Pony AI Becomes Uber's Largest European Robotaxi Partner as Revenue Surges 69% and AI Automation Slashes City-Launch Costs
Q2 2026 earnings call, August 18, 2026
Pony AI used its second-quarter call to lay out what management called a "dual engine" acceleration in both China and international markets, anchored by an expanded Uber partnership that now makes the Chinese autonomous driving company Uber's largest robotaxi partner in Europe. The quarter also revealed a new self-evolving AI system, PonyWorld 2.0, that the company says compresses the engineering effort required to enter a new city from dozens of workers to just a handful, a claim with direct implications for the capital intensity of the entire robotaxi industry.
Uber Deal Expands to 2,000 Vehicles, Total Commitments Hit 4,000
The headline commercial news was the scale of Pony's alliance with Uber, which now includes commitments for more than 2,000 robotaxis across five European cities, part of a broader tally exceeding 4,000 vehicle commitments with Uber and other overseas partners. CEO James Peng explained why Uber chose Pony over other autonomous driving partners it works with globally: "Uber always looks for autonomous driving partners whose technology is reliable at scale, and also whose cost structure brings the attractive economics. That's exactly the two reasons that we can offer on the table." Peng noted that when the partnership began in early 2025, there were doubts about whether Pony's technology could handle Europe's mixed-infrastructure cities, but the launch of Europe's first commercial robotaxi service in Zagreb, Croatia has since validated the model. "With these contracts, we become Uber's largest autonomous driving partner in Europe," Peng said, adding that further fleet expansion is likely as the economics continue to validate at scale.
CFO Leo Wang detailed the mechanics of this joint deployment model, which is structured as a three-party arrangement in most markets: Pony supplies the Gen-7 vehicle and virtual driver software, a mobility platform like Uber or Bolt supplies demand, and a local operator such as Verne in Croatia or ComfortDelGro in Singapore handles fleet management. The model is asset-light for Pony since partners fund the vehicles, and it generates both upfront vehicle delivery revenue and, over time, higher-margin recurring revenue-sharing income. Wang was explicit that the current 4,000 vehicle commitments should be viewed as "a multiyear growth catalyst for 2026 and beyond," rather than a one-time contract.
PonyWorld 2.0 Turns City Expansion Into a Largely Automated Process
CTO Tiancheng Lou offered the most technically substantive disclosure of the call, describing how the newly upgraded PonyWorld 2.0 system automates what was previously a manual, engineer-intensive process of adapting the autonomous driving stack to local traffic behavior. The system works by refining the statistical precision of the company's world model, which needs to capture not just what traffic participants might do but the exact probability with which they do it, since jaywalking or unsignaled lane changes occur at different rates in different cities. Lou explained that the old workflow required "dozens of engineers doing manual work to review local driving issues, analyze root cause of these issues, upgrade the World Model and retrain the onboard models," but PonyWorld 2.0 now handles this automatically, with humans mostly needed for verification. As an example, he cited Zagreb, where local drivers rarely slow down even at blind spots when they have right of way; the system caught this pattern and retrained the virtual driver with minimal human involvement. Lou argued this scalability advantage is durable even against open-source competition, stating that a generic open-source world model is "basically just a 3D video generator" lacking the statistical precision needed to actually train an onboard autonomous driving system.
Financial Results Show Operating Leverage Beginning to Show Through
Total revenue reached $36.2 million, up 69% year-over-year, with robotaxi revenue the standout line at $12.1 million, up 691% and accelerating from 395% growth in the first quarter. Fare-charging revenue specifically grew 849%. Robotruck revenue grew 40% to $13.3 million, while the Intelligent Solutions segment, which includes domain controller sales, grew a more modest 4% in the quarter to $10.8 million, a deceleration the company attributed to delivery timing fluctuations. Gross margin improved to 17.5% from 16.1% a year earlier. More notably, non-GAAP operating expenses rose just 9.6% while revenue grew nearly 69%, narrowing the operating loss margin from negative 285.6% to negative 181.5% year-over-year, an improvement of over 100 percentage points. Net loss narrowed 14.9% to $45.4 million. Cash and equivalents stood at $1.39 billion as of June 30, down modestly from $1.44 billion at the end of the first quarter, with capital expenditures of $32.2 million in the quarter tied to fleet expansion and data center investment.
Operational Efficiency Claims: Three Staff per 100 Robotaxis
Lou provided a specific operating metric that underpins the company's unit economics argument: Pony now needs only three human support staff for every 100 robotaxis, covering charging, cleaning, and maintenance, compared with the one-to-one ratio required for traditional taxi fleets. He noted that Pony's vehicles autonomously navigate to shared public parking lots, locate open chargers, and self-park without human intervention, which he said is a key reason partners are increasingly willing to commit fleets under the joint deployment model. This efficiency claim is central to the company's broader argument that its cost structure, not just its safety record, is what is winning partner commitments internationally.
New Light Truck Vertical Targets an 8 Million Vehicle Market
Management also detailed a new L4 light truck initiative, unveiled four months ago at the Beijing Auto Show, that reuses robotaxi technology and hardware to target urban delivery logistics, a market Peng sized at over 8 million active light trucks in China alone. The vehicle, developed jointly with battery maker CATL, is described as the world's first automotive-grade, fully redundant light truck built for L4 autonomous driving, offering three to four times the cargo capacity and twice the speed of existing low-speed robovans. Pony has already signed partnerships with SF Express and China Post Technology, giving the new segment a logistics customer base before wide commercial rollout, though the business remains nascent relative to the core robotaxi and robotruck lines.
Management Pushes Back on "Demo Versus Scale" Framing
Asked to respond to a comment attributed to Waymo management that "the demo is only 1% of the work," Lou offered a candid explanation of why scaling autonomous driving is fundamentally a probability problem rather than an engineering demo. He noted that a fleet of 100 vehicles driving 300 kilometers a day in one city would need an accident rate low enough to avoid roughly 10 incidents daily at unacceptable failure rates, meaning safety improvements require "multiple 10x jumps in performance," each harder than the last, since fixing one failure mode, such as hard braking, can introduce new risks like rear-end collisions. Lou argued Pony has already cleared this bar through its Tier 1 China deployments and Zagreb operations, positioning the company's current phase as pure scaling rather than continued technology validation, a distinction management clearly wants investors to draw versus less-proven competitors.