Taking a humanoid robot apart means getting through 10,000 to 15,000 components and recovering about four kilograms of neodymium magnet without ruining it. That is roughly double the rare-earth magnet in an electric car. Robert Belt put those numbers in The Robot Report this week, under the heading of recycling. They belong in a different document: the total cost of ownership model a fleet buyer runs before signing anything.
What is inside one, and why it fights back
A humanoid that has worked in a building carries navigation maps of that building, biometric logs, facial recognition recordings, and behavioral patterns of the people around it. All of that has to be destroyed to a standard somebody will sign, so the machine has to be alive enough to wipe before it is dead enough to scrap. The lithium packs keep that window narrow. A pack cut open inside a frame built for motion can go into thermal runaway. Belt notes that stored hydraulic and pneumatic energy turns components into projectiles.
Very little of what survives can be sold on. A carbon-fiber limb that has carried load for three years looks identical to a new one, which is why Belt warns against putting reclaimed frames back into service at all.
The magnets are the only part with real recovery value, and they are the hardest to get at. Crushing a humanoid the way you would crush a car cross-contaminates the neodymium with everything else in the machine, and cross-contaminated rare earth sells as scrap. Getting it out clean takes a trained person with hand tools, working around pinch points and flying fragments. Belt calls the work "a highly technical, high-stakes surgical endeavor." He is working with a recycling firm on humanoids, which is who gets handed the machine at the end.
The decision was made at a design review
Belt's fix is design for recycling: modular cartridges and standardized decoupling joints, a machine built to come apart. That is not a recycling proposal. It is an architecture proposal, and it has to win an argument years before the first unit ships.
Decommissioning is unlike most operating costs in that respect. You can renegotiate a service contract and re-source a supplier. A bonded carbon-fiber assembly stays bonded. Whatever it costs to take the current generation of humanoids apart was set in design reviews that were not about disposal, and recycling expertise arriving later does not move that number much.
Where the cost lands
A capital asset gets depreciated against an assumed salvage value, and a leased asset gets priced against a residual. Either way, a buyer comparing a humanoid to the labor it replaces is running a total cost of ownership model with a terminal line in it. If retiring the machine costs money, that line is negative, and the payback the buyer was shown is wrong in a direction that surfaces in year four, after they have standardized on you.
A demo has no year four in it. Humanoid companies are being judged right now on what the machine does in a video, and no video covers the day it comes out of service.
Humanoids are not a special case here. Any product a customer eventually has to get rid of carries a cost that lands on somebody, and the design decides who. A portable medical device with a sealed battery has the same problem, and so does grid hardware full of coolant. Put the number on your own roadmap: what does it cost to make this thing go away, and has a customer or an investor ever seen that figure?
Dave's take
I don't think the humanoid companies are hiding the number, and it is more likely that nobody has needed it yet. The first invoice for taking one of these apart lands years after the design review that set its size, and that review is the only place the number moves. By the time the invoice arrives, the answer is bonded into every machine already built.
From Dave’s video library
Dave on the same gap in software: shipping something that works is cheap, and changing it later is the bill that arrives afterward.
I’m here to help you scale.
Work With DavePrefer a smaller first step? Book a $500 one-hour working session →
Dave Saunders is the founder of Base Reality Group and a Fractional CPO for product companies. He was a founder and operator at Galen Robotics, where the surgical-robotics platform earned FDA De Novo authorization in 2023, and he managed a 35-patent portfolio licensed from Johns Hopkins. He wrote Founders Who Finish and publishes The Build. More about Dave →