Guide
Tesla Cybercab fleet ownership.
Almost everything written about owning a robotaxi fleet is written about the purchase, which is the part nobody outside Tesla can answer yet. This guide covers that honestly and then spends the rest of its length on the part that is knowable today: what it takes to keep a driverless car earning once you have one.
Written by the team building the operations layer for these fleets. Last reviewed 9 September 2026. We are not affiliated with Tesla, Inc.
Chapter 01
Buying the cars
The purchase is the part with the fewest published facts and the most confident commentary.
Tesla accepts interest from businesses and individuals who want to buy Cybercabs for commercial use, and has said production has started. That is the whole of what is on the record. There is no published price, no delivery window, and no set of owner-operator terms, which means nobody outside Tesla can honestly tell you what a fleet costs to assemble or how quickly it pays for itself.
What that leaves is a decision about readiness rather than a decision about purchase. The people who will move first when terms are published are the ones who already know which markets they want, which vendors they can call on, and what their operating cost per mile has to be for the numbers to work. None of that requires a car to exist.
Treat any figure you see — including ours — as an assumption with a name on it. Our calculator marks the two unpublished inputs explicitly rather than quietly filling them in, because the point of modelling this today is to find out how sensitive your case is to them.
Chapter 02
What operating a driverless fleet actually involves
Ride-hailing worked because the driver absorbed the operations. Remove them and the work does not disappear.
Every ride-hail business in history ran on one unpaid worker. The driver wiped the seat, noticed the warning light, cleared the camera lens, handed back the phone somebody left behind, and decided when the car needed a wash. None of that was on anybody’s cost sheet, and all of it was load-bearing.
Take the driver out and each of those becomes a dispatch: something has to notice, something has to decide, somebody local has to be sent, and somebody has to prove it was done before the car carries another rider. That is the entire operating problem, and it is per-city, per-incident, and unavoidable.
The scale of it is what surprises people. A car earning well is a car that is being sat in constantly, which means it is a car generating cleaning events constantly. The busier your fleet, the more operations work it generates — the opposite of how a software business scales.
Chapter 03
Cleaning and interior condition
The highest-frequency job, with no equivalent in private ownership.
Interior condition is the operating cost with no private-car analogue. A car nobody supervises accumulates spills, crumbs, rubbish and worse, and it cannot tell anyone. It just keeps accepting rides in the state it is in until a rider complains, by which point you have already lost the fare and the rating.
The work splits into three tiers that behave very differently. Routine interior resets are frequent, cheap and best batched at a hub off-peak. Exterior washes matter less for appearance than for keeping camera lenses clear, which is a driving-capability issue rather than a cosmetic one. Biohazard cleanup is rare, expensive, needs a certified vendor and should never sit under an auto-approval ceiling.
The practical consequence is that cleaning cannot be scheduled on a calendar; it has to be triggered by the state of the car. That is harder than it sounds. Tesla’s published fleet telemetry covers location, charge, tyre pressure and vehicle alerts — there is no documented interior-condition field — so today the signal comes from rider reports, from what a vendor finds on the previous job, and from inspection at a staging hub. Direct cabin sensing would be the single most valuable input an operations layer could read, and it is not something we can claim to have.
Chapter 04
Tires, glass and mechanical work
Heavy, high-torque, high-mileage vehicles wear differently from private cars.
A robotaxi covers in a month what a private car covers in a year, and it does it with instant torque and a heavy battery. Tread goes quickly, and summer asphalt makes it worse. Tire pressure telemetry matters more here than anywhere else: a slow leak is a scheduled mobile-tire job, and a rapid loss is a car that needs meeting where it stops.
Glass is the underrated one. A vision-based car with a chipped windshield in front of the camera array is a grounded car. Replacing the glass is the straightforward half; recalibrating the cameras afterwards so the system reads distance correctly is the part that needs a certified technician, and it is the part that decides how long the car is off the road.
Mechanical work is mostly scheduled rather than emergency, but it needs shops that know the platform — suspension, brake-by-wire, powertrain diagnostics. Whether commercially operated cars will carry approved-vendor or warranty conditions is unpublished, and it is the item most likely to reshape what a fleet’s vendor bench has to look like.
Chapter 05
Charging and staging
Where the car sits when it is not earning is an operating decision with a cost attached.
A robotaxi cannot circle the block waiting for a ping. Doing so burns energy and tread for no revenue, so off-peak the car needs somewhere safe to sit — and that somewhere is owned by somebody. Staging is a real line item and a real supply problem, and in dense markets it is one of the harder things to secure.
Charging is the constraint that turns into a stranding risk. State of charge has to be read against distance to a working charger, not against a percentage threshold, because a flat EV cannot be pushed and a rescue charge is cheaper than a tow for everyone involved.
Both of these argue for hubs rather than pure street operation: a lot where cars stage, charge and get cleaned in batches is materially cheaper per car than sending a vendor to three separate kerbsides.
Chapter 06
Insurance and liability
Somebody has to be covered when a stranger unlocks a vehicle you own.
Fleet insurance for commercially operated autonomous vehicles is its own unsettled question, and it is not one an operations layer answers. What an operations layer does have to answer is narrower: who is liable for damage that happens between a vendor unlocking the car and the work being signed off.
There are two workable answers and they trade against each other. Demanding commercial cover from every vendor is safe for the owner and starves the supply side, because the mobile detailer who could clean your car at 3am does not carry garage keepers cover. Backing lighter work with a platform policy widens supply and puts the risk on the platform.
mersul is built for both, with the owner deciding which tier each kind of work may dispatch to. Insured vendors see every category and their own policy is primary. Everyone else is covered per job by a master policy funded from their own payout, and sees cleaning, inspection and triage. The commitment that makes it worth anything is in the terms: no job is dispatched before that policy is in force, and its carrier, limits, deductible and exclusions are published before the first one.
Chapter 07
The software layer
What has to exist between the telemetry and the person holding the vacuum.
The gap is specific. Tesla’s fleet API gives you telemetry, virtual keys and signed vehicle commands. Your city already has detailers, mobile tire vans, glass technicians and tow operators. Something has to turn a signal from the first into a dispatched, verified, paid job with the second, with an audit trail good enough to settle a dispute. Several companies are building into this space, in autonomous fleet operations and in vendor marketplaces, and you should look at them. What we have chosen to build for is narrower: independently owned fleets, where there is no in-house operations team and the vendor bench has to come from the local market.
That is four distinct problems: detection that resolves noisy telemetry into a queue of real incidents; a dispatch decision that ranks local vendors on distance, ETA, reliability and price; access that is scoped tightly enough that granting it to a stranger is reasonable; and verification that closes the job without a human looking at every photo.
It is deliberately not a fleet-tracking dashboard. Anyone can put vehicles on a map. The difficult and valuable part is what happens in the twenty minutes after something goes wrong, and whether it happens at 3am without waking you.
Be ready the day you can buy them.
The purchase terms are Tesla’s to publish. The operating side is not, and it is what decides whether a fleet is worth owning. Join the list and we will tell you where your market sits, and send you every change to the purchase-status page as it happens.