Summary

Khosrowshahi’s core AV thesis: supply is key to winning, and Uber is building the ecosystem to do it — 30+ partnerships (likely Waymo, likely Nuro, Lucid, Nvidia, likely Waabi, likely Wayve, likely WeRide, likely Pony.ai), depots, charging, fleet financing (a just-announced $1B financing line, likely with Santander, for EV/AV fleets), and autonomous insurance. The proof point: AVs on Uber’s network are “30% or more busy” than vehicles not using the network in trips and revenue per vehicle per day — “that 30%… can make a huge difference in terms of your ROI of investing in these expensive cars.”
He sizes autonomy as “another trillion dollar marketplace,” with the same underestimation error as early Uber: “people were sizing the market based on the taxi market. We’re multiple times bigger than the taxi market.” Hardware costs usually fall 30-40% per generation; the Lucid midsize being built with likely Nuro “would be a $60-$70,000 car,” and each AV probably drives 3-4x what a human does.
Uber blew through its full-year AI budget in one quarter and is responding by metering headcount growth as engineer throughput goes “superhuman.” The playbook: frontier models (OpenAI or a cloud model) for exploration — “exploration is go go go” — then swap in cheaper or open-source models once experiences scale. Context: $10B+ free cash flow on over 10 billion trips a year — “we are not a high margin business.”
On the AV foundry question: Khosrowshahi expects traditional OEMs to get to L4-ready systems in 2-4 years, but “the Chinese capabilities in terms of manufacturing both in terms of quality and cost at this point is unrivaled” — a Western low-cost Foxcon equivalent “is being worked on but we’re not there yet.”
His premortem isn’t Uber-specific — it’s public backlash against AI and AVs: “we’ve got to go at the pace that society is prepared for us to move otherwise there will be a backlash.” Early counter-evidence: in Austin and Atlanta, where likely Waymo partnerships run, drivers are earning more and driver sign-ups are rising because AVs “are actually adding incremental demand to the platform.”
The super-app flywheel is compounding: Uber One is at 50 million members growing 50% year-on-year, 13% of Eats bookings come from the mobility app, Reserve went from nonexistent to a $5B+ run rate at 99%+ reliability, and the new Expedia hotels deal hands “the vast majority of the economics” back to members. Membership follows the Amazon Prime pattern: lose money on a member in year one, make it back in years 2-4 — “solidly profitable now.”
Capital allocation, asked “are you Amazon or are you Apple?”: “somewhere in between” — organic growth first (Eats went from under $1B gross bookings to over $100B under his tenure), then AV commitments in the tens of thousands of vehicles (financialized), with buybacks from what’s left. “If you’re building the company right, you’ll do both.”
Management edge, via Barry Diller: go to the source, because “it’s the filtering that gets the edge out of the story… it’s often the edge that gives you an edge.” Companies are organisms that “evolve by mutating” — he hunts troublemakers as mutations, and says AI has accelerated the internal rate of change “by 5x.”

Deep dive

1. “Since when is life about happiness? It’s about impact” — how Dara took the job

After 13 years as Expedia CEO working with Barry Diller, a headhunter’s cold call about the Uber job got a “no effing way.” At Sun Valley, Daniel Ek revealed he’d recommended him — and when Dara protested he was happy at Expedia, Ek answered: “since when is life about happiness?… It’s about impact. Uber is a company that has impact on the world. It’s important. It’s in trouble.” The next morning Dara called the headhunter from the parking lot.
Day one was “complete chaos”: Travis gone, an executive committee running the company for months, the business “hugely competitive” and the company “in the public sphere in many many bad ways.”

2. Order from chaos is vector mathematics

His HBS-seminar answer for stabilizing a broken company: decompose the unassailable problem into dimensions. “It’s vector mathematics… if you solve each dimension and you bring it all together, you can actually solve pretty complex issues.” At Uber the dimensions were distinct: a board “fighting for control” (“focused on who’s going to control the future of the company versus what the future of the company was going to be”) fixed by bringing in chairman Ron Sugar; lost trust with regulators and the public fixed by a listening tour then acting on it; and a talent reset — retaining talent like Andrew McDonald, adding Tony West, while “some folks who were stuck in the old world had to get off.”
On stress: his wife Sid “calls me a robot.” The root is biographical — his family lost everything leaving Iran when he was 9, and he watched it destroy his father, a “giant of a man” who “just lost that spark” and couldn’t begin again in the US. The lesson he took: succeed, but never let fortune break you. “Being stressed out — what’s the point of it? Who cares?”
The immigrant chip on the shoulder “is never satisfied” — and he extends it to parenting: “we’re doing our kids a disservice by giving them too much… it’s the challenges in life that form you… A happy life is not necessarily an easy life.”

3. AI adoption inside Uber: promote the rebels, meter the headcount

Uber is “AI native, or ML native” — the business is digital interaction with probabilistic real-world fulfillment (traffic, cancellations, late food), so it has run AI in production longer than most. His mandate isn’t top-down: he pushes teams to rebuild processes from first principles, not to let AI “optimize 20%, 30%” of an existing one. The adoption pattern is unpredictable — “devs in India who all of a sudden are driving 10x the code commits,” running autonomous agents everywhere — “it’s up to us to promote those individuals who are the rebels inside the company. We want the rebels to win here.”
On the cost of intelligence: “We’re dealing with it now actually. We blew through our AI budget in a quarter… for the whole year essentially.” The response: keep driving adoption, but meter headcount increases as engineers become “superhuman in terms of their output,” and tier the model stack — “the more expensive models to explore… whether it’s an OpenAI model or a cloud model… once we scale some of these experiences we’ll look to bring in more efficient models… or open source.”
The efficiency imperative is structural: “we’ve got over $10 billion in free cash flow, which is great, but it’s on well over 10 billion trips a year… we are not a high margin business” — efficiency funds lower rider prices and higher earner pay.

4. Bigger models are “boring but wonderful” — and they already know where you’re going

The unglamorous near-term AI win: Uber’s feed and search models are “probably 10,000 times bigger than our older models.” Universal search now surfaces rides, Eats, and grocery from one query, and “3/4 of the time we can guess where you’re going and it’s just a one-tap interaction.”
On interfaces in seven years: “I think you’ll be talking to your apps” — inbound interaction goes unstructured via agents, while apps persist for output (“it’s not compelling to say your Uber is six minutes away” versus watching the car). The deeper shift: UIs were historically “optimized for the overall average… Now we’re going to personalize it. When you land in a city, Uber is going to be different for you than when you’re going to work.”

5. The AV playbook: everything is upside down — supply first, demand takes care of itself

The lesson from moving Expedia→Uber: Expedia was demand-first; “At Uber, everything is upside down.” Growth means recruiting drivers, merchants, and couriers first — not the top 10 cities but “the next 50 cities, the next 200 cities” — and “the demand will take care of itself.” Asked what has to be true for Uber to win the AV demand-aggregator slot, his one-word answer: “Supply.”
Uber now has 30+ AV partnerships — likely Waymo, likely Nuro, Lucid, Nvidia (building compute, sensors, and now a software driver), likely Waabi, likely Wayve, likely WeRide, likely Pony.ai — on the thesis that AVs will mirror foundation models: “there isn’t going to be a single winner… there’ll be open-source smaller models as well.”
The offer to AV builders: focus on the driver, Uber builds everything else — depots and charging in regulation-friendly cities, fleet partners, financing (the $1B financing line, likely with Santander, for EV and AV fleets), autonomous insurance, street data fed back to the models, and “when they hit market we’ve got instant demand.” The kicker: AVs on Uber’s network are “30% or more busy” than vehicles not using the network per vehicle per day — decisive for “your ROI of investing in these expensive cars.”
His product observation from riding them: “how quickly magic turns to normal.” First Uber ride ever: “push a button and a car shows up in 5 minutes… Next day, you’re like, ‘Where’s my damn car?’” AVs are the same — magical for two minutes, then routine. “What’s magical now is going to seem normal to all of us 10 years from now.”

6. Waymo is partner and competitor — and that’s fine

Patrick’s pushback: why partner with likely Waymo, which could become a direct competitor at network scale? Dara’s answer comes from travel: OTAs compete with Marriott and Delta, who still want incremental demand “because they have a big box to fill” — “the returns for a hotel that has 70% of its rooms filled versus 90%, it’s night and day.” Same in Eats, where McDonald’s, Starbucks, and Chipotle both list and compete. “I don’t think it’s going to be black or white… There’s going to be an amalgamation of business models.” Likely Wayve, by contrast, licenses one end-to-end model to OEMs and Uber “will be responsible essentially for all the demand”; the coexistence logic extends to likely Zoox and likely Nuro.
His industry-level premortem: AV tech “is going to be safer than human beings,” but AI is already unpopular — “if it’s driving up my electricity cost or if it’s going to cost my cousin’s job, that doesn’t feel that good.” Hence: “we’ve got to go at the pace that society is prepared for us to move otherwise there will be a backlash.” The Uber-controllable version of the premortem is the same as the thesis: losing access to supply.
The early evidence is benign: in Austin and Atlanta (both big likely Waymo partnerships), Uber drivers “are making more money” and driver sign-ups are increasing — “it looks like AVs are actually adding incremental demand to the platform.”

7. The trillion-dollar math: cheap hardware, Chinese Foxcons, and drones that finally work

The success scenario “is another trillion dollar marketplace”: hardware costs “usually” drop “30 to 40% per generation,” the Lucid midsize built with likely Nuro “would be a $60-$70,000 car,” and cheaper transport expands demand — early Uber was sized against the taxi market and ended up “multiple times bigger.” Each AV “probably drives three to four times what a human does,” so fleet efficiency compounds the economics.
On a Foxcon-for-AVs: “It’s happening as we speak.” Traditional OEMs, seeing L4 as “much closer than they think,” are expected to get there “over the next 2 to 4 years”; production goes from hundreds and thousands to hundreds of thousands. But “the Chinese capabilities in manufacturing, both in terms of quality and cost, at this point is unrivaled” — a Western low-cost equivalent “is being worked on but we’re not there yet.”
Why no drone has delivered Patrick’s sandwich a decade after the Amazon video: “battery density” — the battery must lift itself plus payload with the requisite range. Joby is building drones for people now; food and grocery drones “start hitting real scale over the next two to five years,” initially costing more than human delivery. “Two years from now? No. But 5 to 10 years from now it’s going to become more and more normal” — and a 10-15 minute delivery replacing 25-30 “will change how society operates.”
Regionally: Patrick says the Middle East “is going fast”; Dara says services are happening today in Abu Dhabi and Dubai, with entrepreneurial regulators. The US moves via California and Texas but “it’s going to take longer in a New York… a Boston”; Europe is waking up — commercial robotaxi is starting, with London pilots expected “before the end of the year.”

8. The super-app flywheel: cross-platform is the structural moat

Patrick notes that Uber Eats internationally has hit #1 more often than in the US; Dara says the “magic sauce is crossplatform”: 13% of Eats bookings now come from the mobility app — “we get a bunch of free customers.” Add Uber One: 50 million members growing 50% year-on-year, pitched as Netflix logic: “for the same price, you get more content than anyone else.” He thinks the combination yields #1 positions and higher margins than single-line competitors, “proving it out in market after market.”
Membership economics, stated precisely: the ideal program has fixed costs and zero variable cost to serve (travel upgrades, cable, Netflix). Amazon Prime was the first to take on variable-cost membership and walked “through this valley of despair” while public markets misread the mounting losses — “I took a lot of inspiration from that example.” Uber One is the same trade: “the first year of membership is a year where we lose money on you, but we’re going to make money on you 2, 3, 4 years from now.” It’s “solidly profitable now.”
On what supply excellence takes, an honest self-grade: “honestly, we’re okay at it. I think we can get a lot better.” His fix is embodiment — post-COVID he bought an e-bike and delivered food in San Francisco, and drove riders in his Tesla. The asymmetry he learned: a consumer touches the app 30 seconds; a driver has it open 6-10 hours, so “a P95 bug happens to consumers once a month maybe… a P95 bug is happening every single week for a driver.” About 50% of orders are batched — “there’s a reason why things go wrong in the real world.”

9. Hotels and the stretch from on-demand to planned

The hotels logic is data-led: travelers are heavy users — 1.5 billion trips outside home cities last year, ~15% of trips to/from airports — and each added service compounds retention (“more content meant more retention,” proven first with trains in the UK and Spain). The Expedia deal (his old company, full circle) gives “the vast majority of the economics back to Uber One members”: 10% off hotels, 20% off 10,000 hotels.
The end state isn’t booking — it’s in-market magic: pre-booked airport Ubers pulled from your email and flight data, the hotel alerted as you approach, “maybe you can use your Uber app as a key.”
His own premortem on it — kept honest: the brand may not stretch from on-demand to planned. The encouraging precedent is Reserve, rebuilt from on-demand dispatch to true pre-commitment with drivers at 99%+ reliability, now “over $5 billion run rate” from nothing five-six years ago. But “is reserving your ride just fundamentally different from thinking about your vacation 2 or 3 months from now?… I think we can. It’s not a slam dunk.”
A change of mind on marketing: he used to think marketing’s only job was traffic and product’s job was surfacing — “my marketing team told me I was an idiot and I loved it.” Patrick’s framing is that coffee-on-Reserve “isn’t about the ride” but “about the lifetime value… the introduction of a new service”; Dara adds that it should “feel delightful” rather than like an upsell.

10. Diller’s ground truth, troublemakers as mutations, and Amazon-vs-Apple

The Barry Diller lesson: get truth from the source material. As a young Allen Company analyst on the Paramount LBO during the hostile tender, Diller bypassed the MD and VP — “Who built the model? I’m going to talk to that guy, because if I’m going to raise billions of dollars in debt, I want to know I’m good for it.” The generalization: “it’s the filtering that gets the edge out of the story… it’s often the edge that gives you an edge. Everyone is going to have the same reaction to the average.” The failure mode of big-company CEOs: “you get a very thin layer of what’s going on… everything has been processed for you.”
His own operating system, if you shadowed him: brutal transparency plus engineered randomness — “I look for the troublemakers in the company… every company is like an organism… organisms evolve by mutating. Companies that don’t mutate… those are the companies that die.” And urgency: with AI, “the rate of change has accelerated by 5x. It’s the companies that are comfortable with that change that are going to adapt.”
Capital allocation, answering Patrick’s “are you Amazon or are you Apple?”: “somewhere in between.” It’s “more of an art versus a science” — priority one is organic investment (Eats grew from under $1B gross bookings when he joined to over $100B), then AV market development (“making commits to tens of thousands of AV vehicles” — financialized, per the likely Santander deal), with “plenty of cash left over for buybacks. But I prioritize growth. I prioritize innovation over buybacks.”
Two closing threads worth keeping: from Herbert Allen — “he makes bets on people, not companies… great people always stay great,” and Dara’s own best bets came from betting on Diller. And on being wrong, via Diller again: when Barry loses an argument “he’s delighted. He lights up, because he just learned… if I’m not making mistakes, it’s just not very interesting.”

The FDE Role OpenAI and Anthropic Are Betting On, with Rolling AI Bill Ackman: Here's What the Market is MISSING