What 50x Actually Means for Robotaxi: Why Fleet Miles Rewrite Unit Economics

Wall Street keeps counting robotaxi cars, and I think that is the wrong scoreboard.

Tesla has something like 25 unsupervised robotaxis on the road across Austin, Dallas, and Houston, plus hundreds more with a human in the seat. Waymo runs north of 3,000 commercial vehicles across Phoenix, San Francisco, Los Angeles, and a handful of other cities, almost all of them empty up front. Put those two fleet numbers next to each other and the story writes itself: Waymo is winning, Tesla is a rounding error, Elon is failing again. I have been watching this story since the first FSD beta, and I have driven the software in my own car for years. Fleet count is a distraction. The number that decides who owns the multi-trillion-dollar robotaxi market is cumulative miles.

Tesla just crossed roughly 10 billion cumulative Full Self-Driving miles across the entire customer and robotaxi fleet, going back to the first beta. That pile is growing by about a billion miles every month, and the cumulative total doubles roughly every nine months, so ten becomes twenty in the next nine months. Waymo sits somewhere north of 200 million cumulative driverless miles across its whole history, including the old Google self-driving years. Their doubling rate has sped up too, and 2024–2026 public milestones put them near the same nine-month doubling cadence.

Pause on that math with me. Tesla at 10 billion. Waymo at about 200 million. Same ballpark doubling rate. My math says Tesla is roughly 50 times Waymo on cumulative miles. The comparison is not perfect, because most of Tesla's miles are still supervised FSD with a human in the seat while most of Waymo's miles are fully driverless with remote supervisors. Still, for training data — the signal that actually moves the unsupervised stack — the gap is widening every hour in Tesla's favor. Mainstream coverage almost never puts those two numbers on the same slide. When you do, the "Waymo already won" framing collapses.

Why do miles matter this much? Because of Wright's Law. In 1936, Theodore Wright noticed that every time airplane production doubled, unit cost fell a fixed percentage — roughly 15% to 20% per doubling. That pattern has held for nearly a century across solar, batteries, semiconductors, hard drives, and LEDs. For autonomy, the unit is not the car. It is the mile. Every cumulative mile is training data, edge-case experience, safety validation, and cost reduction in the inference loop. The system that doubles miles fastest gets cheaper, smarter, and safer faster. Smart skeptics will say data quality beats raw volume, and I get that. Directionally, though, every doubling still throws exponentially more edge cases at the model, so cost per mile falls on a slope. Tesla's slope is steeper because Tesla can grow miles faster than anyone else.

The cost numbers are already showing it. Morgan Stanley recently pegged Tesla's robotaxi cost per mile around 81 cents. Waymo estimates land between about $1.36 and $1.43. Tesla is already running near 60% of Waymo's cost with a tiny commercial fleet in the first inning. On the price side, Tesla is charging roughly $1 to $1.40 per mile in Austin, while Waymo sits at a premium to ride-hail and Uber today is around two bucks a mile for the rider. ARK Invest's long-run fully autonomous number is about 25 cents a mile. Stack that staircase: Uber near $2, Waymo above that, Tesla at $1 to $1.40 with an 81-cent cost basis, and a scale target near a quarter. Tesla owns the means of production for the whole stack, so its cost curve can keep cascading while rivals buy cars from someone else.

Manufacturing is the other half of the slope. Tesla can build on the order of half a million vehicles a quarter before Cybercab fully ramps. Waymo leans on Zeekr and Hyundai, and combined contracts look like maybe 20,000 vehicles a year at best — roughly 100 times fewer self-driving capable cars per year. That capacity gap feeds the mile flywheel. Pony.ai just published the cleanest break-even benchmark in this debate: about 23 paid rides per vehicle per day. Who gets there first? Waymo at premium pricing with roughly 3,000 cars and a slower cost curve, or Tesla at $1 to $1.40 with an 81-cent cost basis, doubling miles faster, and millions of cars in the global fleet that can flip into robotaxi service the moment regulators and safety levels allow it?

Paid miles are already ramping. On the Q1 2026 earnings call, Tesla reported roughly 1.7 million cumulative paid robotaxi miles in the quarter, up from about 610,000 in Q4 — nearly a triple in one quarter. Most coverage shrugged because 1.7 million looks small next to Waymo, and Waymo deserves credit for getting there first. If Q2 follows the same trajectory as the unsupervised fleet ramps, you are looking at something north of 4 million paid miles. If Q3 follows, maybe above 10 million. By the end of 2026, Tesla's paid robotaxi miles could start approaching the neighborhood of Waymo's annual commercial volume. Waymo spent roughly 16 years crawling from a Google research project to about 3,000 deployed vehicles. Tesla is aiming to put several thousand, maybe tens of thousands, of Cybercabs out of one factory line in the next 12 months, on top of Model Y lines already feeding robotaxi cars. If the ramp holds — and I will say that is a real if — fleet count by the end of next year could sit somewhere between five and ten times Waymo's current commercial fleet, pending safety and regulatory clearance.

Supply is not the bottleneck I used to fear. Tesla already makes on the order of 2 million cars a year, with plants, suppliers, cells, chips, and software locked in. Cybercab production kicked off at Giga Texas in April 2026. The vehicle is FMVSS compliant from day one, which means Tesla is not stuck under the NHTSA 2,500-vehicle exemption cap people (including me) worried about for two years. Tesla's head of engineering, Lars Moravy, confirmed that in a reply to one of my posts. The open question is how hard Tesla wants to push the line in 2026 and 2027, not whether the car is legally capped at a toy volume.

Service geography is still tiny but multiplying. Unsupervised paid robotaxi expanded from Austin to Dallas and Houston on April 18, with roughly 19 cars in Austin, three in Dallas, and three in Houston. Tesla has talked about five more cities in the first half of 2026 — Phoenix, Miami, Orlando, Tampa, Las Vegas — and I doubt all of those hit on time. Every new city still multiplies cumulative miles and training data. The network is unified: a weird intersection in Houston helps the Phoenix car when it launches, and customer cars around the world train the same nets that run unsupervised robotaxis in Texas. Waymo is cutting its reliance on HD maps over time, but it still has to buy cars from other manufacturers. Tesla can crank the factories.

I want three scenarios on the table, because this is not a sure thing.

Scenario one: I am right and the doubling math holds for the next 24 months. Cumulative miles go from about 10 billion to roughly 40 to 80 billion by mid-2028. Cost per mile falls from 81 cents toward 35 to 50 cents. The robotaxi fleet expands from about 25 vehicles toward 5,000 to 25,000, mostly Cybercab, mostly U.S., with international markets opening late 2027. Annual robotaxi revenue gets to a real number — call it north of $5 billion — and Tesla becomes the volume-and-price leader.

Scenario two: I am partially wrong and a regulatory wall, software wall, or public safety incident slows the doubling to every 18 to 24 months. Cybercab gets capped or delayed. Tesla still ends up the largest autonomy player by mile count and cost per mile, but the timeline stretches two to three years. The thesis works. It just takes longer.

Scenario three: I am completely wrong and the doubling math breaks. FSD fails to generalize. AI5 slips again. A Cybercab injures someone and NHTSA pauses the rollout. Waymo's slower, expensive, premium approach turns out to be the right architecture. Tesla's near-term robotaxi thesis gets crushed. Even then, Tesla still owns the AI hardware stack, still has Optimus, still has Energy growing, and still sells roughly 1.6 to 2 million EVs a year with FSD. Autonomy slips. The company does not go to zero.

Why does any of this matter if you do not own the stock? If you own a car, drive a car, or pay to move yourself around a city, the cost of that movement is headed toward an order-of-magnitude drop inside the next seven to ten years. ARK's 25 cents per mile against Uber's roughly two bucks is about an eight-times cut. The average American drives around 13,000 miles a year. At 20 to 25 cents a mile in a robotaxi, annual transport lands near $3,000 — about $250 a month — against something closer to $900 a month for payment, insurance, gas, and maintenance. That math collapses car ownership for a lot of households in the same window. Parking, school drop-offs, aging parents who should not drive — all of it reprices. The same exponent that decides who wins the autonomy race also makes movement cheaper for hundreds of millions of people. That is what 50x actually means to me. Not a fleet snapshot. A learning curve.

Check the video here.