Tesla built a $25,000 robotaxi to win on hardware cost. The model says hardware cost is not what decides this — and the number Tesla optimized is worth about a seventh of the number it didn't.
Per vehicle per year. 110% pre-tax ROIC, 0.9-year payback — if the operating assumptions hold.
Against a $0.20 target that requires supervision ratios nobody has demonstrated.
Per mile vs. Waymo's Gen-6 capex. Real, but ~13% of a fare — not a moat.
Of 420 robotaxis registered in Texas as of 2 Sept. The rest are Model Ys.
Tesla has spent two years marketing the Cybercab on a single number: a sub-$30,000 purpose-built two-seater against Waymo's $80,000–$150,000 sensor-laden Jaguars and Zeekrs. It is a genuinely impressive manufacturing result. It is also, in the arithmetic of a fleet, close to a rounding error.
Amortize a $28,000 vehicle over a 350,000-mile service life and you get $0.08 per mile of depreciation. Do the same for a $90,000 Waymo Gen-6 and you get $0.257. The entire hardware advantage — the thing the unboxed factory, the wireless charging, and the deleted steering wheel were all in service of — is worth $0.177 per mile. Against a fare in the $1.10–$1.40 range, that is about 13% of revenue.
Worth having. Not worth confusing for a business model. The costs that actually decide whether a robotaxi fleet earns its capital back sit somewhere else entirely.
| Line | Bear | Base | Bull |
|---|---|---|---|
| Total miles/day | 84 | 137 | 183 |
| Rides/day | 12.0 | 19.1 | 25.0 |
| Net revenue | $96.90 | $179.10 | $201.88 |
| Revenue $/mi | $1.154 | $1.312 | $1.103 |
| Depreciation | $10.75 | $10.92 | $9.15 |
| Energy | $4.62 | $6.14 | $6.95 |
| Maintenance + tires | $6.30 | $7.51 | $7.32 |
| Insurance | $11.76 | $12.29 | $9.15 |
| Remote supervision | $115.20 | $38.40 | $12.80 |
| Depot + software | $22.00 | $16.00 | $11.00 |
| Total cost | $170.63 | $91.25 | $56.37 |
| Cost $/mi | $2.031 | $0.669 | $0.308 |
| Margin/day | −$73.73 | $87.84 | $145.50 |
| Margin/year | −$25,806 | $30,745 | $50,925 |
| Pre-tax ROIC on vehicle | −80.6% | 109.8% | 203.7% |
Supervision cost = (loaded operator hourly × service hours) ÷ vehicles per operator. Bear 1:5, Base 1:15, Bull 1:50, all at $32/hr loaded. Revenue is net of a 5% take for payment processing and airport/city fees.
Lay the six cost lines side by side on a shared scale and the structure of the problem stops being arguable. In the bear case, remote supervision is larger than every other cost combined — by a factor of three. Even in the base case it is the single biggest line, bigger than depreciation and insurance together.
Scenario identity is carried by the row labels, not by color. Color marks only which line decides the outcome.
The Cybercab's economics are not gated on building a cheap car. They are gated on how many cars one human can watch at once.
Waymo runs roughly one remote assistance operator per three vehicles. At that ratio, run the grid below and read the left-hand column: no achievable utilization rescues the unit until you reach about 160 paid miles per day — above what Waymo, the most mature operator in the world, does today. Move one column at a time to the right and the entire surface flips from red to green somewhere between 1:5 and 1:15.
That is the whole thesis in one picture. Tesla optimized the manufacturing axis. Solvency lives on the autonomy-reliability axis.
Vehicles per remote operator →
Waymo operates near 1:3 today. Tesla has not published a ratio. The base case assumes 1:15 — an improvement of 5× on the industry's current best, and the load-bearing assumption in this entire model.
Musk stated $0.20/mile at the Cybercab unveiling; ARK's Big Ideas 2026 models the same figure at scale by 2030, against $0.40 for Waymo's Gen-6 fleet. TD Cowen's Itay Michaeli is more conservative at roughly $0.30.
Reaching $0.20 requires stacking every favorable assumption simultaneously — and even then you have to push supervision past 1:100 and utilization past anything demonstrated in the field.
All three rungs also assume 20% deadhead, $0.03–0.04/mi maintenance, and depot + software costs cut roughly in half from base. Vehicle life beyond 500,000 miles has never been validated on a 4680 pack.
Treat $0.20 as a 2030-plus asymptote, not a plan. The honest near-term number, if supervision improves as hoped, is somewhere between $0.45 and $0.70.
Tesla's Austin fare — $3.00 base plus $1.40/mile — is currently quoted above Uber on comparable routes: $19.58 against $12.96 in one sampling. That is not a durable position. Robotaxis have no structural cost floor protecting them from a price war with a human-driver network whose supply is already amortized and whose drivers absorb their own vehicle costs.
A third off the fare takes roughly half the margin. The cost structure has no answer to it — every line except supervision is close to physically floored.
Run deadhead — unpaid repositioning miles — from 15% to 50% in the base case and cost per mile falls from $0.72 to $0.62, while annual margin drops from $32,233 to $28,760. Fixed costs simply spread across more unpaid miles.
So the headline metric improves as the business gets worse. Anyone quoting a cost-per-mile figure without stating a deadhead assumption is quoting a number that can be gamed by driving emptier — a caveat worth applying directly to the $0.20 claim, which has never shipped with one.
These are the costs separating attractive vehicle-level ROIC from an attractive business. The base case earns 110% on the vehicle. It does not yet earn anything on the company.
Everything above prices the integrated operator — Tesla, or a fleet owner running its own cars. A separate model circulates for the owner-operator: an individual buys a Cybercab, puts it on Tesla's network, and pays a platform fee. Both models can be true simultaneously. They are two sides of one transaction, and reconciling them turns out to be more revealing than either alone.
The circulating version assumes 250 miles/day, a $0.65/mile fare, and a 30% Tesla take, and concludes roughly $1,580/month with an 18-month payback. A widely-shared critique corrects the charging and tire assumptions down to $863/month and 35 months. Both are wrong, in opposite directions.
| Line | As circulated | Critique | Corrected |
|---|---|---|---|
| Gross revenue | $4,875 | $4,875 | $4,875 |
| Platform fee 30% | −$1,462 | −$1,462 | −$1,462 |
| Insurance | −$750 | −$750 | −$750 |
| Depot, parking, cleaning | −$650 | −$1,100 | −$650 |
| Tires | in depot | in depot | −$356 |
| Maintenance | in depot | in depot | −$225 |
| Electricity | −$180 | −$450 | −$338 |
| Depreciation | −$250 | −$250 | −$750 |
| Total cost | −$3,292 | −$4,012 | −$4,531 |
| Net / month | $1,583 | $863 | $344 |
| CapEx breakeven | 19 mo | 35 mo | 87 mo (7.3 yr) |
Depreciation is the error both prior passes share: $30,000 over a 300,000-mile life is $0.10/mile, or $750/month at 7,500 miles — not $250. The $250 figure implies a 900,000-mile vehicle life, three times the one stated alongside it.
Tires are the clearest case. The critique assumes a replacement set every four to five weeks, implying roughly 7,500-mile tire life. Robotaxi duty cycles are hard on tires, but real fleet data lands nearer 15,000–25,000 miles; at $950 a set over 20,000 miles that is $356/month, not $950 — and it leaves the depot retainer intact rather than consuming it whole. On electricity, $0.25/kWh is Supercharger retail, while the Cybercab is wireless-charging-only and therefore a depot asset by design. The critique's one unambiguously correct point is the platform fee: 30% is a steep take from a party carrying 100% of the hardware risk.
250 miles per day is 75% above Waymo's real-world rate of roughly 143 miles per vehicle per day. And $0.65/mile is less than half Tesla's actual Austin fare of $3.00 plus $1.40/mile — it assumes the price war has already happened and been survived. Relax either one and the model inverts.
The owner-operator needs 201 miles/day just to reach zero — 40% more than the most mature autonomous fleet in the world currently achieves per vehicle. At Waymo's actual rate the owner loses $403/month.
The 30% fee does not create value. It relocates who absorbs the loss.
Run my fleet model at their $0.65 fare with no platform fee at all — Tesla simply owning the car — and it loses money at every utilization from 143 to 250 miles/day. The fare sits below the base-case cost of $0.669/mile. Splitting that shortfall between two parties does not close it.
The owner-operator model has no remote-supervision line, because supervision is Tesla's cost, buried inside the platform fee. So ask whether the fee covers it.
The fee covers supervision only at 1:12 or better — and at exactly that ratio it leaves nothing for FSD R&D, dispatch, routing, customer support, or the insurance backstop. The owner-operator model does not refute the supervision thesis. It silently depends on it.
Run your own numbers. The figures above are one generic case. If you are actually weighing this purchase, the Cybercab Buyer's Calculator takes your city's demand, your insurance quote and your financing, and adds the two things this section does not model: loan payments, and whether the vehicle outlives the note.
One caveat on provenance: no Tesla owner-operator program terms have been published. The 30% take and the $0.65 fare are both assumptions, not announced policy. They should not be read as a plan anyone has committed to.
Where reality actually sits: 420 robotaxis registered in Texas as of 2 September, of which 45 are Cybercabs. Roughly one million cumulative unsupervised miles — under a single day of Waymo driving, which runs about four million miles a week across some 4,000 vehicles. Volume-production language was quietly dropped from the Q2 shareholder letter, battery pack capacity is named as the binding constraint on near-term output, and Musk has conceded material revenue is unlikely before 2027. The 3 September launch was a closed event with no livestream.
The base case is a genuinely attractive vehicle-level return — roughly $31,000 a year per unit, sub-year payback — and it is plausible by 2028–29. But it is gated almost entirely on moving supervision from about 1:3 to 1:15 or better, which is an autonomy-reliability milestone, not a manufacturing one, and one Tesla has not yet demonstrated at scale.
Tesla's stated goal of positive unit economics by mid-2027 is achievable in a narrow sense: a hand-picked set of dense Austin routes could plausibly show positive vehicle contribution. That is a different claim from the fleet earning back its capital, and the two will be easy to conflate when the first favorable number is published.
The number to watch is not production volume, vehicle cost, or cumulative miles. It is vehicles per remote operator — the one figure that decides this, and the one nobody currently discloses.