After a prolonged period of technical validation, leading companies in the autonomous driving industry are now presenting their commercial mass-production results to the market.
Financial reports released by Pony AI show that its revenue reached $90 million in 2025, a 20% year-over-year increase. During the same period, the company's loss narrowed to $76.8 million, a reduction of over 70% compared to the previous year.
The most closely watched aspect of these results is the progress of its autonomous ride-hailing service (Robotaxi). In 2025, Pony AI's Robotaxi business generated $16.6 million in revenue, a sharp increase of 128.6% year-over-year.
The primary driver of this growth comes from passenger fare contributions following the expansion of its vehicle fleet. In the fourth quarter of 2025, passenger fare revenue from Pony AI's Robotaxi service surged by more than 500% compared to the same period last year.
Specifically, Pony AI achieved an operational milestone by reaching positive unit economics per vehicle in Guangzhou and Shenzhen. On March 22, 2026, the seventh-generation autonomous taxi in Shenzhen set a new daily record for average net revenue per vehicle at 394 yuan, with an average of 25 orders per vehicle that day.
Calculations based on this daily revenue and order volume indicate an average fare per trip in Shenzhen of approximately 15.76 yuan. Considering the local pricing structure of a 10 yuan starting fare and 2.7 yuan per kilometer, this suggests that current operations are still confined to limited travel distances.
Regarding this, Pony AI's CFO, Wang Haojun, acknowledged that operations in Shenzhen currently consist mainly of short-distance trips. This is primarily limited by the current service area being concentrated in Shenzhen's Bao'an and Nanshan districts. However, it is expected that as more urban areas in Shenzhen and Guangzhou open up this year, the order mix will shift from predominantly short trips to a combination of short and long-distance trips.
As the total distance traveled increases, metrics like Mean Miles Between Interventions (MPI) are drawing attention. For instance, late last year, a Tesla Model 3 equipped with FSD v14 completed a 2,732-mile journey from Los Angeles on the US West Coast to South Carolina on the East Coast in 2 days and 20 hours. The trip was 100% reliant on FSD, covering complex scenarios including highways, city streets, night driving, and multiple Supercharger stops, all without any manual intervention, sparking significant discussion in the market.
Nvidia's robotics lead, Jim Fan, even exclaimed that "Tesla FSD v14 may have passed the 'Physical Turing Test'."
However, from Wang Haojun's perspective, MPI is not a suitable metric for Level 4 autonomy. "In reality, when reaching the stage of large-scale L4 operations, people no longer talk about MPI; the concept itself becomes irrelevant. Without a human driver involved, the issue of 'takeover' doesn't apply. The main focus for L4 operations is on the scale of deployment. A larger deployment scale inherently leads to a lower accident rate. Beyond that, the proportion of instances requiring remote assistance is what needs attention," Wang pointed out.
He further noted that observing the current landscape reveals that companies like Waymo have already moved away from emphasizing MPI. Conversely, many companies developing L2+ systems still reference MPI as they advance towards L4. A genuine assessment of the situation shows that the key to achieving large-scale L4 operations lies in operational scale and the remote assistance ratio.
Looking ahead, Pony AI has set a target to deploy over 3,000 autonomous taxis in more than 20 cities globally by the end of 2026.
Relying solely on heavy capital investment in a proprietary fleet for such a rapid expansion in capacity would clearly be a significant drain on cash flow. Pony AI's solution involves a dual strategy: expanding into new cities and establishing joint fleet partnerships with third parties.
For city expansion, Pony AI plans not only to increase density in existing first-tier cities in China but also to expand into new first-tier cities like Hangzhou and Changsha.
Under the "joint fleet model," Pony AI effectively transfers the asset-heavy vehicle purchase costs to downstream partners. Third parties, such as Ruqi Chuxing, fund the vehicle purchases and share in the operational revenue. Pony AI operates in the background, generating income by licensing its AI autonomous driving technology.
As this partnership model only began in the third quarter of last year, the current number of vehicles in operation remains small. Wang Haojun anticipates that contributions to revenue will increase in the second half of 2026 as more Robotaxis from these third-party collaborations are deployed.
However, the overall pace of expansion remains dependent on the speed of policy liberalization in each city. Currently, China's urban operational zones lack a large-scale mutual recognition mechanism. This means that every time a Robotaxi company enters a new city, it must go through a gradual process from road testing with safety drivers to final commercial deployment of fully driverless operations.
The pace of policy advancement overseas is similar. Recently, Waymo's co-CEO Tekedra Mawakana stated in an interview that in some cases, Waymo can complete the entire process from city mapping to paid rides in just a few months. In other instances, progress is much slower, particularly in cities or states lacking specific Robotaxi regulations.
Overall, the competitive phase for domestic Robotaxi companies still primarily focuses on deploying as many vehicles as possible to gain a first-mover advantage. As the industry leader, Waymo has already progressed to the stage of competing on order volume, aiming to achieve over 1 million paid Robotaxi trips per week in the US market by the end of 2026.
In this new phase of the Robotaxi race, technology is no longer the sole moat. The players who can first achieve a closed commercial loop with large-scale, sustainable order volume will be the ones who truly remain at the table.
Comments