China's smart driving launches a brutal breakout battle

Author | Zhou Zhiyu Editor | Zhang Xiaoling China's intelligent driving industry is undergoing a fierce reshuffling. Recently, the once unicorn Haomo.AI has been reported to stop operations, a message confirmed by its employees. Almost simultaneously, Pony.ai made a high-profile announcement in Guangzhou that it had seized 40% of the third-party assisted driving market, while Zhiyu, backed by FAW Group’s 3.6 billion RMB investment, gained the endorsement of the “national team.” This contrast of hot and cold is a microcosm of the industry’s tumultuous shakeup. Carmakers are no longer blindly casting a wide net; BYD, Chery, and Geely are integrating in-house intelligent driving resources, or deeply binding with only a few top-tier players. The once-blooming supply chain is rapidly shrinking. In the second half ruled by end-to-end large models, intelligent driving is no longer just an ornamental configuration, but an expensive ticket to AGI. The market no longer trusts flashy PPTs and big talk; only those who hold the entrance to millions of mass-production data have the right to board the Noah's Ark heading for the future. As for those who can't cross the threshold of scale, their names can only be mercilessly swallowed in this torrent. **The Stalled Unicorn** Haomo.AI pressing the pause button might be the most cautionary episode for intelligent driving in 2025. Haomo.AI originated from Great Wall Motors’ Advanced Intelligent Driving System Division, split off in 2019 with Great Wall’s deputy chief engineer Zhang Kai as chairman and former Baidu Automotive’s chief architect Gu Weihao as CEO. This was a typical case of combining traditional automotive industry with internet thinking, and was well favored by capital in its early days. In 2021, it won two rounds of hundreds of millions RMB financing in one year, with Meituan and Hillhouse Capital jumping in. Back then, Haomo had an enviable dream beginning for any startup: not only did it have a steady stream of orders from Great Wall, but also attracted technical executives from Baidu. However, Haomo's most recent financing dates to the first half of 2024, with a Series B round exceeding 100 million RMB. But there was no sign of Great Wall in this round. Initially, Great Wall Motors was Haomo’s important pillar during its period of capital favor, and Haomo’s HPilot 1.0 system was first adopted in the WEY Mocha model. But luck didn’t last. One technical detail: at the hardware selection stage, chasing for cost-performance, Haomo bet on Qualcomm’s Snapdragon Ride chip instead of the industry mainstream NVIDIA Orin. In the era of “high-precision maps + rule algorithms,” the choice seemed smart—cheap and sufficient. However, when the industry rapidly shifted toward Transformer architecture, this became Haomo’s trap. According to industry insiders, because the platform was very difficult to adapt to the new architecture, Haomo engineers had to handwrite operators and hard-adapt code. While competitors ran high frame rates easily on NVIDIA, Haomo struggled with basic operational efficiency. Even worse was strategic arrogance and sluggishness. In July 2023, the market had already shifted from “light maps” to “end-to-end large models.” But at this time, Haomo executives publicly expressed reservations about large model integration, believing it was too early. Yet, just a month after those remarks, Tesla launched the end-to-end FSD V12, rewriting industry rules; ten months later, XPeng mass-produced end-to-end large model cars. This strategic misjudgment led to Haomo’s massive technological drop-off. Meanwhile, Haomo was still pushing its city NOA (No Map Operator) solution. In an internal letter last November, Zhang Kai and Gu Weihao set the next year’s focus as developing the best value urban mapless NOA solution. Great Wall Motors’ patience was eventually exhausted by “PPT deliveries.” Executives bluntly stated at internal meetings that they are only responsible for the competitiveness of their cars, and will use whichever technology is best. Though they admitted there was pressure and difficulty in making the choice, this approach would benefit market health. As Haomo failed to deliver, Great Wall brought in Pony.ai as its intelligent driving supplier, and in November 2024, as Haomo celebrated its 4th anniversary, Great Wall made an exclusive $100 million investment in Pony.ai. Haomo’s expansion to other clients was also lackluster, as it announced partnerships with Beijing Hyundai and others in 2023, but few models have adopted their solutions. Thus, Haomo’s halt wasn’t a coincidence. **Scramble for Entry Tickets** In 2025’s intelligent driving battlefield, competition has escalated. Previously, everyone compared who had the cooler demo; now, it’s about who can cross the “one million cars” physical threshold. “Horizon chips are in one out of every three smart cars.” Lü Peng, VP of Horizon Robotics, boasted at a recent event. This newly Hong Kong-listed unicorn is trying to build a “Wintel”-like alliance in the automotive PC era. Why fight for one million cars? Pony.ai’s CEO Zhou Guang revealed: “Assisted driving in pure highway scenes can hardly form an effective data loop—business model problems arise.” In other words, without millions of mass-produced cars generating data on the roads day and night, any sophisticated code will dry up. In the end-to-end era, data isn’t just a record, but fuel for algorithmic evolution. Only when fleet scale reaches millions and daily activity is high, can it cover enough corner cases—long-tail scenarios—and enable “intelligent emergence" in models. Players below this level, regardless of technology, will have data that is single and ordinary, lacking value. This explains why Pony.ai won out among suppliers, securing orders for core models like WEY Blue Mountain and Tank 500. Zhou Guang shared a detail: to win a top client, his team waited outside the customer’s hotel at four a.m., just for a test drive chance. This “wolfish” hunger paid off—as WEY Highlander MPV with Pony’s solution saw monthly sales jump from 300 at the beginning of the year to nearly 10,000 in October. “Clients are willing to entrust their core blockbuster models to us.” Zhou Guang’s remark shows the new role of intelligent driving suppliers: they are no longer behind-the-scenes component providers, but strategic partners determining model success. A similar story happened when FAW Group joined hands with Zhiyu (formerly DJI Automotive). FAW was willing to bet 3.6 billion RMB on Zhiyu, not for simple financial investment. As the “eldest son of the republic," it needs a super Tier 1 supplier that can connect to vast ecosystems and handle massive data. Zhiyu’s promise to remain independently operated was the core value of the deal. If Zhiyu became FAW’s “Haomo,” it would lose access to data from other brands and its platform value would vanish. FAW values Zhiyu’s cost-performance and openness—this is the lifeline for all carmakers in brutal price wars. Currently, the power structure of the intelligent driving industry is being reshaped. Those who hold “million-level” data entrances are gaining absolute pricing control over the industry chain. Horizon’s chip + algorithm ecosystem, Pony.ai’s hit models, Zhiyu’s “national team” endorsement—all are about seizing the high ground of data. In this scale war, there is no middle ground. Either grow big and become a platform, or be small and gradually wither. For those still proud of having a few thousand unit shipments, the countdown to leaving the table has begun. **The Final Game** It seems the industry has reached a consensus: intelligent driving is no longer just a feature, but a ticket to the next era—AGI (Artificial General Intelligence). According to Zhou Guang and Lü Peng, what they’re pursuing far exceeds helping carmakers build good cars. They are training a general "world model" via cars, the biggest smart terminal. Horizon repeatedly emphasizes its “intuitive system.” Lü Peng believes real experienced drivers rely on intuition 95% of the time, not on calculations. Horizon’s “segment end-to-end” architecture is built to give cars this human-like intuition, so they don’t just mechanically follow "red light stop, green light go" code, but sense, interact, and strategize like living things. Pony.ai has introduced its VLA (Vision-Language-Action) model. Zhou Guang describes a scenario: when vehicles face complex tidal lanes or police temporary directives, traditional rule-based algorithms fail if they can’t read signs or gestures. But with the VLA model and COT (chain of thought) capability, a car can reason like a human: “Though the light is red, the officer is waving me on, and the sign says it’s allowed at this time, so I follow the vehicle in front.” Such logical reasoning is central to general robotics. Today’s smart cars are essentially four-wheeled robots. Every end-to-end large model trained by intelligent driving companies, every strategy for complex road interactions, will eventually be re-used on humanoid robots, logistics robots, and in the broader physical world. Pony.ai’s RoadAGI plan embodies this vision. They seek to make intelligent driving systems not just roadworthy, but capable of commanding robots to deliver food to your door. This “last 100 meters” ambition exposes the ultimate goal—to be the brain of the physical world. In this grand narrative, Robotaxi is no longer an unattainable research project. Both Pony.ai and Horizon have firmly chosen the "Tesla path": using data and hardware from mass-produced passenger cars, upgrading with large models to achieve L4 autonomous driving. When Pony.ai announced launching Robotaxi directly with mass-produced Great Wall Blue Mountain models (costing just 200-300k RMB), the unit economics of Robotaxi finally made sense. Robotaxi officially becomes a consumer-grade service rather than a research project. In this new paradigm, the boundary between assisted and driverless is erased. Millions of private cars, while serving users, also act as stealth data collectors—feeding corner cases to L4 models. This "laying eggs along the road" model smashes the heavy asset barriers of Waymo’s approach. The future of mobility will be defined not by taxi companies with thousands of modified vehicles, but by intelligent driving platforms with millions of mass-production data portals. This also explains why capital markets, even in winter, continue to invest in leading intelligent driving companies. They're not betting on car parts, but on tickets to the embodied intelligence era. For China’s intelligent driving industry, 2025 will be both the cold winter of a bubble burst and the warm spring of value return. Momenta CEO Cao Xudong once predicted: "Next year, China’s city assisted driving landscape will settle—only two or three companies may remain." This is not alarmist. Market share is accelerating toward the leading players like Huawei, Momenta, Horizon, etc. Those in the middle, suffering from technical homogenization and lacking self-sustaining capabilities, are being ruthlessly eliminated. For carmakers, intelligent driving has gone from an optional feature to a core asset. At the final stage, no one wants to hand over their soul completely. Which means the window for third-party suppliers is closing. Unless, like Huawei or Horizon, you can provide carmakers with overwhelming capabilities out of reach in the short term, or, like Zhiyu, offer extreme cost-performance. The future landscape is emerging: one group is self-developed carmakers with million-vehicle sales, using closed-loop data to iterate algorithms; the other is super Tier 1 suppliers serving the rest, pooling data via alliances to fight for a spot in the finals. When millions of vehicles become the life-and-death threshold, for Horizon, Pony.ai, and the rest still at the table, the real game has just begun. Risk Warning and Disclaimer Markets have risks, investment requires caution. This article does not constitute personal investment advice and does not take into account individual users' specific investment objectives, financial situation, or needs. Users should consider whether any opinion, viewpoint, or conclusion in this article suits their particular circumstances. Investment is at your own risk.