The AVxLink Dispatch
← All researchWhat does it actually cost to keep a robotaxi on the road for a year?
What does it actually cost to keep one robotaxi on the road for a year?
By Andy Shelley -
The AVxLink Dispatch
Issue #2 | October 2026
Self-driving isn't self-servicing.
The industry is racing on cars and miles. Tesla's Cybercabs are already carrying riders in Austin. Dallas is next. ARK Invest's October model puts about 8.1 million robotaxis on Tesla's network by 2031.
Those are supply stories. The question owners will ask is quieter and harder:
What does it actually cost to keep one robotaxi on the road for a year?
What we can and can't know yet
Honesty first.
Tesla has not published a Cybercab purchase price. It has not published owner terms. It has not published a network take rate for independent operators.
Remount's Jonas Lamis notes the same gap this week: Remount models a car at about $30,000; ARK's workbook uses about $25,000 as an average Cybercab ASP. Both are assumptions. Tesla has published neither price nor owner terms.
On Wednesday's Q3 call, Tesla put dates on the next steps — Cybercabs in Austin since September 3, nearly two million unsupervised miles by the end of September, Dallas "next month," customer Cybercabs targeted by year-end, and first customer vehicles on the robotaxi fleet around Q2 2027. Useful context. Still not a cost sheet.
So this issue is a labeled cost stack, not a forecast. Where a number is illustrative, we say so.
One car, one year: the assumptions
We frame this around an unsupervised EV robotaxi — Cybercab-first, with a note where Model Y numbers would run higher.
Utilization (assumption): about 40,000 / 65,000 / 90,000 total miles per year (low / base / high). That sits inside ARK's published robotaxi-miles range (30,000–100,000; workbook median near 65,000).
Vehicle CapEx (separate from OpEx): Remount ~$30,000; ARK ~$25,000. Straight-line over five years with $0 residual — Remount's framing — is roughly $5,000–$6,000 a year of capital recovery. We keep that out of the operating stack below.
Network take: unknown. Ride-hail platforms today often take roughly 20–30%. Tesla has not published an owner split, so we do not bake a take rate into OpEx.
The annual OpEx stack
Figures are annual cash costs for one vehicle. Ranges are low / base / high.
| Line | Low | Base | High | Notes |
|---|---|---|---|---|
| Energy / charging | ~$500 | ~$1,200 | ~$2,900 | Cybercab EPA ~0.165 kWh/mi; we use ~0.18 kWh/mi urban. Depot-ish power ~$0.08–$0.10/kWh; higher with more public DCFC. Model Y efficiency is worse. |
| Cleaning / interior resets | ~$2,600 | ~$6,200 | ~$14,300 | No driver to tidy between rides. Illustrative 2–5 resets/week at ~$25–$55 each. |
| Tires | ~$1,400 | ~$2,600 | ~$4,500 | High-utilization EV wear; ~$0.035–$0.05 per driven mile. |
| Maintenance / repairs / glass / consumables | ~$1,000 | ~$2,000 | ~$4,100 | Filters, fluids, alignment, glass, wear items; ~$0.025–$0.045/mi. |
| Insurance | ~$2,400 | ~$3,600 | ~$6,000 | Proxy. Texas commercial rideshare quotes ~$145–$320/mo; no published owner-Cybercab AV rate. |
| Parking / staging / overnight | ~$600 | ~$1,200 | ~$2,400 | Off-street stall or shared yard share; ~$50–$200/mo. |
| Charging access / depot facilities share | ~$400 | ~$1,000 | ~$2,400 | Amortized plug + site OpEx share — not the full depot CapEx. |
| Compliance / registration / permits / inspections | ~$200 | ~$400 | ~$800 | Texas passenger registration is modest; add inspection and early-market compliance work. |
| Recovery / roadside / tow readiness | ~$200 | ~$600 | ~$1,500 | Cybercab ERG: transport with all wheels off the ground (roll-back / flatbed). |
| OpEx total | ~$9k | ~$19k | ~$39k | Rounded. |
Base case in one line: roughly $19,000 a year to operate one busy robotaxi — about $0.29 per driven mile at 65,000 miles — before buying the car, before financing, and before any Tesla network cut.
Add capital recovery and you're in the mid-$20,000s of annual economic cost on a $25k–$30k vehicle written off over five years with no residual. Change utilization, cleaning frequency, or insurance, and the total moves fast.
Where the money goes
In our base stack, the split is roughly:
- Cleaning ~33% — the largest line, and the one with no private-car equivalent
- Insurance ~19% — still a proxy until AV owner products exist in the open
- Tires ~14%
- Maintenance / repairs / glass ~10%
- Staging + depot/charger share ~11% combined
- Energy ~6% — Cybercab efficiency helps; this line is small if you charge at depot rates
- Recovery + compliance ~5% combined
Energy is not the story most owners expect. Labor-adjacent service is.
What scales vs. what stays stubborn
Some costs should compress with volume: energy rates at shared depots, parts and tire purchasing, insurance pools as actuarial history accumulates, and shared charger utilization. ARK's workbook also shows platform COGS and overhead per mile declining over time in its ride-hail tables — that's ARK's model, not a Tesla disclosure.
Other costs stay local:
- Cleaning labor — someone still resets the cabin where the car is
- Tires — miles are miles
- Tow / recovery — especially with Cybercab's all-wheels-off-ground rule
- Deadhead to distant depots — empty miles burn energy, time, and earnings
That last point is Issue #1 in cost form. A facility for around 100 vehicles can still mean $600,000 to $1.5 million+ before land. Spread across cars, that CapEx can be manageable. It does not fix geography: sending a car across town for a five-minute clean is still expensive.
Why a distributed service network matters
The costs that shrink with scale are mostly industrial. The costs that stay stubborn are mostly local.
That is the wedge we care about at AVxLink: connecting autonomous fleets with qualified local providers — tire shops, detailers, parking, charging, recovery partners — so service happens closer to where the cars already are.
We're early. The industry is early. This stack will get rewritten the day Tesla publishes owner terms — and again the first time an insurer posts a real Cybercab commercial rate.
Until then, the useful work is making the unknowns visible and getting the stubborn lines ready.
Let's compare notes
If you're modeling a Cybercab or Model Y robotaxi P&L, we'd like to see which lines you're stressing. Cleaning frequency? Insurance? Deadhead? Recovery partners who can flatbed a wheel-less two-seater?
We're not asking anyone to buy into a finished product. We're comparing notes while the numbers are still soft.
Hit reply or start a conversation with us.
Next issue: Who owns the car vs. who owns the service — and what a recovery-partner checklist looks like when the Emergency Response Guide says every wheel has to leave the ground.
There's a lot to figure out.
Andy Shelley
Founder, AVxLink
avxlink.com
Self-driving isn't self-servicing.