EV fleets – PJFP.com

Uber CEO Dara Khosrowshahi sat down with Patrick O’Shaughnessy on the Invest Like the Best podcast for a long, candid conversation about the forces remaking transportation. There is artificial intelligence inside the company, and there is physical AI out in the real world, meaning autonomous vehicles, robotaxis, and delivery drones. He calls the autonomous opportunity another trillion dollar marketplace and argues it will change how society operates. You can watch the full interview here. What follows is a structured breakdown of the most useful ideas, the strategy behind Uber’s AV bet, and the operating philosophy that runs underneath all of it.

TLDW

Dara Khosrowshahi explains how he brought order to the chaos he inherited at Uber in 2017 by treating hard problems like vector mathematics, and how an immigrant childhood shaped his all-in, low-stress operating style. He describes AI hitting Uber on two fronts at once: much larger digital models that predict rider intent, and physical AI that changes how rides and food get fulfilled in the real world. The conversation covers Uber blowing through a full year of AI budget in a single quarter, metering headcount as engineers become superhuman, the more than 30 AV partnerships with Waymo, Nuro, Lucid, Nvidia, Wayve, and Pony AI, and why supply, not demand, is the whole game. It runs through the coexistence model borrowed from travel and Uber Eats, the Uber One membership flywheel at 50 million members, the push from on-demand to planned travel through hotels and Uber Reserve, the economics of cheaper autonomous cars and delivery drones, the regional race from the Middle East to Europe, and the lessons from Barry Diller and Herbert Allen about getting to ground truth and betting on people. It closes on his capital allocation philosophy of prioritizing organic growth and AV commitments over buybacks.

Thoughts

The most underappreciated line in the whole interview is the budget one. Blowing a full year of AI spend in a single quarter is the clearest signal yet that frontier intelligence is being consumed far faster than even an AI-native company planned for. Dara’s response has quietly become the default enterprise playbook: explore on the expensive frontier models, then scale the proven interactions onto cheaper or open-source models. The deeper tension is that he is simultaneously telling teams to drive adoption and metering headcount, which is the real story of AI in large companies. The productivity gains are showing up as fewer hires, not only as faster shipping.

The supply-first framing is the strategic core, and it inverts the demand-first logic he learned at Expedia. In autonomous vehicles this means Uber does not need to win the self-driving race itself. It needs to own the demand layer and aggregate every AV maker’s supply, the same way online travel agents coexist with hotels and Uber Eats coexists with McDonald’s. The 30 percent higher utilization figure for AVs on Uber’s network is the wedge in that argument. It is the reason a Waymo stays on the platform even while building its own brand, because filling more of an expensive asset’s day changes the entire return on the car.

His premortem answer is unusually honest. Asked what kills the opportunity, he does not name an Uber-specific execution failure. He names AI’s unpopularity with the general public. That is a CEO admitting the gating factor is social license, not technology. The early data he leans on, drivers in Austin and Atlanta earning more and signing up in greater numbers as AVs add incremental demand, is the counter-narrative he is betting the public conversation on. Whether that story holds as AV volume scales from thousands of vehicles to hundreds of thousands is the open risk the entire industry shares.

Underneath the strategy is one repeated instinct: get to ground truth. It shows up in the Barry Diller story about reading the model from the analyst who built it, in his hunt for the troublemakers who keep a company mutating, and in the fact that he bought an ebike to deliver food in San Francisco. It is the same move applied at every altitude, and it is why he frames AI as a chance to rebuild processes from first principles rather than shave 20 percent off the ones that exist. The leaders who treat AI as an efficiency tool will likely lose to the ones who rebuild from the ground up.

Key Takeaways

Dara took the Uber job in 2017 after Daniel Ek recommended him at the Allen and Company Sun Valley conference and told him, when he hesitated, that life is about impact rather than happiness.
He inherited what he calls complete chaos: a board fighting for control, lost trust with regulators and the public, and a committee running the company after Travis Kalanick stepped back.
His method for chaos is to treat it like vector mathematics, breaking a seemingly unassailable problem into component dimensions and solving each one.
Early moves included bringing in chairman Ron Sugar to unite the board, running a listening tour with stakeholders, and rebuilding the executive team with leaders like Andrew McDonald and Tony West.
He credits an engineering mindset and an immigrant childhood for his calm under pressure. His family lost everything leaving Iran when he was nine and rebuilt from nothing.
On parenting, he argues that overcoming challenges is what forms people, and that doing everything for your kids is a long-term disservice disguised as a short-term favor.
Uber has always operated in a probabilistic real world of traffic, cancellations, and late food, so it has used machine learning longer than most consumer companies.
The current inflection is AI on two fronts: larger digital models that predict intent, and physical AI that changes how Uber fulfills in the real world.
Uber’s feed and search models are now roughly 10,000 times bigger than the older ones, enabling universal search across rides, eats, and grocery in a single query.
Uber can already guess a rider’s destination about three quarters of the time, turning booking into a one-tap interaction.
AI adoption is bottoms-up across engineering, legal, and marketing. Developers in India are driving roughly ten times the code commits using autonomous agents.
Dara pushes teams to rebuild processes from first principles with AI rather than settling for 20 to 30 percent optimization of an existing process.
He wants the rebels and troublemakers to win, and treats unpredictable internal adoption patterns as something to find and promote.
Uber blew through its full-year AI budget in a single quarter, which is now forcing it to meter headcount as engineer throughput climbs.
The token strategy is to explore on expensive frontier models, then scale proven interactions onto cheaper or open-source models.
Uber generates over 10 billion dollars in free cash flow on more than 10 billion trips a year, but it is not a high-margin business, so efficiency funds lower prices and higher earnings.
In autonomous vehicles, the thesis is supply: own the demand layer and aggregate every AV maker’s vehicles, the way Uber aggregates drivers and restaurants.
Uber has more than 30 AV partnerships, including Waymo, Nuro, Lucid, Nvidia, Wayve, and Pony AI.
Uber is building the surrounding ecosystem: depots, charging, fleet partners, a one billion dollar Santander financing line for EV and AV fleets, and autonomous insurance.
AVs operating on Uber’s network are about 30 percent busier in trips and revenue per vehicle per day than vehicles not on the network, which transforms the return on an expensive car.
The build, partner, or buy answer is coexistence, mirroring how travel agents coexist with hotels and airlines and how Uber Eats coexists with McDonald’s, Starbucks, and Chipotle.
His public premortem is that AI’s unpopularity, not Uber-specific execution, is the biggest risk, so the company must move at the pace society will accept to avoid backlash.
Early data in Austin and Atlanta shows drivers earning more and more drivers joining, suggesting AVs are adding incremental demand rather than only displacing humans.
AV hardware costs typically fall 30 to 40 percent per generation. A Lucid midsize built with Nuro could land around 60,000 to 70,000 dollars and bring transportation costs down.
Lower cost expands demand. Uber already dwarfs the taxi market it was once sized against, and Dara expects the same dynamic with AVs.
Traditional OEMs are now investing in L4-ready systems and should arrive over the next two to four years. Each AV drives roughly three to four times what a human driver does.
Chinese manufacturing capability and bill of materials are described as unrivaled. A low-cost Western, Foxconn-style player for AVs is being worked on but does not exist yet.
Drones are gated by battery density. Food and grocery drones should reach real scale in two to five years and become normal in five to ten, with Joby and Zipline cited as examples.
The Middle East, including Abu Dhabi, Dubai, and Saudi Arabia, is moving fastest thanks to entrepreneurial regulators. Europe is catching up, with London robotaxi pilots expected before year end.
Uber Eats wins the number one position more often internationally. The playbook is selection plus reliability, amplified by cross-platform upsell, with about 13 percent of Eats bookings coming from the mobility app.
Uber One has 50 million members growing 50 percent year on year. Dara frames it like Netflix, more content for the same price, and accepts a first-year loss for multi-year profit.
Uber is pushing from on-demand to planned through hotels, via a deal with Expedia, and through Uber Reserve, now at over a 5 billion dollar run rate with 99 percent-plus reliability.
His leadership lessons: from Barry Diller, get to ground truth from source material and tell the truth as a leader. From Herbert Allen, bet on people, not companies.
On capital allocation, he prioritizes organic growth and financialized AV commitments over buybacks, while keeping costs growing slower than revenue.

Detailed Summary

From chaos to structure: the 2017 turnaround

Dara came to Uber from 13 years running Expedia under Barry Diller, recruited through a head hunter after Daniel Ek floated his name at the Sun Valley conference. He arrived into what he describes as complete chaos, with the board fighting over control rather than the fate of the company and trust badly damaged with regulators, the public, and employees. His approach was to decompose the situation the way an engineer decomposes a multidimensional problem, solving each dimension and reassembling the whole. Practically that meant a new chairman in Ron Sugar to unite the board, a listening tour to understand stakeholder concerns, and a rebuild of the leadership team that kept strong insiders like Andrew McDonald while adding people like Tony West.

An engineering mind and an immigrant chip on the shoulder

His wife Sid calls him a robot, by which she means he does not get rattled. He traces that to an engineering education and to a childhood upheaval. His family left Iran when he was nine and lost the business his father had built, and he watched that loss diminish his father over the years. The experience produced a durable drive to rebuild and a refusal to let external chaos define him internally. He applies a similar philosophy to his kids, arguing that challenges and the act of overcoming them are what form a person, and that helicopter parenting removes the very friction that builds capability.

AI inside Uber: prediction, agents, and superhuman engineers

Uber has always lived in a probabilistic world where the digital booking is deterministic but the real-world fulfillment is not, so it adopted machine learning earlier than most consumer companies. The newest models are roughly 10,000 times larger than the prior generation and power universal search and destination prediction that is right about three quarters of the time. Internally, adoption is bottoms-up and uneven in a good way, with engineers in India shipping around ten times the code commits using autonomous agents. Rather than mandate from the top, Dara pushes teams to rebuild whole processes from first principles with AI instead of trimming a fifth off the existing ones.

The cost of intelligence

The flip side of fast adoption is cost. Uber blew through its annual AI budget in a single quarter, and that is forcing a real adjustment. Because engineer throughput is climbing, the company is metering headcount increases rather than simply hiring. The operating rule is to keep driving adoption while pursuing efficiency, using frontier models from providers like OpenAI and Anthropic to experiment with new interactions, then moving the scaled experiences onto more efficient or open-source models to bring the per-token cost down. With more than 10 billion dollars of free cash flow on over 10 billion trips, Uber is not a high-margin business, so efficiency directly funds lower prices for riders and higher earnings for drivers.

Why supply decides the AV race

At Expedia, Dara learned a demand-first model where you attract consumers and then build inventory to match. Uber is the opposite, a supply company, where securing every car, restaurant, courier, and retailer causes the demand to follow. Applied to autonomous vehicles, the strategy is to be the go-to-market and demand layer for anyone building a digital driver. Uber wants to aggregate the largest pool of AV supply, just as it aggregates human drivers, so that the companies building the actual self-driving software can focus on the driver while Uber handles distribution and utilization.

Building the ecosystem around the digital driver

Uber now has more than 30 AV partnerships spanning Waymo, Nuro, Lucid, Nvidia, Wayve, and Pony AI, and it expects many winners rather than one, the same shape as the foundation model market. Around those partners it is assembling the connective infrastructure: depots and charging in cities where the regulatory path is opening, fleet partners, a one billion dollar financing line with Santander for EV and AV fleets, and work on autonomous insurance. It is also collecting street data today that can feed the models, so that when a partner’s cars hit the market there is instant demand waiting. The early proof point is that AVs on Uber’s network run about 30 percent busier than comparable vehicles off it, which materially improves the return on a costly car.

The premortem and the public’s patience

Asked what derails the opportunity, Dara points outward rather than inward. The risk is that AI is powerful but unpopular, and the average person experiences it as a threat to electricity costs or a cousin’s job rather than as magic. The same dynamic could hit AVs even though the technology should end up safer than human drivers, which is why questions about emergency services, equitable access, and driver earnings have to be worked through with regulators and communities. The encouraging early signal is in Austin and Atlanta, where drivers are making more money and more are joining because AVs appear to be adding incremental demand. The controllable risk, he says, is access to supply, which is exactly why Uber has partnered with nearly every AV provider across mobility, delivery, and freight.

A trillion dollar marketplace: cheaper cars and delivery drones

Dara sizes the autonomous opportunity as another trillion dollar marketplace. As AV software and hardware costs fall, typically 30 to 40 percent per generation, a Lucid midsize built with Nuro could come in around 60,000 to 70,000 dollars, which starts to lower the real cost of transportation. History says lower cost expands demand, and Uber already became multiples larger than the taxi market it was once compared to. Manufacturing scales from hundreds to thousands to hundreds of thousands of vehicles, each driving three to four times what a human does, with traditional OEMs investing in L4-ready systems over the next two to four years and Chinese manufacturers setting the bar on cost and quality. Delivery drones are further out, gated mainly by battery density, but should reach real scale in two to five years and feel normal in five to ten.

Membership, hotels, and the shift from on-demand to planned

Uber Eats often reaches the number one position internationally by nailing selection and reliability and then layering on cross-platform advantages, with roughly 13 percent of Eats bookings flowing from the mobility app. Uber One, at 50 million members growing 50 percent year on year, is the loyalty engine, and Dara likens it to Netflix in that members get more for the same price. He explains the membership economics through Amazon Prime, accepting a money-losing first year to earn multi-year profit as members spend more across services. The newest expansion is travel: hotels through a deal with Expedia, and a broader move from Uber’s on-demand brand toward planned bookings, proven out by Uber Reserve at a 5 billion dollar-plus run rate and 99 percent-plus reliability. The end state he wants is a trip where Uber pre-books your ride to the airport, knows your hotel, and brings in-market magic to the whole journey.

Operating philosophy: ground truth, troublemakers, and capital allocation

The mentors thread through everything. From Barry Diller, with whom he worked for more than 20 years, he took the discipline of getting unfiltered truth from the source, illustrated by Diller insisting on hearing the Paramount LBO model from the young analyst who built it. From Herbert Allen he took the lesson to bet on people rather than companies, because great people stay great across cycles. In his own practice that becomes radical transparency, a deliberate hunt for the troublemakers who act as the mutations that keep an organism from dying, and a willingness to be wrong, since learning, often through pain, is what he finds interesting. On capital, he treats allocation as an art, prioritizing organic growth, which took Uber Eats from under a billion to over a hundred billion in gross bookings, then AV commitments that can be financialized, with buybacks coming after growth rather than instead of it.

Notable Quotes

“I know who I am, and I’m always going to be that same person. I’m not going to let the chaos of the world affect me mentally.”
Dara Khosrowshahi, on why crisis does not rattle him

“We blew through our AI budget in a quarter, you know, for the whole year essentially. And it is forcing us to adjust.”
Dara Khosrowshahi, on the real cost of AI adoption at Uber

“What’s magical now is going to seem normal to all of us 10 years from now.”
Dara Khosrowshahi, on how fast riders stop noticing autonomous vehicles

“We think it’s another trillion dollar marketplace.”
Dara Khosrowshahi, on the scale of the autonomous vehicle opportunity

“If we do that, the demand will take care of itself.”
Dara Khosrowshahi, on why Uber obsesses over securing supply first

“I’m looking for those mutations. I’m looking for those troublemakers constantly.”
Dara Khosrowshahi, on keeping a large company adaptive

“It’s the filtering that gets the edge out of the story or out of the situation. And it’s often the edge that gives you an edge.”
Dara Khosrowshahi, on a lesson from Barry Diller about going to the source

“If I’m not wrong, if I’m not making mistakes, it’s just not very interesting.”
Dara Khosrowshahi, on why learning, often through pain, drives him

“Meeting her and seeing her operate, I think, finally allowed me to be the person I want to be versus the person I thought I was supposed to be.”
Dara Khosrowshahi, on his wife Sid, when asked the kindest thing someone has done for him

The throughline is that Uber intends to be the demand layer for autonomous transportation the way it became the demand layer for human drivers, while rebuilding its own operations around AI from first principles. Whether the public grants the industry enough patience is the open question Dara keeps returning to. Watch the full conversation here.

Tag: EV fleets

Uber CEO Dara Khosrowshahi on AI, Autonomous Vehicles, Robotaxis, Drones, and the Future of Transportation