Skip to content
Franklin Hugh Money Start the SIE

Robotics as a Service · Part 1

Cybercab

Robotics as a service, pay per mile

Investments · September 3, 2026 · 11 min read

A car that seats two, weighs 3,113 pounds, carries a 48 kilowatt-hour battery, and has no steering wheel, no pedals, and no mirrors. Tesla began building it at its Texas factory in April 2026 with a target price around $25,000, and as of today it carries paying riders in Austin. Tesla does not plan to sell you one. This is the arithmetic of why.

Sold Once

Start with the business Tesla has been in since 2008: it builds a car, sells it, and moves on. At $25,000 and Tesla's company-wide gross margin in the second quarter of 2026, which was 16.8%, one Cybercab sold to a customer produces about $4,200 of gross profit. Gross margin is the share of the sale price left after the cost of building the thing: the cells, the steel, the labor on the line, and nothing else. Out of that $4,200 Tesla still has to pay for engineering, sales, and the overhead of the factory. Then the car belongs to somebody else. Its electricity, its insurance, and its slow loss of value are their problem. Tesla sees the customer again when the software subscription bills.

Revenue once, margin once. That has been the shape of the car business since the Model T, and it is why carmakers trade at a modest multiple of their earnings. A car company earns a sale. It does not earn the car.

Kept

Now keep the car. Tesla's robotaxi service in Austin, running on Model Ys since June 2025, charges $3.00 a ride plus $1.40 a mile, a fare card it set in March 2026. The revenue side of the model is public. The utilization side, how many paid rides a car does in a day, Tesla has not published. So borrow Waymo's. Waymo gives more than 500,000 paid rides a week across a fleet of roughly 3,700 vehicles, which works out to about 135 rides a car a week, or 19 a day. Assume the average ride is five miles, the short end of the trips Austin riders describe.

One Cybercab, kept, one year (illustrative)
LineArithmeticFigure
Rides19 a day, 365 days≈ 7,000
Paid miles7,000 rides at five miles≈ 35,000
Base fares7,000 at $3.00≈ $21,000
Mileage fares35,000 at $1.40≈ $49,000
Gross fares≈ $70,000

Seventy thousand dollars a year from a machine that costs $25,000 to build. That is the sentence a great deal of Tesla's market value rests on. The rest of this article is what the sentence leaves out.

What the Car Costs to Run

Tesla has said its target operating cost for the Cybercab is $0.20 a mile, covering charging, insurance, and depreciation. Morgan Stanley has estimated the operating cost of Tesla's robotaxi service as it runs today, on Model Ys, at $0.81 a mile, against $1.43 for Waymo and $1.71 for a human-driven rideshare. The two figures are four times apart, so the model runs on both.

The same car, after operating cost (illustrative)
LineAt Tesla's $0.20 a mileAt Morgan Stanley's $0.81 a mile
Gross fares≈ $70,000≈ $70,000
Operating cost, 35,000 miles≈ $7,000≈ $28,000
Contribution≈ $63,000≈ $42,000
Payback on a $25,000 car≈ 5 months≈ 7 months

Contribution is what is left of the fares after the direct cost of earning them. Payback is the number of months until those leftovers add up to the price of the car. On these inputs a kept Cybercab earns its own cost back in well under a year and then produces, for whatever remains of its life, something between $42,000 and $63,000 a year. A sold Cybercab produced $4,200, once.

Now the list of what neither figure includes, because it is long and all of it lands on the cost line. The remote operators who take over when a car gets stuck; Tesla does not say how many cars each one watches. The depots, the chargers, and the people who clean a two-seater at 2 a.m. The empty miles between a drop-off and the next pickup, which earn nothing and wear the car anyway. The fare cuts that arrive when Waymo and Tesla share a city and one of them wants the other's riders. The cost of a crash, which in a driverless car is the company's, and which the insurers have not finished pricing. The $0.20 is a target for the car. The service costs more than the car.

The Number That Decides It

Utilization. The model above assumes a Cybercab does 19 paid rides a day, and that assumption was borrowed from a company with 3,700 cars, seven years of commercial operation, and demand it has spent more than $21 billion building. Tesla's unsupervised fleet in Texas was about 25 cars in late April 2026 and 42 in state filings by June, with Model Ys added in Florida and elsewhere since. A car that does five rides a day instead of 19 earns about $18,000 in fares rather than $70,000, and at $0.81 a mile its contribution falls to roughly $11,000. Payback stretches past two years. Still a business. A far smaller one.

Halve the utilization and the model survives. Halve it again and the Cybercab is a cheap car earning a taxi's wage in a city that has not yet decided it wants one.

The Gap Tesla Is Betting On

The reason the Cybercab matters, in this frame, is the first of the five RaaS numbers: the cost of the robot. Waymo's Jaguar I-Pace costs an estimated $150,000 to $200,000 once its sensors are fitted. Its new Zeekr-built Ojai, which entered public service in August 2026, costs roughly $125,000 by Morgan Stanley's estimate, with the base van around $32,000 and the autonomy hardware under $20,000. Tesla's target for a finished Cybercab is $25,000.

Run the same fare card and the same 19 rides through a $125,000 robot and the contribution is unchanged, but payback moves from months to years. Same revenue, five times the capital. If both companies reach the same utilization, the one with the cheaper car earns its fleet back faster, replaces it sooner, and can cut fares further before it loses money. That is the entire thesis of the Cybercab, and it is a thesis about manufacturing cost more than software. It is also why Waymo moved to a Chinese-built van. Both companies read the same arithmetic.

The bet has a condition. Waymo's expensive sensors are the reason it has 3,700 cars operating without a driver, and Tesla's cheap ones are one reason its driverless fleet is counted in dozens. A $25,000 robot that cannot drive is a $25,000 loss. The gap only pays if the cameras-only system does the job at scale, and as of this writing the evidence for that is about 2.5 million paid miles and, in Tesla's own words, zero notable incidents across the latest 380,000. Encouraging, and small.

The Car You May Not Be Allowed to Buy

One detail from the regulations. Federal motor vehicle safety standards were written for cars with a driver, and until this year they required a brake pedal and a steering wheel. A carmaker could build up to 2,500 non-compliant vehicles a year under an exemption. Tesla did not apply for one. It built the Cybercab to satisfy the standards as written where it could, certified it itself, and waited for the rules to change. In 2026 the regulator began removing the brake-pedal requirement for vehicles designed never to be driven by a person and said it would "absolutely" consider doing the same for the wheel.

That decides whether the Cybercab can be sold at all, and to whom. A fleet car on Tesla's own books needs only to be legal to operate. A car sold to a private owner, who then puts it into Tesla's network and keeps a share of the fares, needs to be legal to own. Musk described that arrangement in 2019, with the owner keeping roughly 70% to 75%, and in July 2025 said private cars would join the network the following year. As of September 2026 no owner can enroll a car and no terms have been published. The owner's seat at the RaaS table is a chair with nobody in it.

Scaled to the Milestone

Tesla's 2025 pay package for its chief executive, approved by shareholders in November 2025, pays out in part on one million robotaxis in commercial operation. One million cars at $70,000 of fares is $70 billion a year, before any of the costs above. Tesla's revenue in the second quarter of 2026, from every car, battery, and subscription it sold, was $28.2 billion, or about $113 billion annualized. The milestone, if reached, would add most of a second Tesla, built out of rides.

The cost of building it arrives first. Capital spending in the second quarter of 2026 was $5.79 billion, up 142% in a year, and free cash flow turned negative. Musk told analysts robotaxi revenue would not be "super material" in 2026 and would be "material in a significant way" in 2027. Seven Cybercabs began carrying riders today. The factory that built them can make 125,000 a year.

What Has to Be True

Five things, in the order the model needs them.

  1. The car costs $25,000 to build, at volume. That is a target, and the first ones cost more.
  2. It drives itself in a city without a person watching every car, at a ratio of cars to remote staff that keeps $0.81 a mile from becoming $1.43.
  3. It does something near 19 rides a day, which means the riders show up.
  4. The fare holds near $1.40 a mile while a competitor with $21 billion of funding operates in the same city.
  5. The car lasts. Nobody has published a design life in miles, and the payback math assumes it keeps working after it has paid for itself.

A car sold once is a transaction. A car kept is a bet on all five. The next piece runs the same numbers on a robot that walks.

Education, not advice. This piece models how a product's cost becomes revenue. It is not a view on Tesla's stock or on any security. Franklin Hugh Money does not recommend securities and does not publish the author's personal holdings.
Next · Part 2 Optimus: The Model Employee A worker that costs $25,000 once and about a dollar an hour after, beside a human at $46.

Sources

  • Tesla, Q2 2026 Update, July 2026: revenue, gross margin, capital spending, free cash flow, paid robotaxi miles, Cybercab capacity. Call remarks via Not a Tesla App.
  • Cybercab EPA filing (battery, weight, power, charging), as reported by Electrek, June 15, 2026.
  • Austin fare card, March 2026, as reported by TheStreet.
  • Tesla's $0.20-a-mile target and Morgan Stanley's per-mile estimates, as reported by Basenor; Ojai unit cost via TechCrunch, May 2026.
  • Texas fleet counts via TechTimes, June 2026; Cybercab registrations and launch via Basenor, September 2026.
  • Federal standards and self-certification, as reported by Electrek, April 2026, and CleanTechnica, June 2026.
  • Owner revenue share: Tesla Autonomy Day, April 22, 2019; private cars joining the network, Q2 2025 earnings call, as reported by Fortune.
  • 2025 CEO Performance Award milestones, SEC filing.

← Robotics as a Service