Uber’s Robotaxi Playbook, End of Human Driving & $10B Bet on Robots | Dara Khosrowshahi (Uber CEO)

Summary

• AV demand is arriving before mass production at regular-car prices: in Atlanta and Austin, 80% of riders accept an autonomous match and 20% decline. Dara Khosrowshahi told a 13-year-old that, three years from now, they would still likely want a license because AVs remain “much more expensive than a regular car,” but says that 10 years from now they will not have to drive and will have plenty of choices; within 25 years, humans will be “demonstrably less safe.”
• Uber’s potential AV operating edge is network orchestration at roughly 10 million drivers and 40 million trips every day. Its dispatch system looks five or six seconds ahead rather than simply choosing the nearest vehicle; Dara is “kind of psyched for machines” because they should be more predictable and have higher acceptance rates than humans.
• Autonomous volume remains economically small, leaving insurance pricing and responsibility unsettled. Dara estimated that all AV trips worldwide last year represented less than 1% of the rides Uber added, so insurers still lack the “law of large numbers”; the model may involve the AV provider, platform or a combination, and “all of that is TBD.”
• Uber’s answer to potential Chinese AV disruption is partnership, not retreat from the ecosystem. After Expedia “got our asses kicked” in China, Dara endorsed Uber’s exit but highlighted its unmonetized DiDi stake and work with BYD, Pony.ai, WeRide and hopefully Baidu: “Building bridges to China is the right way forward for us.”
• Dara expects automation to augment more work than it eliminates over the next decade, but leaves the speed of social adaptation open. Companies may automate 20% or 30% of office tasks without reaching 100%, while Uber aims to expand from roughly 10 million people on its platform now to 20 million by 2035; the unresolved risk is that change is occurring “faster than even I expected.”
• Management argues that Uber’s larger cash engine should increase—not suppress—its appetite for failure. The company was losing $4 billion when Dara joined but says it will generate $10 billion of cash flow this year, leading him to insist that as companies grow, their tendency to become conservative is exactly backward: “The exact opposite should be true.”
• Uber’s expansion filter is whether a new business “rhymes” with an existing capability. Rides led to Eats, Freight and air mobility, while AI model design, labeling and testing fit because Uber is also a flexible-work platform; if an idea does not rhyme, Dara says a startup is better equipped to pursue it.
• The physical transition will be slower than the software narrative implies. Dara expects AV growth to move Uber’s network toward EVs because “AVs are EVs,” but inadequate US charging infrastructure remains a constraint and his combustion-engine horizon is still 20 years; eventually, autonomy should make transportation cheaper and extend safe service into today’s transportation deserts.

Deep dive

1. Riders are ready for robotaxis before robotaxis are ready for mass adoption

• Dara said passenger acceptance is already straightforward: when Atlanta and Austin riders learn they have matched with an AV, 80% accept and 20% decline, while those who accept “love the experience.” Regulators “obviously move a little bit slower,” but he sees more excitement than resistance.

• He predicted that a 13-year-old would still want a license in three years because mass-produced AVs remain “much more expensive than a regular car,” but said, “10 years from now, you won’t have to drive and you’ll have plenty of choices.”

• On whether human driving might become illegal, Dara said he did not think the change under discussion would happen in the next 15 years; he then said that “certainly in the next 25 years, humans will be demonstrably less safe.” He favors free will and expects tougher licensing rather than prohibition, with fewer people knowing how to ride a horse; race-car driving should remain because it is “so damn cool.”

• The reality check: Dara estimated that the total of autonomous trips worldwide last year represented less than 1% of Uber’s incremental ride volume. Physical technologies diffuse more slowly than digital ones; “AVs are EVs,” but he still put the horizon for very few combustion engines at roughly 20 years.

2. Uber applies its dispatch intelligence to autonomous fleets

• Uber’s marketplace now coordinates probably 10 million drivers and 40 million trips daily. A hail does not simply go to the closest car: the system predicts the next five or six seconds, perhaps leaving that car free because it expects Peter to hail it and sending another to the current rider to optimize the network.

• Dara is “kind of psyched for machines” because human drivers are unpredictable, can reject assignments and sometimes require redispatch. Machines should state what they will do and deliver higher acceptance rates, although stitching one ride with another into a perfect experience remains a coordination problem.

• His marketplace lesson is supply first. Expedia concentrated on audience; Uber concentrates on liquidity across cars, restaurants and retailers. Build the easiest, most liquid supply, he argued, and—provided there is product-market fit—“the demand just shows up.”

3. Insurance, China and infrastructure remain open layers of the AV stack

• Dara expects layered insurance: the autonomous provider covers the vehicle and takes responsibility for “the quality of their driver,” while another layer protects passengers against everything else. He said Uber would absolutely consider embedding insurance in autonomous vehicles, but whether the model belongs with the driver, platform or a combination “hasn’t been figured out yet.” Credible actuarial pricing needs far more volume.

• An audience member’s challenge was that removing the driver could eliminate an estimated 70% of costs and let cheap Chinese AVs flood emerging markets. Dara’s response was bridge-building: Uber retains a large, unmonetized DiDi stake and works with BYD, Pony.ai, WeRide and hopefully Baidu outside the US.

• Air mobility creates a related real-estate play. Dara expects Joby-style vertiports at airports, city centers and urban edges, designed for multiple vehicles arriving, landing and taking off; Uber’s traffic data can identify sites that maximize business while reducing congestion. On the ground, he wants eventual AV savings extended beyond wealthy cores into transportation deserts.

• An audience member proposed “affordable living as a service”—housing, food, energy, connectivity and mobility for $250 a month—and said it would become a $50 million XPRIZE. Dara offered to discuss Uber’s role in providing transportation to underserved areas.

4. Automation may augment labor, but the speed of transition changed Dara’s confidence

• Dara argued that headlines overemphasize replacement because “the press loves the drama.” His factory example was Chinese OEMs: robots perform more production, but people remain to supervise them and conduct quality checks, even if the plant employs fewer humans than a decade earlier.

• Office automation follows the same pattern in his base case: companies may automate 20% or 30% of tasks, rarely 100%, leaving substantial work. Over the next 10 years he expects “a lot of augmentation” and some replacement, with new work historically emerging as old work changes.

• The hedge matters: AI and automation have advanced over the past five years “faster than even I expected,” despite Dara’s technological optimism. Uber therefore wants 20 million people doing potentially different tasks on its platform by 2035, versus about 10 million now, while private and public enterprise share responsibility for retraining and new work.

• Peter’s related point was that families without a roof, food, water, energy or health security cannot think about what to do next; basic services could give people room to dream. He called Uber a “societal capacitor” able to absorb and release labor as jobs shorten from lifelong commitments to perhaps a week. Dara noted that capital is optimized “sometimes at the expense of labor” and endorsed Peter’s push for workers to own assets.

5. Cash flow gives Uber permission to make larger adjacent bets

• Dara’s cultural anchor is “do the right thing, period”—with “period” literally serving as the value’s description. Employees three or four levels below him must decide whether the right act is speed, caution, risk or folding their cards, preserving the “culture of go-getters” he inherited.

• Scale should expand risk capacity. Uber was losing $4 billion when Dara joined, making a $1 billion mistake potentially disastrous; with $10 billion of cash flow this year, it should place bigger bets and tolerate some failures. Fighting corporate conservatism remains a daily argument with himself, his team and the board.

• New fields must “rhyme with what we do”: rides extended to Eats, bulk movement to Freight and ground mobility to Elevate. AI model design, labeling and testing also rhyme because Uber supplies flexible work; ideas without a shared capability or “right to win” are better left to startups.