Tony Xu on DoorDash DOT robot, 35-country expansion, and the physical delivery wars

Summary

DoorDash CEO Tony Xu used the company's Dash Forward event to announce six products simultaneously, a pace he acknowledges could have been staggered. The centerpiece is DoorDash DOT, an autonomous delivery vehicle seven years in development that travels roads, bike lanes, and sidewalks, tops out at 20 mph, and is about a tenth the size of a car. DOT has already logged hundreds of thousands of real delivery miles in Phoenix, covering roughly 1.5 million people, and Xu says it can handle up to half of all deliveries in markets where regulations permit — specifically those under five miles.

The product stack around DOT is as significant as the vehicle itself. Google Maps isn't precise enough for DoorDash's needs, so the company built its own mapping system over five years, tuning parameters like gate codes, parking spaces, and building entry points. A new autonomous delivery platform sits above all of this, routing orders across DOT, sidewalk robots, drones, and human Dashers based on cost and quality. Smart scales — hardware installed at merchant locations — are already producing a 30% improvement in order accuracy by weighing items before they leave the store.

The hardware-software challenge

Xu pushes back on the Silicon Valley instinct to frame autonomous delivery as a software problem. Hardware manufacturing doesn't scale the way software does, and each 10x increase in fleet size introduces new manufacturing dependencies. But traditional automakers, he argues, underestimate the software side — the L4 autonomy stack, the LLM fine-tuned for physical-world navigation, the mileage data required to make it work. The fourth layer that rarely gets discussed is operations: loading, unloading, maintenance, rescue, and teleoperation. With a robo-taxi, the passenger handles the last step. With DOT, someone has to.

Regulatory sequencing adds another variable. DoorDash has to make multi-year commitments to manufacturing partners before knowing the order in which city permits will come through. The merchant waitlist is long; the deployment timeline is not fully in DoorDash's control.

Scale beyond food delivery

Xu frames DoorDash as a business most investors still misread. The company is now live in 35 countries, gaining share in every geography, and is on track to become the largest third-party marketplace for non-restaurant goods in addition to restaurant delivery. Its B2B software runs inside the McDonald's and Starbucks apps, handling mobile ordering and delivery. The ads business reached $1 billion in annualized revenue faster than any comparable business. Dashmart, originally a network of dark warehouses, is now being offered as a fulfillment service — letting any retailer offer one-hour delivery without building their own infrastructure.

The compounding logic matters here. DoorDash's density and order volume create minimum efficient scale advantages that are mathematically derivable, Xu says. As scale grows, unit economics improve, and those gains get reinvested into the next product layer. He describes it as always building the core and the new simultaneously, with entirely different management playbooks for each.

AI inside the operation

The most concrete AI application Xu describes is data labeling. DoorDash needs to label hundreds of millions of items across cities — parking availability, specific SKUs, apple varieties by store — data that doesn't exist on any search engine or in any training corpus. Doing this manually would have been prohibitively slow. LLMs now handle it at what Xu estimates is a thousandx the throughput of a human operation. On coding, he offers a more measured take: productivity gains are real, but most engineers don't spend most of their time coding, so the weighted-average productivity improvement at scale is less dramatic than headline figures suggest. The gains are clearest at smaller companies where most output is net-new.

The founding mission

Xu traces DoorDash's origin not to a consumer insight but to watching small business owners — his mother among them — operate on 17 days of cash. The long-term vision is a physical commerce infrastructure that lets the maximum number of local businesses survive, rather than a world where one or two large retailers dominate last-mile logistics. The average Dasher works three to four hours a week; 88% already have full-time jobs. Xu frames the platform as a supplemental income layer, not a primary employment substitute.

The bet, taken together, is that whoever builds the most integrated physical delivery stack — mapping, dispatch, autonomous vehicles, merchant systems, warehouse fulfillment — controls the infrastructure layer beneath both digital commerce and whatever agentic AI becomes. DOT is the most visible piece, but the seven years of underlying plumbing is harder to replicate.