Missed the AI rally?
AutoZone might be next.

At a GXO warehouse in Flowery Branch, Georgia, a two-legged robot named Digit spent most of a year lifting totes off a conveyor. By last November it had moved more than 100,000 of them.

When Digit breaks, Agility Robotics puts an engineer on a plane. That works fine at nine sites. It gets expensive at nine thousand. At nine million it’s a comedy.

Everybody is arguing about how many of these get built. Morgan Stanley says a billion by 2050. Rodney Brooks, who co-founded iRobot and has watched this industry miss deadlines for forty years, says that’s fantasy. Tesla wanted 5,000 Optimus units last year and built a few hundred.

Fine. Argue about that. I want to know who fixes them, because in hardware that’s always where the money ends up, and at the moment the answer is nobody.

What breaks

Not what you’d guess. The instinct is to treat a humanoid as a laptop with legs, which would make it a software problem.

Figure actually ran one. A Figure 02, six months on a BMW line, 1,250 hours, 90,000 parts handled. Then they published the post-mortem. The top hardware failure wasn’t the legs. It wasn’t the battery. It was the forearm: a little board routing signals out to the wrist, and cable that flexed every time the arm moved.

So: things that move, and things that carry signal across a joint. Hands, feet, actuators, wiring. Plus a lithium battery that weighs more than everything near it.

And then the failure nobody puts on a spec sheet. It falls over.

Unitree, which sells these machines today, has already handled this the way any sensible company would. From the list of things its warranty does not cover: "Damage caused by tumble or fall when manipulating robot on level ground."

It’s a bipedal robot. Falling over is your problem.

So where does it go?

Here’s the part I find genuinely funny. The thing can walk. It is the first machine in history that can take itself to the shop.

It also weighs between 66 and 198 pounds depending on whose you bought, and the awkward bit is the battery. A healthy robot can fly, as regulated dangerous goods, in a box that costs real money. A damaged one can’t fly at all. US rules put damaged or defective lithium cells on the ground, full stop. UPS won’t touch them without a special permit and a contract. FedEx just says no. So does the Postal Service.

And Unitree, again, has thought it through:

"Please do not return the seriously damaged battery packs; If you have already sent them back, Unitree Robotics will scrap the battery packs and will not return them."

Don’t mail it. We’ll bin it. Please stop.

Then swap it

Which is the obvious answer, and I think the right one: don’t move the broken robot at all. A van drops a working unit, takes the dead one, and you’re down twenty minutes instead of three weeks. That’s your cable box. That’s Zipcar. That’s most of Amazon.

So the whole thing turns on whether you rent or own, and that is being settled right now, in public.

1X will sell you a Neo for $20,000, or rent you one for $499 a month. Agility’s investor deck prices its subscription near $8,500. Unitree just sells you the machine and wishes you luck.

Rent, and the swap model works beautifully. Own, and you’re standing in your kitchen with a broken robot and a phone number.

The vendors would obviously rather you rented. Recurring revenue, fleet control, no aftermarket. They also need the cash. Agility burned $99.8 million last year and told investors it wasn’t certain it had enough for the next one. Companies in that position sell subscriptions.

So let’s say they win. Everything gets rented, everything gets swapped, and the repair work ends up in six enormous buildings outside Phoenix.

Somebody still has to get a hand actuator to a technician by Tuesday.

Which is a business that already exists

AutoZone, O’Reilly and Advance run something like 17,000 stores in this country, organized around a promise to put a part in a professional’s hands in about thirty minutes. AutoZone alone has 367 hubs and 133 mega hubs, the largest carrying north of 100,000 parts. Nobody is funding a second one of those, ever.

And somebody has noticed. Genuine Parts Company, NAPA’s parent, has spent several years quietly buying industrial automation distributors and folding them into a robotics business. The auto parts trade has already decided this is its adjacency. It just hasn’t run the idea through its stores.

Meanwhile the shops have been accidentally rehearsing for a decade. Replace a windshield on a late-model car and you’ve moved the camera mounted behind it, so the car is now slightly wrong about where the lane is. Fixing that takes a level floor, controlled light, clear space and targets at measured distances. A humanoid that’s had an arm replaced has precisely that problem.

High-voltage work is the same story with the nouns swapped. ASE certifies technicians on evaluating a battery pack, isolating and de-energizing the system, protective gear, and assessing a machine that’s been in a crash. Read that list again with a robot in mind.

Almost nobody holds it.

And the scarcest thing in the building isn’t a lift or a scan tool. It’s the zoning. A repair shop is legal permission to do loud, dirty, heavy work at a specific address near where people live, and cities have been clawing that permission back for years. You can’t conjure it onto a retail lot because you’d like to.

If you wanted a hundred robot service centers across the metros where the machines will actually be, the binding constraint wouldn’t be equipment. It wouldn’t be staff. It’d be finding a hundred addresses a city will allow.

Those addresses exist. Somebody is already standing in them.

The law nobody has written

One more thing, for anyone reasoning from the car analogy.

The independent repair industry wasn’t created by the market. It was created by a 1975 warranty statute and a 1992 Supreme Court decision about Kodak refusing to sell parts to independents. The competitive aftermarket Americans treat as the natural order of things is a legal artifact, and a fairly recent one.

There’s no robotics version of either. Parts pairing, software locks, withheld diagnostics, warranty terms that die the second someone else opens the machine. All available today, none of it regulated. Unitree already voids your warranty for "unauthorized disassembly, and shell opening."

So the real question isn’t whether an independent robot repair business is possible. It’s whether anyone gets there before the door shuts.

Obligatory cold water

Roughly 17,000 humanoids sold worldwide last year, by my arithmetic, off Unitree’s own listing prospectus. Spread that across American metro areas and no city has enough of them to keep a single bay busy. Everything above is an argument about a threshold nobody has crossed.

I’m fine with that. Threshold arguments are how you get anywhere early.

Three things to watch

• A manufacturer publishes a real service manual. This one has started. Unitree puts out screw-level disassembly guides for its hands, and Boston Dynamics says every limb on Atlas swaps in under five minutes, with "training and certification paths" coming. Watch whether those paths ever open to anyone outside the customer’s own staff.
• A parts distributor stocks an actuator. Not a press release about robotics. A part number, in a hub, with a delivery time attached.
• An insurer starts writing protection plans on robots. That’s the Asurion move. uBreakiFix is owned by an insurance company, because a repair network is really the delivery mechanism for a financial product.

I don’t know which way this goes. But there are a lot of buildings out there with three-phase power, level floors, calibration targets, a parts truck that turns up in half an hour, and permission from the city to do heavy work near a school.

And there’s an industry a few years away from needing all of it, currently planning to build the whole thing from scratch.