Autonomy ends at the depot. Somebody charges the car, wipes the lidar, vacuums the back seat, rotates the tires and recalibrates the sensors after a parking lot scrape. Below is how that work is actually handled across every autonomous fleet running in the United States, what the parts of it cost where anyone has said, and the long list of things no operator in this industry has ever published.
Two models, and the split matters if you are buying vehicles. The largest operator contracts its depot work out metro by metro, awarding each market separately as it opens. The vertically integrated operators keep it in house and carry the cost. Everyone else is somewhere between, and several have never said.
| Fleet | Where it operates | How servicing is handled |
|---|---|---|
| Waymo | 14 metros, including Phoenix, San Francisco, Los Angeles, Austin, Atlanta, Dallas, Nashville, Miami and San Diego | Contracted out, metro by metro |
| Tesla Robotaxi | Austin, Bay Area, and early Texas metros | In house |
| Zoox | Las Vegas, San Francisco | In house |
| Uber AV programs | San Francisco Bay Area, Austin, Atlanta, Dallas. Houston from 2027 | Contracted out |
| Nuro with Lucid | San Francisco Bay Area. Houston from 2027 | Contracted out |
| Motional | Las Vegas | Contracted, status unclear |
| May Mobility | Multiple US metros | Mixed |
| Avride | Dallas | Never disclosed |
| Aurora | Fort Worth, Houston, El Paso, Phoenix | Own terminals |
| Waabi | Dallas area | Own terminal |
| Kodiak | Permian Basin, Georgia | Customer owned and maintained |
Servicing is awarded one metro at a time, to whoever can put a certified bay and a trained technician near where the cars sleep. Nobody is winning this nationally, and nobody has published what any of it costs. That is the gap we built the company to fill, and it is why the numbers on this site come from operating precedent rather than from anyone else's rate card.
Whatever the platform, the same jobs recur. What changes between a Waymo Jaguar, a Zoox pod, a camera-only Tesla and a Class 8 truck is how many of them are hard.
A robotaxi depot draws four to twelve megawatts, according to the charging developers building them. One converted Brooklyn site runs seven megawatts across twenty five stalls, and operators put the requirement at four to six depots per mid sized market to keep dead miles down.
Today a person plugs the cable in. Robotic charging exists but no operator has named a deployment.
The least glamorous and most constant job. Vomit, spilled drinks, drug paraphernalia and worse, escalated to dedicated staff when it is bad enough. Weekend nights run hottest.
No operator publishes a cleaning cadence. Reports of automated cleaning robots at Tesla hubs trace to a single permit filing found by a fan account.
A lidar dome under a bug strike can lose its field of view in seconds. Waymo's Zeekr carries around ten three inch wipers across its camera and lidar housings. Kodiak plumbs water and compressed air to every sensor pod.
One supplier, in an undated interview, puts a bonus fluid bottle at five to ten litres, emptied in as little as four hundred miles. On a robotaxi duty cycle that is a refill every day or two.
Physical drives pulled and swapped, wireless offload, software pushed, mission configuration set, registration checked. This is a large share of what depot staff physically do to each vehicle, and it is invisible from outside.
Heavy electric platforms on a stop and start urban cycle, with regeneration taking load off the friction brakes and putting it into the tires.
No operator has published a wear rate, a rotation interval or a replacement interval on any autonomous platform. Anyone who quotes you one is guessing.
The one that decides whether ordinary shops can touch this work. Calibration is triggered by collision, windshield replacement, sensor removal and, for radar, wheel alignment. It needs a level floor, a controlled ride height, a full tank, a clear bay and nobody in the car.
Kodiak built its whole sensor architecture around avoiding it: pods arrive pre calibrated and swap out in about ten minutes, three bolts and three cables.
Where the money actually goes. In AAA testing in 2023, replacing a sensor bearing side mirror was 71 percent ADAS cost. A single calibration adds about two and a half days to a repair, two or more adds four.
Not one AV operator has named a collision repair partner. Zoox is hiring to own the problem internally. Nobody has said who does the work.
Repositioning, inspections, charging queue management, connectivity troubleshooting, and getting a stalled vehicle off a live street. The depot floor is a logistics problem before it is a mechanical one.
Waymo's San Francisco depot labor runs roughly twenty nine to thirty one dollars an hour through a contractor, per one worker interview in August 2026.
Most numbers circulating about this industry come from vendors or from arithmetic done on two unrelated press releases. These six come from claims data, regulatory filings and peer reviewed work.
Per repairable vehicle, close to double the figure five years earlier. Calibration appeared on under one percent of appraisals in 2017.
CCC Intelligent Solutions, Nov 2025Share of direct repair program estimates including at least one calibration, up from 26.9 percent a year earlier. Over half appear on supplements rather than the first estimate.
CCC Crash Course, Q3 2025Keys to keys goes from 13 days with no calibration to 15.5 with one and past 17 with several. For a fleet, downtime is the real invoice.
CCC Intelligent Solutions, 2025Across 13.8 million California trips and 86 million miles. Deadhead per trip improved from 5.1 miles to 2.8 over the period, so the problem is being worked.
Findings, peer reviewed, May 2026, from CPUC filingsThe binding constraint on this industry is zoning and a path to power, not capital. Four to six depots per mid sized market.
Charging and fleet operators via Axios, Apr 2026Growth in the share of repair estimates carrying an ADAS calibration line, 2025 over 2024, with calibrations per repair up about ten percent.
Enlyte Envision Trends 2026We went looking for each of these specifically. These are not things we failed to find. They are things that do not exist in public, anywhere, from any operator. If a vendor quotes you one of them, ask where it came from.
Not one operator publishes an uptime percentage, an out of service rate, mean time between failures, or hours out of service per vehicle. The figures in circulation are third party trackers sampling the apps from outside, and vendor case studies with the client's name removed. Nothing from an operator.
There is no credible published cost per vehicle per year for AV depot servicing. Every search for one returns investor speculation and calculators built by content farms.
Every fleet servicing agreement disclosed in this industry has been announced without financial terms. Not one vehicle count, rate, or margin is public, including from the listed companies that have signed them.
No operator has published a calibration interval, procedure, bay throughput or annual calibration labor budget. The widely repeated claim that Waymo calibrates on every warehouse exit traces to one venture capital blog post from 2020 with no corroboration since.
No operator publishes dwell time per session, kilowatts per stall, or energy per vehicle per day. Depot level power draw is known only because a charging developer said so out loud.
No operator publishes how often a vehicle is cleaned or what fraction of vehicle hours it costs. The public record is a handful of depot visits and one anonymous worker interview, none of which give a cadence.
Nothing platform specific exists. The nearest adjacent evidence, from electric Class 8 trucking, says plainly that there is not much formal data yet, and attributes what it sees to the powertrain rather than the autonomy.
No AV operator has a named collision partner or certified body network in public, and no post collision sensor recalibration arrangement has been disclosed by anyone. This is the largest white space in the industry.
The pattern in section 01 is the whole argument. Fleet servicing is being awarded metro by metro, to whoever can put a certified bay and a trained technician near the vehicles. That is a network problem, not a vehicle problem, and it does not care whose logo is on the car. If you are looking specifically at Tesla, the Cybercab page has the cost model and our estimates for the terms Tesla has not published.
Charge management, interior and exterior cleaning, sensor surface cleaning and fluid service, pre service inspection, data offload where the operator's stack allows it, and the yard logistics that keep vehicles moving through a bay instead of sitting in one.
Scheduled and unscheduled work on heavy electric platforms running urban duty cycles, through certified partner shops rather than owned depots, so capacity scales with the fleet instead of ahead of it.
Static and dynamic calibration to manufacturer procedure, in bays built to spec, with the prerequisites actually met: level floor, controlled ride height, correct tire size, clear working envelope. Documented every time, because the record is the product.
Coordination from first notice through teardown, repair, recalibration and release, with the calibration captured on the first estimate rather than a supplement. Over half of calibrations currently land on supplements, which is where the time goes.
Per vehicle service history, calibration evidence and inspection records, in the shape regulators and insurers ask for. States are already collecting AV fleet data through the California PUC and the Texas DMV, and that direction only goes one way.
Real estate, power and people are the constraint everyone in section 01 is fighting over. We are assembling the network ahead of the vehicles rather than after them, which is the only way the first metro is ready when a fleet lands in it.
Whether you own vehicles today, operate depots that could take AV work, or are underwriting somebody who does. We will come back with what the servicing actually costs in your market, built from the same sources as this page.