Dara Khosrowshahi on Uber, AI, and the Autonomous Vehicle Future

Uber’s Dara Khosrowshahi argues that a company living in the physical world plays AI differently from a purely digital one: own the demand and the marketplace, partner for the self-driving stack, and treat autonomous vehicles as another trillion-dollar market to aggregate.

Key ideas

1. Uber is a physical-world company, and that changes the AI question. Its experiences are digital-first but fulfilled in the messy, probabilistic real world — traffic, cancellations, late food — so Uber has run machine-learning tools far longer than most. The new frontier is physical AI: autonomous vehicles and drones reshaping how the service is delivered, not just how it is booked.
2. Own the demand; partner for the driver. Uber runs over 30 AV partnerships (Waymo, Nuro, Lucid, Nvidia, Wayve, Pony.ai) and lets those firms build the digital driver while Uber supplies depots, charging, financing, insurance, and instant demand. AVs on Uber’s network run 30%+ busier than off it — the utilisation edge that justifies the expensive hardware.
3. Supply is the flywheel, not demand. The lesson Khosrowshahi carried from Expedia is inverted at Uber: recruit drivers, restaurants, couriers, and merchants first — in the next 50 and next 200 cities, not just the top 10 — and ‘the demand will take care of itself.’
4. Rebuild processes from first principles, but meter the cost of intelligence. Uber blew through a year’s AI budget in a quarter; Khosrowshahi now pairs ‘go-go-go’ exploration on frontier models with efficiency and open-source models at scale, and meters headcount as engineer throughput rises.
5. Leadership is engineered through constant challenge. Break chaos into its component parts (vector mathematics); seek out the ‘troublemakers’ as the mutations that keep the organism alive; go to the source for unfiltered ground truth; and take delight in being wrong, because that is where learning lives.

Summary

The physical-world thesis

Khosrowshahi frames Uber against the purely digital technology company. Its interactions begin in the app — a determinate structure — but resolve in the real world, where outcomes are probabilistic and things go wrong. That has made Uber a machine-learning-native business for far longer than most. Two AI waves now converge on it: larger models that know more about the user (feed and search models roughly 10,000 times bigger than the old ones, guessing the destination three-quarters of the time in a single tap), and physical AI — autonomous vehicles and drones — that changes the delivery form factor itself. He calls the pace of change the fastest he has seen.

Owning demand in the autonomous transition

The build-partner-buy question resolves into coexistence, a pattern Khosrowshahi knows from travel: online travel agents compete with Marriott and Delta yet drive incremental utilisation for them, because a hotel at 90% occupancy beats one at 70% every time. Waymo, Nuro, and Wayve will build their own channels and still want Uber’s demand. Uber’s role is the go-to-market layer around the driver — securing depots, charging, fleet financing (a billion-dollar Santander line for EV and AV fleets), autonomous insurance, and the data collected on today’s streets. The premortem that worries him is not competitive but social: AI and AVs are powerful yet unpopular, and Uber must move at the pace society will tolerate or face backlash. The controllable risk is access to supply — hence a partnership with nearly every AV provider.

Supply as the operating flywheel

At Expedia, Uber’s host built hotel and flight inventory in response to demand. Uber runs upside down: recruit the drivers, merchants, groceries, and couriers first, and demand follows. The largest opportunity is geographic breadth — the sparse suburbs and the next 200 cities, not the top 10. Excellence at aggregating supply, he concedes, is still a work in progress; it comes from putting employees in the suppliers’ shoes. Khosrowshahi bought an e-bike to deliver food and drove his Tesla in San Francisco to feel the courier’s day — a driver keeps the app open 6-to-10 hours, so a P95 bug that hits a rider monthly hits a driver weekly. ‘Building with heart’ is the stated value.

AI internally: adoption, cost, and efficiency

Uber’s AI adoption is bottoms-up and uneven — developers in India suddenly committing 10x the code via autonomous agents, uptake showing up in ‘random corners’ from legal to marketing. Khosrowshahi’s own push is to rebuild systems from first principles rather than shave 20-30% off an existing process. But intelligence is expensive: Uber consumed its annual AI budget in one quarter, forcing metered headcount and a two-speed model strategy — expensive frontier models to explore new interactions, cheaper or open-source models to scale them. On a business with $10bn+ of free cash flow spread over 10bn+ trips, efficiency funds lower prices and higher earner pay.

Marketplace design and leadership

Around the core marketplace sit the membership and planning layers. Uber One (50 million members, growing 50% year-on-year) follows the Amazon Prime playbook through its ‘valley of despair’ — the first year loses money on a member who becomes durably profitable later. Cross-platform selling is the structural edge: 13% of Eats bookings come free from the mobility app. Uber Reserve stretched the brand from on-demand to planned (a $5bn+ run-rate that did not exist five years ago), opening the door to hotels and travel. On leadership, Khosrowshahi credits Barry Diller for the discipline of going to the source for unfiltered ground truth, and Herbert Allen for betting on people over companies. He hunts for the ‘troublemakers’ — the mutations that keep a company from dying — and treats being wrong as the point: the delight is in learning.

Speakers

• Dara Khosrowshahi — CEO of Uber since 2017; previously CEO of Expedia for over a decade; steered Uber from post-Travis chaos to profitability and through the AV transition.
• Patrick O'Shaughnessy — host of Invest Like the Best; CEO of Positive Sum and chairman of O’Shaughnessy Asset Management.

See also

• Hierarchy of Marketplaces — the marketplace-quality frame; Uber’s supply-first flywheel and utilisation edge are a live case of demand-side aggregation.
• Latent Demand — Khosrowshahi’s claim that securing supply surfaces demand that was there all along, and that cheaper transport uncovers a market multiples larger than taxis.
• Brian Chesky on Founder Mode, the CEO Job, and Redesigning Airbnb for AI — a fellow marketplace CEO on the same show, on redesigning a physical-world platform for the AI era.
• Patrick O'Shaughnessy — the host.