Uber CEO Dara Khosrowshahi on self-driving's future, changing business model, job displacement

Summary

• Uber’s autonomy strategy is to be a demand platform for more than 20 partners, subject to safety and economics. Khosrowshahi wants AVs to be “multiple times safer than a human being,” which he says is achievable; he also argues autonomy could eventually save millions of lives, lower mobility costs and expand the on-demand market. Falling LiDAR costs—from $20,000-$30,000 five or six years ago to $300-$500—support the economics, though vehicles remain expensive.
• Khosrowshahi argues robotaxis reinforce Uber’s marketplace moat because demand density drives utilization. A vehicle on Uber’s network might travel three minutes to collect a 10-minute ride versus 15 minutes for a standalone fleet, producing more “revenue per car per day.” His pitch to Tesla is that Uber is currently “your ticket to the maximization of that revenue.”
• Uber expects to take balance-sheet risk on fleets temporarily, then move the assets to financial investors. Khosrowshahi’s 10-year model resembles hotel brands: Uber supplies demand while outside owners finance fleets and maximize utilization across networks. Uber will use its balance sheet to prove the model, then move those assets off balance sheet.
• The $20 billion buyback does not signal retreat from autonomy investment. Uber generated over $8.5 billion of cash flow in the past 12 months, with top line growing 18% and bottom line 35%; Khosrowshahi says it can “walk and chew gum at the same time.” The host projected autonomy taking ridesharing from 1-2% of global rides to 20%, but Khosrowshahi did not endorse that specific forecast.
• Uber’s distribution thesis extends from robotaxis into eVTOLs, sidewalk robots and drones. The company invested in Joby, while Khosrowshahi estimates robots and drones together could address “50-plus percent” of delivery TAM. Getting food out of restaurants and into apartments leaves the other half unresolved.
• Management sees little driver displacement for five to seven years but a serious societal problem over 10-15 years. Uber can slow recruitment as AVs arrive; Khosrowshahi says Austin’s incumbent drivers still earn as much or more after Waymo’s launch. Beyond that transition, even new work such as AI labeling does not provide a complete answer: “I don’t have a neat answer for it.”

Deep dive

1. Safety—not sensor ideology—is Uber’s admission ticket

• Khosrowshahi says Uber has over 20 autonomy partners across mobility and delivery, with Waymo “the best of the best” in Austin and Atlanta. Asked how many Chinese Level 4 services run without safety drivers, he did not quantify; he cited Baidu, WeRide and Pony.ai and called their safety record “excellent.” More Texas deployments arrive in this year’s second half, initially with safety drivers expected to come out “this year and especially going into next year.”
• His Tesla-versus-other-AV framing is architectural: redundant cameras, radar, LiDAR, HD maps and heavier onboard compute are “cheat codes or good engineering” that simplify perception. Tesla chooses camera-only, no HD maps and tighter compute—cheaper hardware, but “tougher on the software.”
• Uber’s test combines technical approach with evaluation, through a dialogue across different safety approaches: a partner should be “multiple times safer than a human being,” then clear the economic hurdle. Khosrowshahi says Waymo and Chinese players show that safety level is achievable. LiDAR falling from $20,000-$30,000 per unit five or six years ago to $300-$500 for solid-state units helps, though the cars remain expensive.
• He also frames autonomy as a long-term market expander: lower hardware costs could reduce the cost of mobility, make on-demand mobility available to many more people and ultimately save millions of lives.

2. Demand density remains the robotaxi moat

• Khosrowshahi’s defense of Uber’s moat survives removing the driver: a hybrid human-and-AV network already aggregates demand, producing a three-minute pickup for a 10-minute trip instead of a 15-minute collection drive. More revenue-generating miles create higher “revenue per car per day.”
• The Uber Eats analogy carries the logic: McDonald’s keeps a direct channel but also works through Uber, DoorDash and other marketplaces to maximize utilization. Waymo already works with Uber in Austin and Atlanta; riders rate the experience highly and return for the new cars, privacy and “freaking cool” factor.
• His pitch to Tesla’s “digital shepherds” is blunt: Uber is currently “your ticket to the maximization of that revenue,” and owners excluded from its demand pool could under-monetize their vehicles. The host said Tesla is looking to go it alone; Khosrowshahi noted Elon Musk prefers “full stack,” said the market may support multiple winners and expressed interest in partnering.

3. Uber will warehouse fleet risk before finance takes over

• Khosrowshahi’s 10-years-down-the-line end state resembles hotels: Hilton or Marriott supplies the brand while financial owners hold the buildings. Uber can similarly supply demand while pure financial investors own AV fleets and chase utilization across networks. Until then, Uber will take balance-sheet risk based on known market-level revenue, prove the model and later move fleets off balance sheet.
• The host pressed whether the $20 billion buyback should instead fund AV bets, against current ridesharing penetration of 1-2% and his own forecast of 20% with autonomy. Dara’s answer was “not either or”: trailing-12-month cash flow exceeded $8.5 billion, top line grew 18% and bottom line 35%, leaving room for aggressive AV investment over the next three to five years.

4. Autonomous distribution expands beyond the road

• On mobility’s “Z axis,” Uber is a Joby investor and plans to work with the company as eVTOL vehicles become available. Khosrowshahi argues that cities have put homes and businesses into the third dimension while transportation infrastructure remains two-dimensional; the host linked that gap to traffic “getting worse and worse and worse.”
• Slow sidewalk robots such as Serve and Cartken fit deliveries of a mile or less; drones fit spread-out suburbs without high-rises. Together they could cover “50-plus percent” of delivery TAM. The unsolved remainder is literal first and last mile—moving food out of restaurants and into apartments.
• His broader call is categorical: any food business not “deep in delivery” will lose share, while rising labor costs push food, grocery and retail businesses toward roboticization. Uber does not plan to build those systems; it wants to remain the capital-light demand network feeding asset-heavy operators, with selective investments where needed.

5. Driver displacement is delayed, not resolved

• The host brought the political externality into view, citing civil unrest in Wuhan around limits on self-driving licenses and saying Waymos in Los Angeles were “called to their death.” He asked what happens when robots displace the drivers who built Uber, Lyft and DoorDash.
• Dara separates timing from destination: for five years, platform growth should absorb incoming robots. In Austin, Uber can turn down driver recruitment as AVs arrive, while current drivers are making as much or more money after Waymo’s launch; he expects more human drivers and couriers for five to seven years. At 10-15 years, however, “this is going to be a real issue.” Uber is adding work such as AI labeling through Uber AI Solutions, but his endpoint is: “I don’t have a neat answer for it.”