The key race: kilometers driven empty-seated, not cities announced
A robotaxi business is a utilization problem wearing a self-driving costume. The number that compounds is cumulative driverless distance on public roads, because that is both the safety denominator and the data flywheel. Vehicle count and geofence area only matter if they produce kilometers. Waymo is still the only operator that publishes a fully autonomous public-road mile pile at national scale and a weekly rate (about 4 million miles a week, more than 220 million fully autonomous miles through March 2026). Baidu Apollo Go publishes the largest autonomous-kilometer total, 350 million-plus as of its August 18 Q2 print, of which 220 million km were fully driverless as of May. Tesla's unsupervised Robotaxi program crossed 1 million miles on September 3, 2026, two orders of magnitude behind both.
City-count press releases hide that gap. Tesla's unsupervised map now covers six metros; Waymo's paid public map covers 14; Apollo Go cites 28 cities. Those are not the same product. Tesla's Austin geofence is hundreds of square miles served by a fleet that was in the tens of unsupervised cars for most of 2026 and is now 420 vehicles registered in Texas, including 45 Cybercabs. Waymo's 1,400 square miles sit under about 4,000 commercial vehicles. Apollo Go does not publish a global fleet total at all. The tracker on this page is built so those series can be read together without pretending they were measured the same way.
Who's ahead: Baidu on kilometers, Waymo on US paid trips, China on vehicles
Split the leaderboard by the unit. On disclosed autonomous kilometers, Apollo Go is ahead. On fully driverless public-road miles with a weekly rate, Waymo is ahead, and it is the only US operator whose safety record has been independently scored per mile (IIHS, July 2026). On paid US trips, Waymo is still in a different league: about 500,000 a week since late March, against a year-end goal of 1 million, while Zoox has carried nearly 1 million riders lifetime and Tesla does not publish a weekly trip figure. On vehicles, the Chinese public companies are the ones adding metal in thousand-unit steps: Pony.ai went from 270 robotaxis in April 2025 to 1,975 on June 30, 2026; WeRide's robotaxi fleet is 1,800-plus inside a 3,400-vehicle L4 mix that also includes buses and sweepers.
Zoox is the first US paid service in a purpose-built vehicle with no steering wheel or pedals, live in Las Vegas since August 10 including Harry Reid airport since September 3, on a roughly 65-car permitted fleet. That is a product milestone, not a volume one. Tesla's Cybercab entered paid Austin dispatch on September 4, the same week NHTSA opened Audit Query AQ26002 into how Tesla self-certified a controls-free car. Avride (Dallas, Uber, ~200 cars, safety operator still in the seat), Motional (Las Vegas, Uber, supervised), May Mobility (shuttle-like, supervised), DiDi's R2 trials in Beijing and Guangzhou, and AutoX's quieter Shenzhen operation fill out the field. Cruise is on the chart as a flatline at zero: GM shut the robotaxi business in December 2024.
What decides it: what a kilometer means, and whether the map has cars on it
Comparability is the whole problem. Waymo's series is fully autonomous public-road miles. Apollo Go's headline pile mixes earlier safety-driver operations with driverless km; the driverless subset is the fairer comparison and is not restated every quarter. Pony.ai's 100 million km includes robotruck testing. WeRide's 55 million km is all L4 products. Tesla Full Self-Driving miles on customer cars are not robotaxi kilometers and are not on this page. A log-scale distance chart is the honest default, because a linear axis makes everyone except Waymo and Baidu disappear.
Coverage has the same trap. A 288-square-mile Austin geofence with 20 unsupervised cars is not 288 square miles of service. WeRide's 60-plus cities include testing and permits; Waymo's 14 are paid public metros. Tesla's Texas registration count of 420 is not an unsupervised in-service count, and it excludes Florida. The basis label on every series exists so those facts stay attached to the line.
The money: thousand-car fleets still lose money, and Waymo is borrowing
Scale has not yet meant profit. Pony.ai's robotaxi revenue hit $12.1 million in Q2 2026, up 691 percent year over year, on a 1,975-car fleet that is still targeting 3,500 by year-end. WeRide posted record Q2 revenue of RMB 231.7 million and a 37.5 percent gross margin, with an adjusted EBITDA loss still in the hundreds of millions of yuan. Apollo Go says it has reached unit-economics breakeven in its largest China city despite low fares. Waymo, already the best-capitalized operator after a $16 billion equity round at $126 billion in February, was in talks in early September for more than $3 billion of first-time debt. Even the leader is funding fleet growth and compute with leverage.
Tesla's bet is that a vision-only stack plus a cheap purpose-built car inverts that cost curve. The Cybercab is in paid service and in an NHTSA certification audit at the same time. Zoox paid for the other path: a two-year NHTSA exemption capped at 2,500 vehicles a year. Uber is the demand layer for everyone who does not own a consumer app, brokering WeRide, Apollo Go, Pony.ai, Motional, Avride, and a Lucid-Nuro premium effort that is still employee-only.
What to watch through 2027
Watch whether Waymo's weekly paid rides break the 500,000 plateau toward 1 million, and whether Alphabet's late-October Q3 print restates fleet, miles, or only the same weekly-ride line. Watch whether Tesla's unsupervised mile pile, which jumped from 380,000 on July 22 to 1 million on September 3, keeps that slope once Cybercab is more than a few dozen Texas cars, and whether NHTSA's Cybercab audit stays an information request. Watch Pony.ai's path from 1,975 vehicles to the 3,500 year-end target, and WeRide's from 1,800 robotaxis toward 2,600.
Watch Apollo Go's China permit thaw after the March 31 Wuhan stall and the nationwide pause on new robotaxi permits, and whether the 350 million km figure gets a fresh driverless split. Watch Zoox convert San Francisco from free waitlist to paid once California deployment permits land, and whether the Uber integration in Las Vegas actually goes live after missing summer. Watch DiDi's R2 trials become a disclosed fleet. The throughline is unchanged: driverless already works in paid service. The next year grades who can put kilometers on the board every week, not who can publish a new city.
Sources
- Waymo blog - 200 million-plus fully autonomous miles; 4 million miles per week; 220 million through March 2026
- Baidu - Q2 2026 results (23 million-plus rides, 350 million-plus autonomous km, 28 cities)
- Electrek - Tesla Robotaxi 1 million unsupervised miles
- Pony.ai - Q2 2026 exhibit (1,975 robotaxis as of June 30)
- WeRide - Q2 2026 results (1,800-plus robotaxis, 3,400 L4 vehicles)
- Bloomberg - Waymo first debt raise talks
