Uber has some autonomous ride-hailing lessons for Tesla

Alongside Uber’s earnings report Wednesday, where the ride-share service missed earnings estimates and provided underwhelming guidance, CEO Dara Khosrowshahi laid out a candid assessment of what the company has learned so far from its push into autonomous ride-hailing.

The remarks, meant for Uber investors, emphasized the difficult and uneven path autonomous ride-hailing will likely to take on the road to Uber’s “multitrillion-dollar opportunity.” Obviously, Khosrowshahi’s comments are meant to support his own business’s strategy, but they also likely bear some truth for competitors like Tesla, which is beginning to scale its own early autonomous ride-hailing efforts.

Today, Uber is working with Google’s Waymo to offer fully autonomous rides through its platform in Austin, Atlanta, and Phoenix. It also operates a limited robotaxi service with Avride in Dallas, where an onboard safety specialist still sits behind the wheel, and is testing a robotaxi fleet with Lucid in the San Francisco Bay Area.

Tesla, by contrast, is working on gradually removing safety monitors from its ride service in Austin and expanding a Bay Area fleet that relies on vehicles running supervised Full Self-Driving with drivers. During its earnings report last week, Tesla said it plans to expand its robotaxi service to seven additional US cities, including Miami, Dallas, and Las Vegas, in the first half of this year.

AVs and incremental growth

One of the central questions for Uber is whether autonomous vehicles will cannibalize its existing business or expand it. Khosrowshahi said early data suggests the latter.

“In Austin and Atlanta, where hundreds of AVs are operating on the Uber network, our overall (AV and non-AV) trip growth has significantly accelerated,” he said, adding that those operating zones are now among Uber’s fastest-growing in the US. Growth was driven both by new riders trying Uber for the first time and by higher usage among existing customers.

That dynamic suggests that the rapid expansion of robotaxi services in cities like San Francisco may not necessarily come at the expense of traditional ride-hailing.

Scaling from one city to the next

Khosrowshahi cautioned against extrapolating too much from early success in a handful of markets.

“No two cities are alike, and going from proof of concept to mass scale will be far more challenging,” he said, pointing to the unique mix of density, income levels, trip lengths, nice weather, and permissive regulatory conditions that have made San Francisco an unusually fertile ground for AV deployments.

Tesla CEO Elon Musk, meanwhile, has struck a far more optimistic tone. Last month, he said Tesla expects to be operating robotaxis in “dozens of major cities” by the end of the year, even if approvals are required on a city-by-city or state-by-state basis. He has also said Tesla’s existing robotaxi fleet — roughly 500 vehicles — could “double every month.”

AV utilization remains a hurdle

Khosrowshahi also argued that stand-alone AV fleets will struggle to achieve high vehicle utilization at scale, given how uneven ride-hailing demand is across hours, days, and seasons. Peak demand during rush hour contrasts sharply with long stretches of low demand, leaving expensive autonomous vehicles idle.

Data from the California Public Utilities Commission analyzed by The Driverless Digest found that while Waymo’s idle time has declined, vehicles still spend roughly 30% of their miles traveled waiting between trips.

Khosrowshahi said Uber’s hybrid model, combining autonomous vehicles with human drivers who can log on and off based on demand, is designed to smooth out that variability.

Tesla, by contrast, has promoted the idea that customers could earn money by lending their vehicles to its robotaxi fleet when not in use. That could potentially be useful if vehicle owners operate their cars outside busy times, but that’s just not when most people are driving. Musk has also floated the idea of using parked Teslas as part of a distributed AI inference network, though he has not described how owners would be compensated.

Most profits aren’t in big cities

“An often repeated myth is that the vast majority of US trips and profits are concentrated in the top cities,” Khosrowshahi said. “The truth is that the US market comprises a long tail of thousands of cities, suburbs, towns, and rural areas with significant diversity in market characteristics and regulatory requirements.”

The company’s top 20 cities, he said, represent only 30% of Uber’s US gross bookings and 25% of its profits.

Of course, Khosrowshahi is talking his own book here. Uber already does traditional ride-hailing in big and small cities across the country.

Tesla, meanwhile, has said that thanks to training on data from existing drivers around the country, it can expand driverless ride-sharing everywhere including outside dense urban cores. Musk had previously said that robotaxis would be available to half the population of the US by the end of 2025.

Edge cases are a big deal

Despite rapid progress, Khosrowshahi said autonomous vehicles remain far from meeting the level of reliability and ubiquity cities and riders expect. He pointed to gaps in geographic coverage, weather disruptions, and infrastructure failures that have sidelined AV fleets — situations where human drivers continued to keep Uber’s network running.

Those concerns echo warnings from AV experts like Phil Koopman, an associate professor at Carnegie Mellon University, who has argued that each order-of-magnitude increase in robotaxi fleet size brings new, hard-to-predict challenges that remain difficult for autonomous systems to handle without human intervention.

“At 1,000 cars you can expect to see things that require a lot more training to handle than you’ll get from a few million miles of material used to create training data,” Koopman wrote. “A fleet at this size will experience frequent novel unstructured situations that people can figure out how to muddle through, but that tend to flummox robotaxi technology.”