Every autonomous vehicle the public has ridden in until now has been a conversion. Waymo buys a Jaguar, or now a Zeekr built in Ningbo, ships it to Mesa, Arizona, and bolts on a lidar-and-radar suite that costs more than the car underneath it. Morgan Stanley estimates the current Waymo at $125,000 a unit.
The Cybercab is not a conversion. Two seats, no steering wheel, no pedals, no lidar, no second life as a consumer car. Musk has confirmed it retails under $30,000. The average new car sold in America in July went for $49,855.
Chart E — The asset the network is built on. Waymo per-unit costs are Morgan Stanley estimates via The Next Web (2026). The Cybercab figure is Musk's confirmed sub-$30,000 retail price (Fox Business, 19 Feb 2026), used as an upper bound on build cost. US average new-car transaction price, July 2026: Kelley Blue Book / Cox Automotive.
That single comparison is the thesis in one picture. If the software holds, this is the first autonomous product that can be built at the scale of an ordinary car, at the price of an ordinary car, and run at a cost per mile nothing with a human in it can approach.
Four things that actually matter
1. The constraint moved from the regulator to the factory. On 27 July, Nevada capped Tesla at ten robotaxis. On 20 August, the same authority voted unanimously to allow up to five thousand. A 500-fold change in twenty-four days. Until August the honest bear case was that regulators would not let the fleet get large. In Nevada, that argument is now dead — and factories are the one problem Tesla has repeatedly solved.
2. Vertical integration is not a slogan here; it is a number. Tesla designs the car, casts and assembles it, makes the battery, fabricates the inference chip inside it, writes the software, owns the charging network, underwrites the insurance and operates the fleet. No competitor holds more than two or three links of that chain. It converts into one figure that decides the industry — Tesla is targeting roughly $0.30 per mile in long-run operating cost. A network with that marginal cost does not have to win a price war. It sets the price and everyone else decides whether they can survive at it.
3. Demand is not a fixed pie. The standard mistake is to size robotaxi as a share of today's taxi market. That is exactly what was said about Uber in 2013. New York went from about 14 million for-hire rides a month before 2014 to 25 million by end-2017 — a 79% larger market. In Manhattan roughly 70% of that was cannibalisation; in the outer boroughs almost all of it was net new demand that had never existed, because the price had always been too high. Tesla is at the same pressure point with a bigger lever: a five-mile ride at $8.25 instead of $19.69 is not a discount, it is a different product, competing with the bus and with the second car in the driveway.
4. Every robotaxi on the street is a free advertisement for the software. The system driving in Austin is the same one Tesla sells to owners for $99 a month. Trust in autonomy is not built by a spec sheet — it is built by watching one go past your window a hundred times and nothing happening.
Chart F — The flywheel that pays for itself. Sources: Tesla disclosures compiled by Not a Tesla App (Q2 2026); Tesla Q2 2026 earnings call; Lars Moravy on historical take rate (Sep 2025).
Paid FSD customers: 400,000 in 2021, 1.48 million now, up 56% year on year. Fleet take rate: single digits a few years ago, 15.2% last quarter. Model S and X buyers: 50–60%. New North American deliveries leaving with FSD attached: 55%. Since February the one-time purchase no longer exists, so every future sale is recurring — on a car already built, already carrying the computer, at an incremental cost close to zero. Robotaxis build trust; trust lifts the take rate; the take rate compounds a near-pure-margin software line across an installed base of millions. Same fleet, monetised twice.
And the part almost nobody is pricing
A humanoid robot cannot carry a lidar mast. No roof to mount it on, no budget to buy it, no power envelope to run it. A robot has to perceive the world the way a person does — cameras, neural network, real time, on a chip small enough to fit inside it. That is precisely, and only, what Tesla has spent a decade building.
So the robotaxi is not just a business. It is the public proof of the perception stack Optimus depends on. JPMorgan's August factory tour found Tesla explicitly "aligning humanoid data collection with its FSD methodology" — same cameras, same training pipeline, same silicon, same team — with external Optimus sales as early as H2 2027 and long-run capacity ambitions of a million units at Fremont and ten million in Texas.
If you believe the robotaxi thesis, you have already assumed away the hardest part of the robot thesis. The market prices these as two separate bets. They are one bet with two payoffs.
What it is worth — and where to get the numbers
We have built the whole thing out as three separate businesses, valued separately and summed: the robotaxi network, the car-and-energy business with the FSD subscription attached to it, and Optimus as risk-weighted optionality.
The short version: our base case sits meaningfully below today's price, and our bull case is close to five times it. The distance between them is almost entirely one variable, and it is not the one most commentary focuses on.
"Tesla: the three-legged thesis" — the complete build. All three legs with every assumption exposed, the sum-of-the-parts table, a fleet-size-versus-Optimus-volume sensitivity grid, the live valuation spreadsheet, and a fully sourced data pack that grades every input A, B or C so you can see exactly which numbers are Tesla's claims and which are independently verified.
The problem with being right about Tesla
Here is what every Tesla bull learns eventually: being right about the company is not the same as making money on the stock.
The thesis in this note plays out over years. The stock does not wait for it. Tesla has traded between $297 and $499 in the last twelve months alone. Anyone who bought the story at the wrong point in that range spent a long time being right and poor at the same time. Timing an entry on a name this volatile is close to impossible, and holding through the drawdowns is a test most people fail.
That is exactly why we do not simply buy and wait.
At Proflex, we run a defined options structure on names like this — rolled put spreads with overlapping short-dated calls — designed to harvest the volatility itself while the long-term thesis plays out underneath. The volatility that makes Tesla painful to own is the same volatility that pays us to wait for it.
Our subscribers have seen significant returns in each of the last three years using this approach — without needing to call the top, catch the bottom, or sit through a 40% drawdown hoping the thesis lands. The long-term case does not have to be a waiting game just because the stock is volatile.
The complete walkthrough of a live Tesla structure — strike selection, roll mechanics, what we do when the stock gaps against us, how the position is sized, and the full trade-by-trade record. No paywall.
Sources and full data pack accompany this note. Tesla discloses no robotaxi ride count, revenue or fleet size; figures marked as estimates are ours, not the company's.