By Qingcheng Finance and Qinchun
Edited by Liuzi
On August 18, Pony.ai released its 2026 first-half performance report: total revenue reached $70.47 million, nearly doubling compared with the same period last year; revenue from the Robotaxi business surged to $20.60 million, up 534%; the net loss rate fell sharply from 255.8% to 140.3%, suggesting that economies of scale may be starting to take effect.
But on closer inspection, the situation isn’t that simple. In the first half, its absolute net loss was $98.86 million, up 9.1% year over year; in Q2, Pony AI Inc.’s attributable net loss was $59.80 million, higher than $53.10 million in the same period last year. Revenue is rising, but losses are growing even more— the old problem of “growing revenue without growing profit” still persists.
More worth examining, however, is that its gross margin is below 17%, leaving a noticeable gap versus leading peers. Although Guangzhou and Shenzhen have claimed that per-vehicle Robotaxi operations have turned profitable, profits at the financial statement level still appear to be far away.
For Pony AI, are these current performances the costs that must be paid during an expansion phase, or is it the business model itself that needs to raise questions? How much time will the market still give it to prove itself? These questions are worth breaking down.
01 “Three cracks hide underlying concerns”
Judging purely by revenue growth, Pony AI’s half-year report is indeed eye-catching.
In the first half, its total revenue was $70.47 million, up 98.9% year over year; second-quarter revenue alone reached $36.2 million, up 68.8% year over year. The net loss margin dropped from 255.8% in the same period last year to 140.3%, narrowing by more than 115 percentage points. These numbers easily make you think of a technology company in a period of rapid growth.
But if you move your focus away from revenue growth rates and look deeper into the income statement, some signals worth paying attention to are beginning to emerge.
The most obvious aspect is still losses. In the first half, Pony AI’s net loss in absolute terms was $98.86 million, up 9.1% from $90.64 million in the same period of 2025. The second quarter is even more nuanced: net loss calculated under U.S. Generally Accepted Accounting Principles (GAAP) was $45.4 million, narrowing 14.9% year over year.
This looks like a positive signal, but when broken down, you find that the net loss attributable to Pony AI’s shareholders (Pony AI Inc.’s net loss) is $59.8 million—about $14.5 million more than the two figures suggest. Where does this gap come from? Mainly from the “co-built fleet” model: Pony AI and partners such as Qichens Tech, Uber, and Verne jointly fund and hold the vehicles, while Pony AI only provides the “AI driver” technology and charges technology licensing fees and revenue share.
Under this structure, some of the losses generated by fleet operations are borne by the partners and therefore are not included in the losses attributable to Pony AI’s shareholders. But the revenues generated by the fleet are consolidated into the company’s total revenue. Objectively, this creates an effect of “diverting” losses.
Therefore, as the proportion of co-built fleets continues to rise, this structural difference needs to be re-understood. After all, if you exclude share-based payment compensation, fair value changes of trading financial assets, and a long-term investment impairment of $25 million, Pony AI’s net loss in the second quarter is about $44.7 million, basically flat compared with the same period last year. In other words, the narrowing loss at the GAAP level may not show an obvious improvement once one-off factors are removed.
At the same time, the gross margin data is even more worth scrutinizing. In the first half of 2026, Pony AI’s gross margin was only 16.9%, up slightly by 0.6 percentage points from 16.3% in the same period of 2025. For comparison, WeRide—which released its earnings report a week ago—saw its gross margin jump from 30.6% in the first half to 36.6%. The gross-margin gap between the two leading L4 companies has widened from around 14 percentage points a year ago to nearly 20 percentage points.
This means that with the same scale of road testing and fleet deployment across multiple locations, Pony AI can cover operating and R&D costs with less than 17 yuan out of every 100 yuan of revenue, while WeRide’s figure is close to 37 yuan. And this gap is not a natural result of scale effects. Even as the revenue scale doubled, the gross margin barely moved. This suggests the root issue is not in the “quantity,” but in the “quality.”
In addition, the diversification among business lines is also worth noting. In the second quarter, Pony AI’s Robotaxi revenue surged 691.2% year over year. Passenger fare revenue jumped 849.3%, becoming an undeniable growth engine. Meanwhile, revenue from autonomous driving trucks was $13.3 million, up 40.0% year over year. While the growth rate is steady, it is far from the same scale as Robotaxi’s explosive growth.
It is also worth noting that the intelligent solutions business shows a different trend. In the second quarter, revenue from this segment was $10.8 million, up only 3.9% year over year, a slowdown compared with previous quarters. As the “second engine” that used to support growth, it now faces the reality that its growth momentum is weakening.
This also raises a question worth thinking about: as the company’s revenue mix gradually shifts toward the Robotaxi single business, will its overall risk resilience face new challenges? After all, changes in the pace of policy adjustments, fluctuations in regional markets, and even safety incidents in extreme cases could all have a more direct impact on an income model with a high concentration of revenue from a single business.
02 “Where is the root cause?”
Of course, if you place these “cracks” within Pony AI’s strategic framework, you may find they are not random operational mistakes. They likely relate to a few key choices. One core one is the trade-off of the “asset-light” model.
Pony AI co-founder and CFO Wang Haojun has stated clearly in public: “Under the co-built model, we will definitely not do vehicle assets; we will definitely do AI drivers. That is the core value.” He breaks down the Robotaxi value chain into four parts: AI driver technology licensing fees, vehicle asset depreciation costs, the cut taken by customer acquisition platforms, and fleet operations and management.
Pony AI’s strategy is to take only the “technology licensing fees plus revenue share,” while leaving heavy-asset segments like vehicle procurement and daily operations and maintenance to partners.
The appeal of this model is obvious. In 2026, many newly added vehicles at Pony AI are deployed using the co-built model. Since it does not need to pay upfront purchase costs of several hundred thousand yuan for each vehicle, capital expenditure pressure is greatly diluted, and cash reserves can be preserved.
But on the other side of the coin, this model also, to a certain extent, gives up profits from vehicle assets and operations. In the co-built fleet, Pony AI only takes a portion of the cash flow; a significant share of per-vehicle revenue flows to the partners. This revenue structure also, to some degree, locks in an upper limit on gross margin.
The deeper issue may not be so much the above, but rather it is related to the choice of the pure L4 route.
Currently, Pony AI’s three major businesses—Robotaxi, Robotruck, and intelligent solutions—are all focused on L4 autonomous driving. For comparison, WeRide follows a dual-wheel route of “L4 building a technical barrier” and “L2++ releasing scale dividends.”
Latest financial report data show that in the second quarter, WeRide’s WRD 3.0 shipments for L2++/L3 solutions were about 30,000 units, and it has already received production-destined appointments from more than 30 vehicle models. These driver-assistance businesses have high gross margins and stable cash flows, which can provide some support for sustained L4 investment.
Objectively, Pony AI is not without L2++ technical capability. Its L4-level autonomous driving system is downward compatible with L2++ functions, so there is no technical barrier that is difficult to overcome. The reason for not doing so is more a strategic choice.
In the narrative logic of the capital markets, the “pure L4” label implies a more thorough futurism and also means a larger valuation upside in terms of imagination. But as the industry moves from the technology validation stage into large-scale deployment, the impact of lacking middle-to-short-term “cash-generation” businesses is more likely to be noticed by the market.
By contrast, peers taking the “L4 and L2++ in parallel” route provide relatively stable cash flows through their driver-assistance businesses, which can, to some extent, feed back into L4 R&D and expansion. The difference in how the two models cycle capital is becoming clearer as competition in the industry deepens.
In addition, another reality that cannot be ignored is that technological advantages have not yet been effectively converted into a commercial moat. In terms of technology, Pony AI is undoubtedly a top player in domestic L4 autonomous driving. In April 2026, it released PonyWorld world model 2.0, giving AI the ability to self-diagnose and evolve directionally.
However, the decisive factor in the current Robotaxi battle is shifting from whose technology is better to whose operating costs are lower and whose order density is higher.
Technology determines whether the vehicles can hit the road, but operational efficiency determines whether the vehicles can become profitable. Building operational capabilities—including dispatching and staffing for ground operations, establishing maintenance and upkeep systems, and local adaptation in city expansion—requires continuous investments of capital and manpower. For a pure L4 model, this portion of investment depends entirely on external financing or internal cash reserves, making its sustainability a long-term focus of market attention.
03 “How far is the break-even turning point?”
In Pony AI’s first-half results, the highlight of “per-vehicle UE turning positive” is undoubtedly exciting.
In November 2025, the seventh-generation Robotaxi achieved monthly per-vehicle operational profitability in Guangzhou; in February 2026, Shenzhen followed suit. As a result, Pony AI became the first company in China to complete city-level Robotaxi profitability validation in frontline cities.
Turning per-vehicle profit into reality is indeed a milestone worth marking. It means that in a city, once the size of the fleet reaches a certain level, the daily operating revenue per vehicle is already enough to cover its directly related variable costs, including vehicle maintenance, charging or energy replenishment, insurance, roadside support, and so on. This shows that Robotaxi’s per-vehicle economic model can hold true in specific urban conditions.
But there is still a huge gap between per-vehicle profitability and overall company profitability. Per-vehicle UE only covers variable costs directly related to vehicle operations, while the company-level R&D spending, management expenses, and market expansion costs are massive fixed expenditures that require a sufficiently large fleet base to spread and absorb. Co-founder and CFO Wang Haojun reiterated the company’s earlier judgment at the earnings communication meeting: only after accumulating deployment of 40,000 to 50,000 Robotaxis does it have a chance to turn free cash flow positive.
As of the end of the first half of 2026, Pony AI’s global Robotaxi fleet size was 1,975 vehicles. From 1,975 vehicles to 40,000–50,000 vehicles is a distance of more than 20x. The company’s goal is to expand the fleet size to more than 3,500 vehicles by the end of 2026. Even under the most optimistic expansion plan, accelerating overseas and domestic markets at the same time to reach the 40,000–50,000-vehicle threshold would likely still take several years.
And the bigger challenge lies in the average fare per trip. As far as we know, Waymo’s average fare per trip in the U.S. market is about $15 to $20, several times higher than Pony AI’s domestic market. With the same order density, the monetization ability differs dramatically.
Against this backdrop, overseas markets are understandably expected to play a major role. As it turns out, Pony AI has locked in an overseas deployment plan for more than 4,000 Robotaxis, including over 2,000 vehicles deployed in five European cities in cooperation with Uber. In theory, overseas markets with higher fares per trip can significantly improve the quality of revenue per vehicle.
At a financial results communication meeting, Wang Haojun said that under the co-built model, the vehicle assets are borne by the partner, which can significantly improve capital efficiency, but there is a timing lag because it is affected by the vehicle production and delivery schedule. However, he also admitted that although per-vehicle revenue is higher in overseas markets like Europe, the company will not lower its scale threshold of 40,000 to 50,000 vehicles. Overseas scaled deployment still takes time, and the company has not locked in a specific year of profitability.
So, how big is the time window the market leaves for Pony AI to prove itself? This may still come down to a more fundamental metric: how long the cash on the books can still be “burned.”
As of the end of the first half, Pony AI’s total cash and cash equivalents, restricted cash, short-term investments, and long-term restricted investments amounted to about $1.39 billion (about RMB 9.435 billion). Just in terms of the numbers alone, this funding reserve still places it in the top tier within the autonomous driving industry.
But the pace of “burning cash” is also accelerating. In the first half of 2026, the net cash flow from operating activities was a net outflow of $118 million, widening by about 48% compared with the net outflow of $79.57 million in the same period of 2025. Roughly calculated, this means the company “burned” about $59 million per quarter on average in the first half of this year.
At this static burn rate, Pony AI’s $1.39 billion cash reserve could theoretically support about 23–24 quarters (nearly 6 years). But it’s important to note that this estimate is based on the assumption that the cash-burn pace will no longer increase significantly.
In other words, abundant cash reserves give Pony AI a relatively comfortable space for trial and error, but it is not unlimited. A company can continue to raise funds from capital markets, but ultimately what determines its value is whether it can generate positive cash flow through its own business. The market will likely need to remain patient for the arrival of that day.