A framework, not a forecast
Driverless ride economics come down to a short list of variables. Most of them can be observed today. The handful that decide the answer cannot. This page walks the list, says what moves each line, and is straight about where the blanks are.
This page is general information about how the economics work. It is not investment, tax or financial advice.
Start here
Four inputs decide the answer and not one of them is settled: what insurance costs for a driverless mile, which cities grant permission and on what terms, what a worn autonomous vehicle is worth once its duty cycle ends, and how much of the day a vehicle actually spends carrying someone. Any figure published today is a guess wearing a suit.
A model with four open inputs is not a forecast. It is a sentence with four blanks in it, and whoever fills those blanks controls the result. Two careful people can build the identical spreadsheet, make no arithmetic error between them, and finish a world apart.
The blanks are not even independent, which is the part that gets missed. A city that narrows where vehicles may operate shrinks the area a fleet can serve, which lowers utilisation and lengthens the empty drive between rides. A poor safety record raises the insurance line and lands on residual value at the same moment, because the people who would buy used autonomous vehicles are reading the same claims history the insurer is. Move one variable here and you have usually moved two more without touching them.
So this page does the thing that survives contact with reality. It names the lines, explains what pushes each one up or down, and marks plainly which ones nobody can fill in yet. For the wider background this page takes as read, how driverless ride services are run and where they operate, the cluster's main explainer is Robotaxi Hub.
The revenue side
Revenue on a driverless vehicle is not a fare. It is a fare multiplied by how much of the time the vehicle is doing paid work, and the second half of that sentence swamps the first.
Utilisation is the share of a vehicle's available hours spent actually carrying someone. It dominates because almost every other line is charged against the vehicle whether or not anyone is aboard. Insurance runs while it sits. Depreciation runs while it sits. The depot bay is rented while it sits. Lift utilisation and the same fixed stack is spread across more paid work, which is arithmetic rather than optimism. This is why the ordering matters: a modest gain in utilisation outweighs a large saving on almost any single cost line, and it is also why a model that treats utilisation as a rounding detail is not a model.
Deadhead is every mile driven with no passenger in the vehicle: repositioning towards the next request, drifting to where demand is expected, driving to a charger, driving back from one. Those miles burn energy, tyres and duty cycle and produce nothing at the other end. A human driver decides for themselves when to move and when to sit. A fleet decides for every vehicle at once, in software, which turns deadhead into a dispatch policy rather than a habit, and therefore into something an operator can be measurably good or bad at.
Ride demand is not spread evenly across the day. It stacks up around commute hours, again late at night at weekends, and thins out in between. That hands a fleet the problem airlines and hotels have lived with for decades: size for the peak and much of the fleet stands idle off peak, size for the average and you turn away the busiest hours of the week. The shape of that curve in a specific city, rather than the total number of rides in it, is what decides how much of a fleet can be kept busy.
With nobody being paid per trip, the cost of one more ride on a vehicle already out on the road falls towards energy plus wear. That is a floor, and floors get tested. Once more than one fleet serves the same city, price is the most obvious thing left to compete on, and the fare a rider pays drifts towards that floor. Anyone assuming today's fare holds for years is quietly assuming the outcome of that fight.
The cost side
Costs on a driverless vehicle fall into three kinds: the ones charged per mile, the ones that run per hour whether or not the vehicle moves, and the ones that arrive in a lump when something breaks. Blending those three together is the most common fault in a homemade model, because it hides the fact that idle time is expensive.
Energy cost is consumption per mile multiplied by the price of electricity, and the second half moves through the day. Commercial power is not a single price: it is cheaper at some hours than others, and a site's bill can also carry a charge set by the highest rate of draw it reached that month rather than the total it took. Fast charging does not change how many kilowatt hours a vehicle needs. It changes when it can buy them, because a vehicle that fills quickly can wait for a cheap window, and a slow one has to take whichever hour it is standing in.
Dwell is the charging itself. Turnaround is everything else between one rider getting out and the next getting in: queueing for a stall, moving onto it, plugging, unplugging, pulling back out, and any cleaning squeezed in between. Turnaround is frequently the bigger of the two, and it is far more fixable, because it is a site layout and dispatch problem rather than a battery problem.
There is no driver in the seat to spot a spill, decline a passenger or ask somebody not to eat. Detection happens after the event, usually by camera, and the person who reports the problem is often the next rider, who is already sitting in it. Each incident creates two costs rather than one: cleaning the vehicle, and the trip that had to be cancelled or refunded around it.
A driverless vehicle carries sensors that must stay pointed exactly where the software believes they point. A kerb strike, a windscreen replacement or a light collision can mean recalibration before the vehicle is allowed back into service. That is a line a conventional vehicle does not have at all, and the expensive part is not the labour, it is that the vehicle is standing in a bay instead of working while it happens.
Insurance is priced from loss history, and driverless miles have a short one, so cover has to be priced on thin evidence that will keep moving in one direction or the other. There is a structural difference too: with no driver, liability tends to sit with the operator and the manufacturer rather than with an individual behind the wheel, which changes who buys the policy, who defends a claim, and how long a claim takes to settle.
These are two costs, not one, and the second binds harder. Land can be rented on ordinary commercial terms. Electrical capacity has to be delivered by a utility on a utility's timeline, and a depot can easily hold more vehicles than its electrical service can charge. When somebody says a charging site is full, they usually mean the power is spoken for, not that the parking has run out.
Vehicles that get confused need a person somewhere who can advise them. The cost driver is not the wage, it is how many vehicles one person can cover, and that is not a fixed figure: it depends on how often the vehicles ask for help. As the software improves, the same staff cover more vehicles and the cost carried by each vehicle falls without anybody being paid less. It is one of the few lines on this page that should get cheaper on its own.
The vehicle is being consumed the entire time, on two clocks at once: miles driven and years elapsed. A third clock runs beside them, which is whether the sensor and compute generation on board is still supported. A vehicle can be mechanically sound and still be finished as a working asset, because the stack it carries is no longer maintained. That third clock is the one people leave out.
The three that decide it
Everything above can be estimated from things we already watch vehicles and depots do. These three cannot. They carry the widest range of possible values and the least evidence, which is the worst pairing a model can have.
Nobody can quote a settled price for covering vehicles with no driver at scale, because the loss history that would justify one is still being written. The signal to watch is not a headline, it is the moment insurers publish standing terms for driverless operation instead of negotiating each fleet separately. When cover becomes something you shop for rather than something you argue for, this stops being an unknown and becomes a line item.
Permission is neither national nor permanent. It is granted jurisdiction by jurisdiction, usually with conditions attached about where a vehicle may travel, at which hours, in what weather, and whether a person has to be aboard. Those conditions are economic instructions in disguise. A tighter operating area and a shorter permitted day both cut straight into utilisation, which is the term that swamps everything else. Read the conditions, not the announcement.
There is barely a second-hand market for high-mileage autonomous vehicles yet, so what one is worth when its working life ends is the least evidenced figure in the model, and it sits right at the end, where an error has the whole duty cycle to compound. We keep that question in its own place: what a used driverless vehicle is actually worth depends on who the buyer is, and today it is not obvious there is one.
Why site design is an economics question
Charging is the one line that sits on both sides of the page. The minutes spent taking energy are minutes not spent carrying anyone, so the layout of a charging site stops being a facilities matter and becomes part of the operating model.
Three things about a site move that number, and only one of them is the charging speed.
That is why a site built as somewhere a vehicle stops, with somewhere for a person to be while it does, is a different proposition from a row of posts in a car park. EV Cafe is being built in Anaheim on exactly that reading, and it is a useful thing to hold this checklist against. It sits in central Orange County right off Interstate 5, close to the SR-57 junction, which answers the deadhead question. It is planned with 72 Superchargers, which answers the stall availability question. It is planned for charging up to 500kW and is open to all EVs rather than one brand, which answers the rate question. It is due to open early 2027.
Notify list
One message when the Anaheim site opens its doors. That is the entire promise, and it is the only thing this form does. No newsletter, no sequence, nothing else attached.
This page is general information about how the economics work. It is not investment, tax or financial advice.
Straight answers
Not today, and not from anyone. There is no retail route to owning a vehicle you can dispatch driverless: the vehicles running now are operated by the companies that build or license them, and permission to run without a driver is granted to an operator under conditions, not attached to a vehicle when it is sold. If a scheme is offered to you that says otherwise, the thing to ask for is the operating permission it runs under and whose name is on it.
We do not publish figures, and the reason is specific rather than cautious. The largest lines in the model, insurance for driverless operation and the value of the vehicle at the end of its duty cycle, have no settled market price for this use yet. A cost per mile built on those two would be an invention presented as research, and it would date badly. What travels better is the list of lines and what moves each one, which is what this page is.
Four things, in rough order of how much they will move the picture: insurers publishing standing terms for driverless operation rather than negotiating one-off arrangements; the conditions attached to city and state permits, particularly limits on operating area and hours; the first real sales of high-mileage autonomous vehicles, which is when residual value stops being an assumption; and any operator disclosure of how much of the day its vehicles are occupied.
Because time spent taking energy is time not spent carrying anyone, and distance to the charger is deadhead in both directions. That makes stall availability, site location and the rate a vehicle can actually accept into economic variables rather than facilities details. It is the one part of this list a site operator can change directly.
No. This page is general information about how the economics of driverless ride services are put together. It is not investment, tax or financial advice, it is not an offer of anything, and it does not suggest that any outcome is available to any reader. If you are weighing a real commitment, take it to a professional who can look at your circumstances.