Abstract
Paid Cybercab rides opened in Austin on Friday, September 4, and the Texas registry now shows 45 Cybercabs, every one registered to Tesla. Nobody outside the company can value one from disclosure, because Tesla has published no Cybercab price, no build cost and no operating figures. What it has published is enough to build the frame and see which gaps decide the answer.
This piece works a single car all the way through: what it costs, what it earns, what it costs to run, what the tax code does to it, and what is left. The arithmetic is deliberately simple so a reader can move any assumption and see the answer move. Start from a $30,000 build cost — an assumption, and the piece says so every time it uses it.
The short version. A working Cybercab grosses about $70,000 a year, so four years of fares is $280,000. Against that, $30,000 of car and federal tax at 21%. With no allowance for running the thing, four years leaves $197,500. That is not a valuation, it is a ceiling — and the gap between it and reality is a number Tesla has never disclosed.
Running costs are the whole argument. Charging, cleaning, insurance, depot space, tyres, maintenance and remote supervision are real and undisclosed. At 25% of fares the four-year figure is $142,200; at 40% it is $109,020; at 57% it is $71,000 — the exact point where Tesla would have done just as well selling the car to somebody who lends it back. So the entire question of whether Tesla should operate its own fleet reduces to one unpublished number, and 57% of revenue is the line it has to stay under.
The tax treatment is stranger and smaller than it looks. A Cybercab's EPA filing puts it at 3,113lb curb and 3,730lb GVWR, far under the 6,000lb threshold, so it is a passenger automobile and the luxury-auto caps normally limit year one to $20,300. Tesla escapes them through a separate provision — the statute excludes vehicles used directly in transporting people for hire — and deducts the full $30,000 in year one. But every route deducts the same $30,000 eventually and defers the same $6,300 of tax, so the exception is worth one year of timing on $9,700: $2,037 a car. Over four years it changes the answer by nothing at all. And none of it touches reported earnings, which run on book depreciation regardless.
Table of contents
- Start with the number Tesla has not published
- Step one: what it costs
- Step two: what it earns
- Step three: what it costs to run, and why that is the whole answer
- Step four: the tax, which is stranger than it looks
- Step five: why none of that moves reported earnings
- Putting it together
- The other way to own it: sell the car
- What moves the answer most
- Where this lands in our model
- What to watch
- Primary sources worth reading
1. Start with the number Tesla has not published
There is no Cybercab price. The sub-$30,000 figure people quote is a consumer price Musk has described, not a disclosure, and it was not addressed at the September 3 launch, which was invite-only with no livestream and no Musk appearance. There is no build cost either, no bill of materials, and no unit volume.
So this piece adopts $30,000 as the build cost and says so at every step. It is the single biggest assumption here. The useful thing is that it turns out not to be the assumption that decides the answer — section 9 shows what does.
Two things are disclosed and worth having in hand. Tesla has been running paid robotaxi rides since June 22, 2025, fourteen months. And a registry pull on September 5 shows 45 Cybercabs and 387 Model Ys registered as autonomous in Texas, every one of them Tesla's — the same count our reading of the federal audit file used to bound the "1,000" printed on it. No private owner has been paid a fare.
2. Step one: what it costs
For a Cybercab, $30,000, assumed.
For the cars Tesla is actually running today, the number is derivable. From the June-quarter release: automotive revenue of $20,516M, less $146M of regulatory credits and $364M of leasing, gives a vehicle-sales line of $20,006M across 480,126 deliveries — $41,668 a car. Tesla reported a 16.9% automotive gross margin and said credits contributed 0.6 points, so the cars themselves earned 16.3%: $6,792 of gross profit on $34,876 of cost.
Model 3 and Model Y were 97.4% of those deliveries, so that average is essentially a Model 3/Y average. Tesla publishes no per-model split.
Two things that number is not. It is not profit — the consolidated operating margin last quarter was 1.4%. And it is not a robotaxi cost; the cost of running a car as a taxi appears nowhere in it, which is section 4.
Running total: −$30,000.
3. Step two: what it earns
A working robotaxi grosses about $70,000 a year. That is ours, from August's fleet ladder, and it survived its first real test this month: the first Cybercab receipts cluster at $31 to $33 an occupied hour, and $70,000 a year is $192 a day, about six occupied hours.
The word working is load-bearing. A car earns nothing while it is parked, and nobody outside Tesla has measured how many hours a day these cars are occupied. Six is a middle estimate inside a wide band, not a measurement.
Four years of fares: $280,000.
Running total: +$250,000, before running costs and tax.
4. Step three: what it costs to run, and why that is the whole answer
Here is the line that decides everything, and it is empty.
Charging, cleaning, insurance, depot space, tyres, maintenance, remote supervision and customer support are all real costs of operating a taxi fleet, and Tesla discloses none of them. This is where most analysis picks a plausible cost per mile and carries on. We are not going to — but the shape of the answer can be laid out completely, because the whole range fits on one table.
Four-year net per Cybercab, after tax, at each level of running cost
| Running cost | Per year | Four-year net |
|---|---|---|
| Nothing | $0 | $197,500 |
| 15% of fares | $10,500 | $164,320 |
| 25% of fares | $17,500 | $142,200 |
| 40% of fares | $28,000 | $109,020 |
| 50% of fares | $35,000 | $86,900 |
| 57% — break-even | $40,037 | $70,975 |
Running costs cover charging, cleaning, insurance, depot space, tyres, maintenance and remote supervision. Tesla discloses none of them, so every row here is a scenario rather than an estimate. Each line takes $280,000 of fares over four years, subtracts running costs at that share of fares, subtracts the $30,000 assumed build cost, and subtracts federal tax at 21% with the whole car deducted in year one. The 57% row is the break-even against selling the car to an owner who lends it back at a 30% take, which returns $70,975 after tax. Fares of $70,000 a year and the $30,000 build cost are both R40 assumptions.
Read the bottom row first. At 57% of fares, operating a Cybercab returns exactly what Tesla would have got by selling the car to someone who lends it back — the comparison section 8 works through. So the entire strategic question has a single threshold:
Below 57% of revenue, Tesla should own its fleet. Above it, Tesla should be selling cars to fleet operators.
That is the most useful thing in this piece, and it needs no view on what the true figure is. It is also worth noting that Tesla published a business-to-business interest form on the day of the launch, headed "Help Us Build Our Robotaxi Network," collecting a company name — which is the structure that makes sense if running costs are high.
Chamath Palihapitiya's widely shared "Signed up! Let's build a huge fleet" is a form submission on that page, not a contracted order. No price, volume or delivery date has been published for a Cybercab fleet sale.
For the rest of this piece: 25% of fares, $17,500 a year. An assumption, chosen as a mid-case, not a finding.
5. Step four: the tax, which is stranger than it looks
Three statements circulate about Cybercab depreciation. Each is true; together they give the wrong answer, because a fourth is missing.
Automobiles are five-year property. True, and long-settled.
100% bonus depreciation is back. True — the 2025 tax act made it permanent for qualified property acquired and placed in service after January 19, 2025, covering anything with a recovery period of 20 years or less.
So the whole car comes off in year one. Not by itself, no.
The missing fourth: the luxury-automobile cap. It applies to a passenger automobile even when bonus otherwise would, and for a vehicle first placed in service in 2026 it limits the deduction to $20,300 in year one, then $19,800, $11,900 and $7,160. The usual escape is weight — the rule is about vehicles under 6,000lb. The Cybercab does not escape: its EPA filing puts it at 3,113lb curb and 3,730lb GVWR, barely half the threshold.
What does escape is the use, not the vehicle. The statute defining "passenger automobile" carves out any vehicle used by the taxpayer directly in the trade or business of transporting persons or property for compensation or hire. That was written for taxi and livery fleets, and Tesla collecting fares from the public in Austin is the paradigm case, not an edge one. Outside the definition there is no cap and the full cost comes off in year one.
Year one on a $30,000 Cybercab, US federal rate of 21%
| Year one on $30,000 | Deduction | Tax at 21% |
|---|---|---|
| No bonus, ordinary schedule | $6,000 | $1,260 |
| Bonus, inside the cap | $20,300 | $4,263 |
| Bonus, driven for hire | $30,000 | $6,300 |
| Over the car's whole life | $30,000 | $6,300 |
| What the exception buys | $9,700 | $2,037 |
Without bonus depreciation a five-year asset takes 20% in year one. With bonus but inside the luxury-automobile caps, a vehicle first placed in service in 2026 is limited to $20,300, then $19,800, $11,900 and $7,160; a $30,000 car finishes in two years because only $9,700 remains after year one. With the statutory exception for vehicles used directly in the trade or business of transporting persons for compensation or hire, no cap applies and the whole basis goes in year one. The Cybercab's EPA-filed 3,730lb gross weight is far under the 6,000lb threshold, so the exception is the only route to the full deduction. Every route deducts the same $30,000 and defers the same $6,300, so the exception buys one year of timing on $9,700. Reported earnings run on book depreciation and are unaffected by all three.
Now the part that gets lost. Every route deducts the same $30,000 and defers the same $6,300. Depreciation rules decide when a deduction lands, never how much. Under the caps a $30,000 car finishes in two years — $20,300 then the remaining $9,700 — because the year-two cap of $19,800 is far above what is left. Under the exception it finishes in one.
So the for-hire exception is worth one year of timing on $9,700: $2,037 a car. Over a four-year horizon it changes the final figure by nothing, because both routes have finished long before year four. It is a real cash-flow benefit in year one and it is not a valuation input.
Two more caveats that matter more than the exception does. A deduction only converts to cash against taxable income, and without it becomes a loss carried into a later year. And it is a deferral, not a saving: basis goes to zero, so proceeds on a used robotaxi come back as ordinary-income recapture.
Running total, at 25% operating cost: four-year federal tax of $37,800.
6. Step five: why none of that moves reported earnings
Bonus depreciation is a rule about the tax return. Tesla's reported earnings run on book depreciation, which spreads the car over its useful life regardless of what the tax return did. The two schedules run in parallel; they do not add.
On a $30,000 car over four years, book depreciation is $7,500 a year. That is what hits operating income, in every one of the four years, whichever tax route the car took. The difference between the book and tax schedules sits on the balance sheet as a deferred tax liability, peaking at $4,725 at the end of year one and unwinding to zero by year four.
Scale it to a fleet worth talking about. Ten thousand Cybercabs is a $300 million outlay. Year-one tax deduction: $300M with the exception, $203M under the cap — $63.0M against $42.6M of cash tax, a $20.4 million difference, once. Book depreciation over the same period is $75 million a year and does not move at all.
So the tax angle is a free-cash-flow story, not an earnings story. Anyone modelling an EPS boost from bonus depreciation is counting a deduction the income statement never sees.
7. Putting it together
The table at the top of this piece is the whole walk, and it is short enough to check by hand: $280,000 of fares, less $70,000 of running costs at the 25% mid-case, less the $30,000 car, less $37,800 of federal tax, leaves $142,200 over four years.
The honest way to state that is as a range with its driver named. At no running cost the answer is $197,500; at 25% it is $142,200; at 40% it is $109,020; at 57% it is $71,000 and Tesla should have sold the car. One undisclosed number spans the entire result.
Applying the identical method to a Model 3/Y at its derived $34,876 cost gives $193,648 with no running cost and $138,348 at 25% — close enough to the Cybercab to say something useful: at these fare levels the vehicle's build cost is a second-order input. A $4,876 difference in cost moves a four-year answer by about $4,000. Fares and running costs move it by tens of thousands.
8. The other way to own it: sell the car
Tesla has a second option on every car it builds: sell it, and take a cut of the fares if the buyer enrols. At Autonomy Day on April 22, 2019 Musk said Tesla would take 25% to 30% of revenue from rides on customer-owned cars. That is seven years and four months ago and has never been restated — not at the 2024 shareholder meeting where he described the "Airbnb" model, not on the January 2026 call where he repeated it, not at Thursday's launch.
On the assumed $30,000 cost, a Cybercab sold at Tesla's own 16.3% margin fetches $35,842 and earns $5,842 of gross profit. Add a 30% take on $70,000 of fares for four years — $84,000 — and after tax Tesla keeps $70,975.
That is the number the operating case has to beat, and it is why 57% is the threshold in section 4. Note what makes it low: Tesla gives up 70% of the fares. No depreciation rule anywhere in section 5 is worth a fraction of that. The reason to keep the car is the fare split, not the tax code.
9. What moves the answer most
In descending order, which is the practical output of the whole exercise:
- Running costs. Spans the entire result, from $197,500 to $71,000. Undisclosed.
- Fares. At the $56,000 a year the segment's own reported revenue currently implies, the 25%-cost case falls to $109,020; at the $120,000 our model assumes it rises to $260,700. Ours, and soft.
- The fare split, if Tesla sells. 100% versus 25-30% is the difference between $142,200 and $70,975.
- Discounting. Everything above is undiscounted, which flatters ownership, since the sale price arrives at once and the fares arrive over four years. At 8% the advantage of keeping narrows by roughly a fifth.
- Build cost. $30,000 against $34,876 moves the answer about $4,000. It is the assumption everyone argues about and close to the least important.
- The for-hire tax exception. $2,037 in year one, and zero across four years.
10. Where this lands in our model
Our Tesla model runs robotaxi as a capacity line — vehicles deployed, times utilisation, times revenue per vehicle — with 1,000 cars in commercial service at the June-quarter basis. It lists a network take rate on owner-supplied cars among the segment's revenue streams and carries zero dollars of it, because Tesla has named no take rate since 2019.
The place this piece rubs against it is the fare assumption, and it cuts against us. The model carries $30,000 of revenue per deployed vehicle per quarter — $120,000 a year — where this piece uses $70,000 for a working car. They are not the same denominator, since the model's is per deployed vehicle at 45% utilisation. But the segment's own reported June revenue across 1,000 vehicles is nearer $56,000 a year, so our $70,000 sits between what the line earns and what the model assumes it will. It is the figure a reader should distrust first.
The model also carries no segment-level tax line, so nothing in section 5 has anywhere to land.
11. What to watch
- Any operating-cost disclosure. Cost per mile, per vehicle, or a robotaxi gross margin. It is the one number that turns this range into a valuation.
- A named fleet buyer, with volume and price. The interest form produces leads; a disclosed order would price the sale option directly.
- A take rate on the record. Anything Tesla states for owner-supplied cars replaces a 2019 number and makes that stream modellable.
- Occupied hours per car per day. Fares are the second-biggest driver here and rest entirely on an estimate of utilisation nobody has measured.
- A Cybercab retail price. It would replace the $30,000 assumption everything above is built on — and, per section 9, move the answer less than any other input on this list.
12. Primary sources worth reading
- Tesla's second-quarter 2026 results — the revenue, credits, leasing, delivery and margin lines behind every per-car cost figure.
- The Cybercab's EPA filing — 3,113lb curb, 3,730lb GVWR, the specs that put it inside the luxury-auto cap.
- 26 U.S. Code § 280F — the caps, and at (d)(5)(B) the for-hire exception that removes them.
- Revenue Procedure 2026-15 — the 2026 dollar limits: $20,300, $19,800, $11,900 and $7,160.
Automotive revenue, regulatory credits, leasing, deliveries, gross margin and operating margin are as Tesla reported for the quarter ended June 30, 2026. The Cybercab weight and battery specifications are from its EPA filing of June 2026, and the depreciation caps are the IRS's own. The 25-30% take rate is Musk's, from Autonomy Day on April 22, 2019, never restated; his owner-fleet remarks are from the shareholder meeting of June 13, 2024 and the call of January 28, 2026. The 45 Cybercabs and 387 Model Ys are from a September 5 registry pull. Ours, not Tesla's: the $30,000 Cybercab build cost and the $35,842 price implied from it; the per-car cost and gross profit derived for Model 3/Y; $70,000 of annual fares and every figure built on it; operating cost as a percentage of fares; four-year book depreciation; a 21% federal rate with no state or foreign tax; and the reading of the depreciation statutes, which is not tax advice and is fact-specific.
