Ask AI · What key challenges does Pony.ai face in scaling to profitability?

Author: Jia Lele, Editor: Zhao Yuan

After Pony.ai released its 2025 financial report, the U.S. stock market fell 14.66%. Although Chinese concept stocks generally came under pressure that day, the figure was still very striking. After the Hong Kong market opened, it also followed the same downward trajectory as the U.S. market, closing with a drop of nearly 14%.

The financial report shows that in 2025, Pony.ai’s total revenue was $90 million (about RMB 629 million), up 20% year over year. The company recorded a net loss attributable to shareholders of $134 million, narrowing 51.13% year over year.

Among that, in the fourth quarter of 2025, revenue was $29.13M, down 18% year over year, but the company turned a profit during the quarter. For an autonomous driving company still in the early stage of commercialization, this is an important signal.

In addition, on the operating metrics side, Pony.ai also said that Robotaxi revenue and passenger fares surged, and both Guangzhou and Shenzhen achieved a positive turnaround in per-vehicle profitability.

So why doesn’t the capital market buy the story?

I. Unstable growth structure—profitability depends on Moore Threads

In the $23.43 million of book net profit in the fourth quarter, the largest component was a $132 million gain from fair value changes of “trading financial assets.”

This is widely seen as resulting from Pony.ai’s investment in Moore Threads, whose stock price rose after it went public. As a shareholder, Pony.ai benefited.

On the one hand, this is a classic case of “paper wealth”—the stock wasn’t sold, so the money didn’t actually come in. On the other hand, this has little to do with the company’s core business.

If you strip out these investment gains and other non-cash items like equity incentives, the company’s adjusted net loss for the fourth quarter is $48.64 million, which is 18% higher than the $41.19 million in the same period last year. In other words, losses at the operating level have not narrowed.

Let’s get back to how the core business is performing.

Pony.ai’s business is divided into three parts: autonomous driving mobility services, autonomous driving truck services, and technology licensing and applications.

In the fourth quarter, Pony.ai’s total revenue fell 18% year over year. In particular, technology licensing and applications dropped from $20.10 million to $9.40 million—essentially cut in half.

The technology licensing and applications segment mainly refers to collaborating with leading semiconductor chip suppliers to develop L4 autonomous driving controllers, and providing autonomous driving solutions to customers—for example, tailoring sensor solutions for sensor suppliers. Downstream customers include OEMs, autonomous delivery vehicles, and autonomous driving vehicle fleets, among others.

Regarding the decline in this area, Pony.ai explained that it was due to the “timing of revenue recognition under project-based contracts.” In other words, large orders were signed in other quarters, so they didn’t show up as revenue this quarter.

But that inevitably raises concerns: will this kind of big ups and downs become the norm? Ultimately, if there are enough projects, revenue will be less affected by the project delivery schedule, which would make performance more stable.

For autonomous driving truck services, in the fourth quarter of 2025, revenue was flat with the same period in 2024, with a slight increase of 1.2%.

Carrying the growth banner is autonomous driving mobility services. Fourth-quarter revenue was $6.70 million, up 159.5% year over year. Of that, passenger fare revenue increased by more than 500%.

Pony.ai’s seventh-generation Robotaxi was投入 into commercial operations in concentrated fashion in November last year. That month, Guangzhou achieved a positive turnaround in per-vehicle operating profitability. Currently, Shenzhen’s Robotaxi has also achieved a positive turnaround in per-vehicle operating profitability. On March 22, the seventh-generation Robotaxi in Shenzhen recorded a new historical high for average daily net revenue per vehicle of 394 yuan, and the average number of orders per vehicle that day reached 25 orders.

In the words of Peng Jun, Chairman of Pony.ai’s board of directors and CEO, Pony.ai has “successfully validated the business model.”

As of March 26, 2026, Pony.ai’s Robotaxi fleet exceeded 1,400 vehicles. Looking ahead to 2026, Pony.ai’s goal is: Robotaxi fleet size exceeding 3,000 vehicles, and more than 20 cities rolled out both domestically and internationally.

From a full-year perspective, the growth rates of autonomous driving mobility, autonomous driving truck services, and technology licensing and applications were 128.5%, 0.6%, and 19.7%, respectively.

Of Pony.ai’s three legs, only one is running quickly. The other two—what they show so far—is weaker growth potential and less certainty.

In addition, Pony.ai’s cash flow data is also rather weak. In 2025, net cash outflow from operating activities was $165 million, expanding 48.9% year over year. Free cash flow was -$209 million, and the gap increased 71% year over year.

So although Pony.ai says it was an excellent year in 2025, investors still have difficulty not focusing on these not-so-strong numbers.

II. Burning cash for growth—how long can it continue?

Regarding stock performance, according to data from Futu Niuniu, Pony.ai’s actual revenue in the fourth quarter was higher than expectations. “Moore Threads’ stock price rise thickened Pony.ai’s net profit” was also within expectations. Yet after the financial report was released, the stock still dropped sharply.

(Source: Futu Niuniu)

Pony.ai and WeRide (Wen Yuan Zhixing) are both racing to become the first Robotaxi stock. With roughly the same timing and similar valuations ($4.2 billion and $4.5 billion) when they listed on Nasdaq, both experienced a post-listing drop (breaking issue price). Even the founders are “from the same school” (Baidu).

By the time Pony.ai listed back in Hong Kong, its market capitalization was more than twice that of WeRide.

After the disclosure of the 2025 financial reports, Pony.ai quickly slumped while WeRide rose quickly, and the gap in market capitalization began to converge rapidly.

When losses cannot be contained, the scale and growth rate are important reference points for valuation. In 2025, Pony.ai and WeRide had comparable overall revenue: RMB 629 million versus RMB 685 million, but WeRide’s growth rate was far higher than Pony.ai’s.

Focusing on Robotaxi, whether by scale or growth rate, WeRide is still ahead.

So the rapid convergence of the market-cap gap between the two may be the market’s recalibration of relative valuation.

As for when it can achieve true profitability—rather than turning UE positive after deducting multiple costs—Pony.ai is still at a stage where even a single quarter’s revenue cannot cover research and development expenditures. For example, using 2025 data, the two projects were $29.13M and $60.52M, respectively. Combined with the cash flow data mentioned above, Pony.ai has not yet formed a closed loop in its “cash-generating” ability at the operating level.

In interviews with the media, Peng Jun once said that when Robotaxi scale reaches 50k vehicles, the company expects to become profitable, with the timing of profitability estimated to be 2028–2029.

Therefore, the most important task for Robotaxi is to achieve scaling. Only with scale can Pony.ai bring down the cost of vehicle procurement and the production costs of components, and spread out R&D spending.

Against the backdrop of the Robotaxi industry accelerating commercialization due to technological maturation, falling costs, and policy loosening, the race is speeding up.

On the technology side, although leading companies have achieved L4 operations in partial areas, fully solving long-tail scenarios remains challenging.

If end-to-end large models or sensor-fusion solutions do not continue to make sustained breakthroughs when handling extreme weather and complex road conditions, it will lead to a delay in the “fully driverless” goal. Companies would need to maintain high R&D and operating expenses for the long term, and businesses whose revenue sources are relatively single will face the risk of a break in the capital chain.

Beyond technology, the battle for market share is also intensifying.

After achieving UE positive with an owned fleet, Pony.ai, through a “co-built fleet” approach, has exported proven operating capabilities to partners. It has opened vehicle ownership, operations, and other parts to partners, tapping into industry resources.

Currently, Pony.ai has reached cooperation on the co-built fleet model with partners including Toyota, Ruqi Travel, Aitibo, Verne, and others. In addition, Pony.ai also needs to begin its global commercialization journey for autonomous driving via a “capital-efficient” model—meaning Pony.ai provides a validated autonomous driving platform, while local strong partners are responsible for fleet assets, operations, and market access.

However, whether it can replicate scaling across different contexts still depends on multiple issues.

First is the escalation of the cost race.

In April 2025, Pony.ai released its seventh-generation autonomous driving system, claiming a 70% cost reduction. After that, the stock price in the U.S. market rose by more than 30% for two consecutive days—this is also the beginning of the market-cap divergence between Pony.ai and WeRide.

Baidu has already compressed the cost of its sixth-generation complete vehicles to the 200k-yuan level. For XPeng, which has newly entered the Robotaxi business, it can both make vehicles and conduct in-house smart driving R&D. This means XPeng can factor in driverless driving from the design source.

Pony.ai and WeRide are also continuing to reduce costs, but they have not provided specific data on complete-vehicle costs.

Second is the dual-sided nature of the business model.

For co-built fleets—this kind of asset-light model—the advantage is faster expansion.

Pony.ai’s per-vehicle profitability under the self-operated model is achieved through coordination across multiple areas, such as unified fleet dispatching and standardized ground maintenance processes.

But under the “co-built fleet” model, partners’ operating capabilities vary. It remains unknown whether the efficiency of the self-operated system can be replicated, and whether both sides can be profitable under the asset-light approach.

In addition, as the number of partner fleets increases, as partners control fleet asset and operational data, will Pony.ai’s technology iteration lose proactive control of the data closed loop? This requires Pony.ai to obtain more assurance regarding data ownership.

Only with data from fully driverless Robotaxi can the gap between the world model and the real world be continuously narrowed. In other words, the real operational data from Robotaxi fleets is an irreplaceable component of the technology moat.

Finding a balance between “fast” and “stable” is a challenge Pony.ai cannot avoid on the path to scaling.

Author’s statement: Personal opinion, for reference only