Tesla's robotaxi strategy looks a lot like Waymo's
Tesla is rolling out a geofenced taxi service in Austin backed by remote operators.
We are two months away from the Ride AI summit in Los Angeles on April 2. We’re bringing together industry leaders to discuss what’s working and what comes next for autonomous vehicles. I will be there, and we’ve got a great lineup of speakers, including Mobileye CEO Amnon Shashua, Kodiak CEO Don Burnette, and executives from Waymo, Zoox, and Wayve—with many more speakers still to be announced.
For a limited time, Understanding AI readers can click here to get a ticket for the discounted price of $500. You can also click here if you’d like to apply to speak at the conference—we’re especially interested in applications from people with hands-on experience building autonomous vehicle technology.
Elon Musk is a master showman who seems to do everything he can to keep Tesla in the news and its stock price high. Tesla generated a lot of headlines last October when Musk unveiled the Cybercab, a driverless robotaxi with no steering wheel that will supposedly ship in 2026. During that event, held at a Warner Brothers studio lot, Musk wowed the crowd with a staff of humanoid Optimus robots serving drinks. Subsequent reports indicated that they were likely controlled by remote human operators.
Waymo takes the opposite approach. For example, a couple of weeks ago, Waymo announced in a tweet that Waymo employees would begin testing Waymo’s driverless vehicles on freeways in Los Angeles. Waymo employees have been able to take driverless freeway rides in the San Francisco area since last August, and in Phoenix since last May.
When will customers be able to start taking driverless freeway rides? Waymo hasn’t said. But I bet when it happens, it’ll be announced with a bland tweet and rolled out gradually across Waymo’s various service areas.
Waymo is a subsidiary of Alphabet, which means it doesn’t need to hustle for outside investment. This means hype has little upside for Waymo. And so for many years, the company’s public relations strategy has seemed calculated to discourage media coverage.
But while Waymo hasn’t gone out of the way to tout its expansion, Waymo is expanding. Waymo’s service in Los Angeles became fully open to the public in November, and Waymo expanded its Los Angeles service area just this week. Waymo is now actively testing services in Austin and Atlanta, with Miami coming soon.
Waymo is preparing to expand its San Francisco service down the peninsula to Silicon Valley. And Waymo says that in 2025 the company will take “road trips”—the first step toward launching a new service—in at least 10 additional cities, including San Diego, Las Vegas, Washington DC, New Orleans, and Tokyo.
One bottleneck for Waymo’s expansion is vehicle manufacturing. Waymo says it has more than 700 in-service vehicles in its commercial fleet. I expect that number to get a lot higher in the coming months.
Last month, Autonomy Central published drone footage of around 2,000 Jaguar IPACE vehicles parked outside a Waymo manufacturing facility in the Phoenix area. Close-up photos showed that many of the vehicles have cut-outs where sensors will be added. So by the end of the year I expect Waymo to have thousands—not hundreds—of vehicles in its commercial fleet.
Of course I know what Tesla fans will say here: Tesla produced 1.77 million cars last year alone. So while Waymo is growing quickly—weekly rides grew from 10,000 per week in May 2023 to 150,000 rides per week in October 2024—Tesla has the potential to grow much faster.
But I think this underestimates the complexity of building a nationwide robotaxi service. Yes, you need capable software and a fleet of vehicles. But you need a lot of other things too: depots, on-site support staff, remote operators, state and local permits, and relationships with first responders in each community where the service operates.
Tesla fans have long faulted Waymo for building a taxi service that only worked in a few specific geographic areas. But as Tesla has gotten closer to launching its robotaxi service, something funny has happened: Tesla’s rollout strategy has looked more and more like Waymo’s!
Tesla’s geofenced robotaxi service
Last October, Musk said that unsupervised FSD would roll out first in California and Texas. In December, Reuters reported that Tesla was in talks with the city of Austin, Texas, to launch a robotaxi service there. In an earnings call last month, Musk said that Tesla would launch a robotaxi service in Austin in June then in California later in 2025.
Whether or not Tesla manages to hit these deadlines—and they seem pretty optimistic to me—they represent a major shift in Tesla’s strategy. Tesla is rolling out its robotaxi gradually, one city at a time, just as Waymo has done over the last five years.
In his January earnings call, Musk insisted that this didn’t represent a major change.
“While we’re putting our toe in the water gently at first, just to make sure everything’s cool, our solution is a generalized AI solution. It does not require high-precision maps of a locality. So, we just want to be cautious. It’s not that it doesn’t work beyond Austin, in fact it does.”
The case for Tesla having a distinctive strategy rests heavily on this point about high-precision maps. Tesla’s vehicles do have maps, they are just less detailed than the ones Waymo uses. Musk has long portrayed this as a key advantage, arguing that it will be too expensive for Waymo to create and maintain higher-density maps nationwide.
When I asked Waymo co-CEO Dmitri Dolgov about this last October, he told me that the cost of mapping was “not a significant factor in terms of our expansion into places. It is a contributor to the overall cost but there are many other things and it is not at the top of the list.”
I find Dolgov’s argument more convincing here. Once Waymo has its fleet up and running in a particular city, Waymo vehicles will be able to collect fresh sensor data automatically as they drive around. Moreover, Tesla has tweeted about developing “fleet-scale auto-labeling” technology that can automatically turn sensor data into a high-density map. I would be surprised if Waymo didn’t have similar technology.
In that January earnings call, Musk predicted that Tesla will “most likely release unsupervised FSD in many regions of the country” before the end of 2025. But this seems like a classic case of Musk over-optimism.
A number of companies have attempted to build driverless taxi services over the last decade. Every one of them found that they needed to do many months—if not years—of testing with safety drivers before they were ready for fully driverless operation.
Moreover, the complex logistics of entering new markets—applying for permits, building depots, hiring local support staff, training first responders, and so forth—means it can take a year or more to launch in a new city.
I would not be surprised to see driverless Teslas in “many regions” before the end of the decade. But this year seems unlikely.
Tesla is hiring remote operators and upgrading vehicle hardware
Since 2018, Waymo has had a teleoperations team that gives vehicles remote guidance if they get confused or stuck. Tesla fans have long pointed to this as evidence that Waymo’s technology isn’t fully autonomous. But in November a Tesla job listing revealed that Tesla now has a teleoperations team responsible for “providing remote access to our robotaxis and humanoid robots.” Tesla has apparently discovered what Waymo learned more than six years ago: even the best self-driving software occasionally needs remote assistance.
Tesla fans have long argued that Waymo is being held back by its expensive hardware. Since 2016, Tesla has claimed that every new vehicle had the chips and cameras required for full autonomy. But during last month’s earnings call, Musk acknowledged that Tesla would need to upgrade the HW3 computer chips in Tesla vehicles sold between 2019 and 2023.
“That’s going to be painful and difficult, but we’ll get it done,” Musk told Tesla shareholders.
In short, as Tesla gets closer to launching a driverless taxi service, it’s becoming more obvious that Tesla’s strategy isn’t that different from Waymo’s. And this could make it difficult to maintain Tesla’s astronomical stock price.
When Waymo last raised money in October, investors valued the firm at $45 billion. As I write this, Tesla is valued at $1.04 trillion. Tesla’s self-driving ambitions are a major factor in that valuation, as many investors expect Tesla to race ahead of Waymo in the self-driving race.
If investors conclude that won’t happen, they could revise Tesla’s stock dramatically downward. But I wouldn’t necessarily count on it—Musk is very good at keeping Tesla shareholders excited about the next big thing.
"Tesla now has a teleoperations team responsible for “providing remote access to our robotaxis and humanoid robots.” Tesla has apparently discovered what Waymo learned more than six years ago: even the best self-driving software occasionally needs remote assistance."
Why shouldn't self-driving companies rely on remote operators to handle edge cases (<1% of the time)? It's important to avoid mishaps like cars driving in circles or holding up traffic, which give people a bad impression of self-driving cars.
As you mentioned in an earlier post, Tesla has an even bigger problem: FSD needs help from the driver very often....as I recall once every 1000 miles. That's probably both because they're behind on software and they still haven't figured out they need lidar. Maybe those things will change, but for now Tesla has a long way to go.