Uber CEO Dara Khosrowshahi：自动驾驶的未来、商业模式变化与岗位替代

the driving force behind one of America's most influential companies. Record high today for Uber. 100% up last year. Waymo and Uber have announced a partnership. When you have a CEO that's done what Dara has done, you set the bar higher and higher. The impact we have on society is significant. We hope to keep building on that impact going forward and I'm quite optimistic about what the future's going to bring. Ladies and gentlemen, please welcome Uber CEO Dara Khosrowshahi. [Music]

Good to see you. All right, Dara, I wasn't sure if you were aware, but I was an early investor in Uber.

I've heard you say it once or twice.

I'm curious how my investment's doing.

I don't know.

Okay. Based on today, today's looking pretty good. So autonomy is the discussion I think everybody wants to have in the timeline. How many partners does Uber have in autonomy today?

We have over 20 partners across both mobility and the delivery business. I'd say mobility is now in the field as we speak. Obviously, we've got a partnership with Waymo, who I think is the best of the best in Atlanta and Austin. But there are a number of other players that we are partnered with, including a number of Chinese players.

Autonomy in China is hitting the big time, and a lot of these companies that want to expand outside of China, we're partnering with. In the U.S., even in the second half of this year, we will have a couple of partnerships hit the road in Texas, and then in Europe and the rest of the world. So you will see—we've announced a bunch of partnerships—we're doing a ton of work with these partners. You'll see these cars hit the road with safety drivers eventually, and the safety drivers will come out this year and especially going into next year. We're going to have a significant number of cars on the road.

How many players in China have Level 4, with no safety driver, today? And what's your assessment of those companies and their safety record?

It's a different market, obviously. Baidu, WeRide, and Pony.ai are all on the road today with no safety driver. We are partners with all of them. Their capabilities are amazing. You can imagine driving in China, in these big cities, is quite a complex undertaking. They take safety just as seriously as the Western companies do, so I think their safety record is excellent.

Ultimately, we think autonomy can be both superhuman in terms of safety and can save millions of lives over the course of time on the road. Over a period of time, as the cost of especially the hardware stack comes down, we think that it can bring the cost of mobility down and make mobility on demand available to many, many more people than it is now. So it's going to be a very big market expander for us.

So that was the debate, Dara, that maybe kind of exploded a little bit on X between you and Elon, where you guys, I mean, very respectfully, were just debating the pros and the cons. Maybe just set it up for the folks in the audience: the difference between Elon's approach and the Waymo approach, and maybe the relative pros and cons as you see it.

Yeah. I mean, they're the ones building the cars, so I'm, to some extent, a very, very, very interested bystander. But the way I put it is, Elon's approach depends on excellent software to do a bunch of the heavy lifting. Whenever you're building a product, there may be some cheat codes that you undertake. You could call them cheat codes or good engineering.

Some of the things you see in early systems are camera, radar, and LiDAR—multiple sensors, redundancy on the sensor stack—to make sure that your perception algorithms are seeing the world as it really is. Elon is doing camera-only: tougher on the software, cheaper for the hardware.

The second big difference, I would say, is that many of the players use HD maps. What HD maps do is essentially map out an area so that it's much easier for the software to determine what the permanent aspects of a certain view are—the lines on the road, traffic lights, et cetera. Because of the HD maps, it's very, very easy for that piece of software to determine what's permanent and then what's impermanent: vehicles, people, et cetera. So it makes the job of the software much easier—to figure out what's going on and then determine what to do. Elon's approach doesn't depend on HD maps, and again, it makes the job of the solver harder.

The other significant factor is compute. When you look at the compute in many of the other players, in terms of FLOPs and memory, et cetera, in the back of the car, it's pretty expensive and pretty extensive. I think Tesla's approach is with a much tighter compute stack.

Do you see a world where you try to put your distribution into all those solutions, assuming that everybody's amenable to working with you and it meets your threshold for what you're looking for?

I'd say safety comes number 1.

Exactly.

We have a certain safety case that we want to make sure that our partners adhere to or exceed.

And sorry, is that an eval, or is that certain rates that they have to publish to you? How do they demonstrate it to you?

It's the technical approach and the eval together. And listen, it is a dialogue, right? Different people take different approaches to safety. We want to make sure that, showing up on the Uber platform, it is as safe as it can be.

Our definition of safety is multiple times safer than a human being, which is achievable. Waymo is showing that it's achievable, and many of the Chinese players are showing that it's achievable as well.

If it meets our safety criteria and the economics are attractive, then we'll do business with them. The economics improve as the cost of hardware comes down. LiDAR was $20,000 to $30,000 apiece 5 or 6 years ago. Now solid-state LiDAR is $300 to $500 apiece. So the cost of hardware is coming way down.

It is going to need to continue to come down because these cars are very expensive. We want to be the platform, and we want to essentially help the entire AV ecosystem thrive. We think there's enough economics for the network player to have a great business, and for the software providers and the vehicle owners to have great businesses.

Then, obviously, there's fleet operations in terms of housing the cars, recharging the cars, and all of the on-the-ground work that's necessary as well.

When you get to a tipping point—I'm going to assume in driverless miles, let's say—one of the most interesting things I've thought of is whether you could tell a city how it should actually be designed for optimal traffic.

I wouldn't say we work for optimal traffic. I think, theoretically, it's possible, but that's something Google and lots of other players can help with. We're certainly helping cities in terms of where you should put charging infrastructure, for example, parking, drop-offs, et cetera, to help traffic flows.

I think we can be a partner for cities, and we do have a small operation where, essentially, we offer data for free for cities to embark on city planning, so to speak.

Do they take it?

Some do. Some of the more sophisticated cities take it, but I wouldn't call it a big part of our business.

So, Dara, I want to ask you about the business-model impact of basically robotaxis, or self-driving. In the old world, Uber's network effect was a marketplace effect, where you connected drivers and riders. If you had the most geographic density in an area, then you could promise riders faster pickups, and the drivers got higher utilization. That was a very powerful network effect.

But we're moving into a new world where anyone who has a fleet of self-driving cars, in theory, could just make them available to the public and start competing. How do you see that impacting your model?

Do you have to go from being an asset-light business to now owning all these cars and deploying them? Is that a good thing or a bad thing for your business?

I think the same economics apply, right? A fleet owner is not going to have as many vehicles available in a certain market as, let's say, a network like ours, and we will have a hybrid network. We're going to have humans and autonomous cars together, and that's going to continue for a while. The autonomous machines aren't going to replace all humans, at least for the foreseeable future.

For us, if you're part of our network, you're going to get more requests than the player who's doing a standalone operation because we already have the demand. The requests are going to come from much closer. Instead of a pickup that's 15 minutes away for a 10-minute ride, you're going to get a pickup that's 3 minutes away for a 10-minute ride. So the utilization—in terms of revenue-generating miles as a percentage of total miles driven—is much, much higher on our network.

If you have fleet player A going direct as a standalone, and fleet player B working with us, fleet player B will have much more business and many more miles creating revenue as a percentage of the total miles driven. As a result, each of their cars is going to get much more revenue per car per day than the fleet player who isn't working with us.

Ultimately, even if you think about Uber Eats, there's this drama: Do you go direct only, or do you work with a marketplace? The fact is, every major food player—McDonald's has a direct channel—but they have a brand, and they want that brand to create as much revenue as possible.

So they have a direct channel, and they work through our marketplace, DoorDash's marketplace, and other marketplaces as well, because that's how you drive utilization.

I think that most of these players—there are going to be some players like Waymo and Tesla who can build their direct channel—but we think if they want to drive maximum economics out of these really expensive cars for now, they're also going to want to work with us.

Do you think you will need to buy and deploy your own fleets, or can you rely purely on third-party fleet owners?

Ultimately, if you look at the end state, I think all of these cars are going to be financeable. If you look again at the hotel business—I used to be in the travel business—a Hilton or a Marriott, who's the brand, doesn't own any of their hotels. Those hotels are owned by financial-only players.

I think 10 years down the line, there are these things called REITs, real estate investment trusts. You're going to have fleets. You're going to have financial owners that own big fleets of cars that are on our network, maybe on other networks.

The new Enterprise kind of thing.

I'd say it's going to be more financial players. Hertz and Enterprise are operators. These are going to be like Blackstones of the world, and they own fleets and are just trying to monetize those fleets as much as possible. That's the end state.

Between now and the end state, we will take balance-sheet risk because we can sign up. We know exactly how much revenue a car can produce in a given market because cars are already producing revenue. We can sign up for the revenue. We will prove out the business model, use our business and our balance sheet to prove out the business model, and then at some point the whole thing is going to get financialized and we'll be able to take it off balance sheet.

Is Waymo willing to work with you?

Actually, Waymo is working with us now in Austin and Atlanta.

Okay.

In Austin and Atlanta, if you're using Uber, you can be picked up by a Waymo. Our customers love it.

Is it the driverless aspect of it that they love?

I think one thing is that they're new cars. They're really nice cars.

Yeah.

It's kind of freaking cool.

Yeah.

You do have privacy in that car as well. I think the combination of it works out really well. We see customers who experience the product rate it really highly and use it again. It's just an absolutely dynamic product.

We've mostly only spoken about the X-Y axis, and we have a couple of our friends who've built businesses that are trying to launch these eVTOL businesses, and some of our other friends who are experimenting with small drone delivery. Tell us where all of those things play in your infrastructure going forward.

We're absolutely believers in eVTOLs. We're an investor in Joby, and we are going to work with them as those vehicles become available. We know that there are some other vehicles, but I think that the Z axis, if you want to call that, makes a ton of sense.

Listen, cities of the world have essentially been built in the third dimension because there's only so much that you can expand in the X and Y dimensions. Businesses have expanded in the third dimension. Residences have expanded in the third dimension. But our transportation infrastructure has only expanded in 2 dimensions.

So it's no wonder that traffic just keeps getting worse and worse and worse, because that third dimension isn't available.

We are absolutely believers in both eVTOLs and drone delivery. On the delivery side, there are 2 areas that we're working on. One is sidewalk robots. It's easier technology to develop.

Explain what that is—a sidewalk robot.

Sidewalk robots—there are some of them in Los Angeles and Santa Monica. They are autonomous vehicles that drive on the sidewalks. They drive pretty slowly, and they're very, very safe. They look kind of cute, and they're appropriate for deliveries that are a mile or less. They're for deliveries in a tight space, so there's a certain addressable market for us where those sidewalk robots work.

We're working with Serve, Cartken, and a number of other players in the U.S., in Japan, and in a number of other markets. Then on the other side is drone delivery. Drone delivery is appropriate for markets where they're more spread out—suburban, with no high-rises, et cetera. Those 2 together, we think, can cover 50%-plus of our delivery TAM, so to speak.

But then there's another 50% that we're going to have to work on in terms of the first and last mile—coming out of the restaurant and then getting the food into your apartment as well. Humans take care of their own first and last mile, but you need something to take care of the first and last mile of the food. That's where the challenge is going to come in, and we're working with a number of players to see how we can get that first and last mile for food.

I want to talk to you for a minute, if I may, about the balance sheet. One of the great early insights we had at Uber was around profitability. The press and the narrative was, “Oh, Uber could never be profitable.” I would talk to TK, William, and all the guys in New York about it, and they're like, “Yeah, we could flip it at any moment in time to $2 more a ride. We would lose no rides, and it would be wildly profitable.”

In fact, under your stewardship, Uber has become a money-printing machine, to the point at which you announced a $20 billion stock buyback.

Yes.

I saw it and I said, “Wow, this is just incredible.” However—

Did you tweet about it by chance?

I might have. Once in a while, I'll retweet you and give you a little shine. But I did have this thought: I had Chris from Neuro on the program, and you have this great partnership to put 20,000 Lucids on the road.

Wrong podcast, but keep going.

The other podcast. I'm wondering how you think about the war chest, the money-printing machine, and deployment of that asset. How do you decide between a $20 billion stock buyback and putting $300 million into Nuro? We had Travis on the podcast, and he said he's had many opportunities to look at things like Pony.ai, which has been in the press. It would be pretty great to have the original founder—I don't know, you've got a couple of billion laying around—maybe help him have Pony.ai come to the West.

How do you think about deploying that capital in order to continue to grow? Where are we at—1% of rides globally are ridesharing, approximately?

A little more, but it's between 1% and 2%. It's a very low number.

It's clear it's going to go to 20% with autonomy. If we all believe that, and that's obvious, is that the best use of the capital? How do you make that decision?

The good news for us is it's not either-or. We can walk and chew gum at the same time. In the past 12 months, we've had over $8.5 billion of cash flow. The business is growing 18% top line and 35% bottom line, so that cash flow is going to grow by a lot over the next 3 to 5 years.

We announced the $20 billion buyback because, in looking at areas in which we could invest aggressively—for example, in autonomous vehicles, because we should, because it's an enormous opportunity, whether it's vehicles or fleets, et cetera—we are very comfortable that we've got enough capital to be super aggressive there appropriately and, at the same time, buy back our stock.

There's a great company we know of. The management team can get a little better, but they're okay, and we think it's a great deal. So it's not an either-or. It's an and for us, and we're lucky to be in that position at this point.

You have a very big business in Uber Eats. It competes with folks like DoorDash.

Mhm.

When Travis was on the pod a few weeks ago, maybe a month ago, he talked about the robotization of food and all of that. Can you just talk to us about your vision of where all of that stuff goes?

We actually work with Travis and his CloudKitchens business. He's also built a restaurant technology business in Otter.

I do think that any food business that is not deep in delivery is going to lose share, period, for the foreseeable future. Every food player, grocery player, and even retail player has to get into delivery and on-demand delivery. Otherwise, they're missing the most attractive segment of consumers out there.

As the cost of labor is going up, all of these businesses are increasingly investing in roboticization. It's not something that we're getting into, but as more healthy, delicious food becomes available at lower prices, our delivery business will benefit a lot.

I'm hearing consistently from you—and you can just tell me if this is wrong—that you are becoming increasingly an asset-light, highly liquid distribution network. You have this incredible network effect. You have these hundreds of millions, maybe approaching a billion, users, and you can just pour them into all of these things.

We essentially bring demand to the assets that are driving the movement of people and things, food, and groceries, and these are all asset-heavy businesses. The next incremental piece of demand that comes from our network is incredibly valuable for them, and we can do so while staying largely capital-light.

At the same time, to the extent that I can use my capital to invest in the AV ecosystem or fleets, et cetera, we can also do that.

There's one company that wants to go on its own. A friend of ours runs it. I think you've probably had some conversations—obviously, you have. What's the best pitch to Elon to put 100,000 robotaxis into the Uber fleet while still doing his own thing? His app is doing spectacularly well, and the pilots are doing well, so he'll obviously figure it out. But what's your best pitch to him to join the Uber network?

I think the pitch is simple: if you're looking to maximize the revenue of those robotaxis—

Today?

Today, we are your ticket to maximizing that revenue. To the extent that you're looking to have these fleets owned by people—you know, the digital shepherds—which is an amazing vision that Elon has, those owners will under-monetize if they're not able to monetize their assets on the Uber network. If there's a competitor offering those vehicles on the Uber network, the monetization of those vehicles is going to be superior, and those digital shepherds are going to go elsewhere. So I think that's the pitch. Again, Elon kind of believes in full-stack.

Yes.

He's proven it, and I think this market is large enough for there to be multiple winners. In the end, we would love to partner with them.

But at this point, they're looking to go it alone, and I think the market is large enough to carry a number of winners in this.

I see a tough question about humans. I was talking to Will Barnes, who ran Uber originally in Los Angeles and then half of the country for Travis. Will Barnes had a pretty amazing insight, because in the early days, we had humans protesting humans competing for—taxi drivers versus rideshare drivers.

In China, in Wuhan, in fact, there's been a lot of civil unrest, and they're talking about limiting the number of licenses for self-driving cars because of the disruption that would happen if young men who have those jobs are not able to have a job. We saw the Waymos get called to their death here in Los Angeles, and that was a pretty clear message as well. How do you think about that group of people losing their jobs? These are the drivers who built the Uber network, who built Lyft, who built DoorDash, and China's overwhelming concern about this.

These are robots taking human jobs, and there's a lot of discussion about this. I think maybe in the tech industry we don't talk about it head-on.

I think this is a big issue for AI in general and for job displacement. You see it with younger graduates as well. For us, at least for the next 5 years, the number of robot cars coming onto the platform is not going to displace people, because the platform is just growing so quickly that we can very easily take that demand.

In a market like Austin, or other markets in which we're launching autonomous vehicles, we'll turn down the driver-recruitment machine so the robots can come in, and the drivers who are currently driving on the platform can make as much money. Austin drivers are now making as much or more money than they were before we introduced Waymo.

For the next 5 to 7 years, we're going to have more human drivers and delivery people, just because we're growing so quickly. But I think 10 to 15 years from now, this is going to be a real issue. Jason, I don't have a neat answer for it.

Now we're finding other kinds of work. We've got drivers and couriers labeling AI data. We have a whole Uber AI Solutions business. One way to look at Uber is that we are a platform for work. Transportation is the first kind of work, and now we're expanding into other kinds of on-demand work as well to be able to adjust the kind of work available to people who want to earn on our platform.

But I think long term, this is a big, big societal question that we're going to have to struggle with, and lots of others are going to struggle with it too.

Absolutely.

All right. Thank you very much.

Thank you. I really appreciate it.

Great. Crushed it. Thank you, my brother.