeds manufacturing — and metal-fueled propulsion. Water recycling feeds life support — and electrolytic propellant. Waste heat from processing warms habitats and process loops. Nothing valuable is thrown away, because “away” costs $2,700 a kilogram to replace.
Policy is bending in this direction too. ESA’s Zero Debris Charter — signed by well over 200 companies, institutions, and governments across more than 30 countries — targets the end of new debris generation by 2030, and ESA’s own Clean Space initiative has studied debris recycling as a long-term complement to removal. Sustainability in orbit is moving from aspiration to procurement language, and the operators who can treat end-of-life spacecraft as assets rather than liabilities will own the resulting market.
Honesty requires saying what stands between here and there: no one has yet melted a captured piece of real debris in orbit, microgravity metallurgy is early, and the autonomy to run an industrial plant without human hands does not exist. Those are exactly the gaps we find worth working on.
The Road Ahead
SDI’s contribution to this future is deliberately ground-first. Our on-prem AI research cluster trains the perception and materials-classification models orbital recycling will need. Our 3D scan-to-print pipeline is a terrestrial rehearsal of the digitize-analyze-refabricate loop at the heart of any foundry, orbital or otherwise. Our robotics and reinforcement-learning research targets manipulation and process control under uncertainty — the core competence of an unmanned recycling plant. And our commitment to offline edge AI reflects the reality of where this intelligence must ultimately run: far from any cloud, on its own power, making good decisions alone. The circular economy in orbit will be built by many hands. We are building its mind.