Author | Zhou Zhiyu
Horizon has started stockpiling ammunition for the next battle.
According to Wall Street News, Horizon is accelerating the development of the next-generation intelligent driving chip Journey 7 series, which is based on the new generation of Riemann architecture, with performance comparable to Tesla's AI5, and is planned for release and mass production next year. At the same time, Horizon will also launch an integrated cockpit and driving solution, expected to be unveiled during this year's Beijing Auto Show.
These two moves are becoming increasingly urgent, as evidenced by the 2025 financial report just released on March 19. Revenue reached 3.76 billion yuan, a year-on-year increase of 57.7%, with NOA chip shipments surging nearly fivefold—both volume and price have risen as expected, but Horizon's CEO Richard Yu devoted more time in the earnings call to discussing the future. The reason is simple: the Journey 6P is still ramping up, with an absolute ASP of only about $60, and the release of high-value products has just begun.
The changes in the industry this year are faster than those within Horizon itself. NVIDIA's Thor has been adopted, Nio and XPeng's self-developed chips have entered mass production, and Qualcomm is moving from cockpit to intelligent driving. The intelligent driving chip market is evolving from a single model of &#34;chip companies supplying, car manufacturers purchasing&#34; to a dual-track pattern of self-developed chips by car manufacturers coexisting with third-party solutions. Horizon needs to seize the window of opportunity in the tightening next cycle.
<h2>Holding the Cards</h2>
Wall Street News has learned that Horizon has listed the integrated cockpit and driving intelligent system as the most important strategic product for this year. The April Beijing Auto Show will be the first public appearance of this solution.
Horizon is already the number one in the intelligent driving domain for domestic chips—holding a 47.7% market share in the domestic brand ADAS market, and 44% in the NOA market for vehicles priced below 200,000 yuan. However, just focusing on the intelligent driving domain has its limits. By 2025, total chip shipments are expected to exceed 4 million units, with 1.8 million units of mid-to-high-end chips supporting NOA, nearly five times that of the previous year, but only accounting for 45% of total shipments.
Richard Yu calculated that the $60 ASP has a tenfold upward potential compared to the average price of urban NOA products. While shipments are increasing, the proportion of truly high-value segments remains low.
The cockpit domain is a different world. Qualcomm monopolizes with over 70% market share, and Snapdragon 8295 is almost synonymous with cockpit chips in China, while the next-generation Snapdragon 8797 has secured contracts with Leapmotor's D series and SAIC-GM. Many domestic companies have attempted to directly challenge Qualcomm in recent years, but none have succeeded.
Richard Yu also candidly stated in the earnings call: &#34;We have no intention of engaging in close combat with Qualcomm in the mature cockpit market.&#34;
What he aims to do is to change the dimension—integrating the intelligent driving and cockpit domains onto a single chip, using the full-stack capability of &#34;chip + intelligent driving software + intelligent system OS&#34; to carve out a niche that neither Qualcomm nor NVIDIA can reach.
Qualcomm's shortcoming lies in the intelligent driving software ecosystem, as high-level intelligent driving solutions largely depend on third-party algorithm companies like Yuanrong and Momenta, lacking a complete intelligent driving software stack. NVIDIA's shortcoming is in the mass production of integrated cockpit and driving solutions—it has been collaborating with a luxury car manufacturer overseas on an integrated cockpit project for several years, but has yet to achieve mass production. Horizon believes it is currently the only player capable of seamlessly integrating from chip to intelligent driving software to intelligent system OS
Yu Kai's judgment is that entering the cockpit from intelligent driving is a &#34;more dimensional strike&#34; than entering intelligent driving from the cockpit.
The integration of cockpit and driving is becoming a clear industry trend. Tesla's FSD+Grok has shown the industry the imaginative space for the integration of cockpit AI and autonomous driving, while Li Auto and XPeng have merged their intelligent driving and cockpit teams at the organizational level. In the low-end market, BAIC's Arcfox has already equipped its integrated cockpit and driving solution on models priced at around 100,000 yuan.
Industry insiders have analyzed to Wall Street News that if current intelligent driving suppliers cannot secure a position in this trend, they may lose both the intelligent driving and cockpit markets in the future.
The cost calculation also stands firm. The integration of cockpit and driving can save on hardware costs such as wiring harnesses, cooling, PCB area, and domain control enclosures. Yu Kai specifically mentioned the soaring prices of memory chips this year—just the memory alone is expected to save car manufacturers thousands of yuan. In an industry environment focused on cost reduction, this is a calculable expense.
However, the cockpit is ultimately Qualcomm's main arena, having operated five generations of products. The Snapdragon digital chassis, covering over 350 million vehicles globally, along with its mature multimedia ecosystem and toolchain, cannot be easily overshadowed by a new chip.
<h2>The Battle Ahead</h2>
The Journey 7 addresses the issues of the next cycle.
Wall Street News has learned that the development of Journey 7 is progressing smoothly, and there is high confidence internally in aligning its performance with Tesla's AI5. Previously, Yu Kai also expressed to Wall Street News that in comparing parameters with Tesla's next-generation AI5, Journey 7 has &#34;significant advantages.&#34;
Unlike Journey 6, which was led by the chip team, the product definition of Journey 7 comes largely from the algorithm team. The vehicle-side model is rapidly evolving, expanding from millions of parameters to billions, and the chip must reserve space for model iterations over the next two to three years. Allowing algorithms to define the chip itself is a signal.
Competitors are already in plain sight. The demand for chip computing power is rising from hundreds of TOPS in the past to thousands of TOPS. The C7H chip disclosed by Horizon and its joint venture with Volkswagen, based on the Journey 7 architecture, has an AI computing power of 500-700 TOPS and is built on a 3-4nm process. According to Horizon's past product naming conventions, the C7P, which is a level above C7H, has a computing power of at least 1000 TOPS.
Moreover, the focus of the industry is shifting. Simply stacking peak computing power is no longer the only standard; memory bandwidth, data transfer efficiency, and inference accuracy may all be the real bottlenecks for deploying large models on vehicles. Tesla's FSD, which operates at 144 TOPS with HW3.0, remains the efficiency benchmark—deep collaboration between software and hardware is more important than merely stacking computing power.
The time window is narrow. Musk has stated that AI5 is about to complete tape-out and will be mass-produced in 2027. Journey 7 is targeting the same timeline. Before that, Horizon must prove its software iteration capability with the mass production scale of Journey 6P+HSD.
Financial reports show that Horizon's R&amp;D expenses for the entire year of 2025 will be 5.15 billion yuan, a year-on-year increase of 63.3%, far exceeding the 3.76 billion yuan in revenue—Yu Kai's attitude is not afraid of R&amp;D investment, as HSD is not only the core product for winning urban intelligent driving but also the technical foundation for achieving L3/L4
HSD only started mass production last November, with shipments reaching 22,000 sets by the end of the year, and a target of 400,000 sets this year. The models that first launched with HSD, which is positioned as a core configuration, have a high-end version sales ratio of 83%, breaking the industry's convention of &#34;low configuration for volume, high configuration for show.&#34; During this year's Spring Festival, the proportion of user intelligent driving mileage reached 41%—close to the critical point where machine driving mileage exceeds human driving mileage at 50%. These figures indicate that consumers are willing to pay for effective intelligent driving, but also show that Horizon's high-value products are currently concentrated in a few models, making this year crucial for whether they can replicate from a few models to dozens.
The MPI roadmap provided by Richard Yu is: this year to increase from tens of kilometers to hundreds of kilometers, next year to thousands of kilometers, with some regions reaching tens of thousands of kilometers, thus meeting the operational conditions for L4 level Robotaxi.
In the second half of the year, Horizon also plans to launch a Robotaxi pilot in collaboration with ecological partners. This path is highly similar to Tesla's, as FSD has already achieved over a thousand kilometers of comprehensive MPI in the U.S., with Robotaxi testing initiated in some areas. Horizon's pace is about a year behind, but the direction is consistent.
<h2>Who Remains at the Table</h2>
A larger change is underway.
The competitive landscape for intelligent driving chips is evolving from a &#34;competition among chip companies&#34; into a multi-party melee. On one hand, Nio, XPeng, and Li Auto's self-developed chips have either started mass production or are close to it, while BYD and Xiaomi are also making moves. These former customers are beginning to build their own capabilities, shrinking the space available for third-party chip companies in the high-end market.
On the other hand, algorithm companies are also extending into the chip segment. The first-generation assisted driving chip from XinHangTu, closely collaborating with Momenta, is expected to start mass production this year and has secured clients such as SAIC Volkswagen and Beijing Hyundai. In the future, it is not ruled out that Momenta's designated clients will shift to models equipped with this chip.
Momenta's founder, Cao Xudong, previously predicted that &#34;the competition in intelligent driving will end in 2026, with only three participants ultimately succeeding.&#34; Huawei's Qian Kun has already established a foothold in the high-end market, while the direct confrontation between Horizon and Momenta in the market below 150,000 is just beginning.
Horizon’s advantage lies in the scarcity of its ecosystem. It is currently the platform with the largest shipment volume and the most cooperative car companies among domestic intelligent driving solutions, with over 95% of shipments delivered through ecological partners. However, this open ecological model also has its tensions. As HSD extends from chips to software solutions, Horizon is involved in both hardware and software, blurring the boundaries of its role. The Journey 6P was previously exclusively used by HSD, but is now open to partners such as Yuanrong and Bosch—while there are more partners, some have begun to simultaneously engage with NVIDIA and other chip manufacturers.
Going overseas is another variable. Horizon has accumulated mass production designations from 11 car companies and over 40 overseas models, and has secured designations from three international car companies in markets outside of China, with a total lifecycle shipment volume reaching 10 million sets. However, Momenta has entered the global OEM system from the beginning with a posture of &#34;algorithm globalization,&#34; covering over 160 models including General Motors and Nissan. The competition for overseas markets is also accelerating
Yu Kai revealed that the revenue growth expectation for the next few years has been raised from 50% to 60%, with confidence stemming from a robust pipeline and the ramp-up of mass production for the Journey 6 series. However, he is also aware that Horizon faces not only a product competition but also a competition of ecosystems and time.
An undeniable fact is that Horizon's R&amp;D expenses for the entire year of 2025 are expected to be 5.15 billion yuan, a year-on-year increase of 63.3%, with an adjusted operating loss of 2.37 billion yuan. In the chip industry, the conversion of R&amp;D investment into product competitiveness requires a cycle of three to four years, while the iteration speed of autonomous driving models is once every six months. The battle for the Journey 6 is not over, yet the R&amp;D for the Journey 7 must be accelerated. This resource pressure of fighting on two fronts is not easy for any company.
The landscape for autonomous driving chips is taking shape. A period of lag may lead to structural gaps. Horizon must also accelerate its pace to maintain its hard-won leading advantage among domestic autonomous driving solution providers

TSLL

TSLZ

TSDD

The acceleration of the second half of intelligence

- Horizon Robotics is advancing its next-gen autonomous driving chip, the Journey 7 series, anticipated for release and mass production by next year, following significant revenue growth in 2025.  
- The company aims to merge driving and cabin controls into a single chip solution, tackling competition from established players like Qualcomm and NVIDIA, while navigating a rapidly evolving industry landscape.  
- Horizon's research and development expenses are expected to reach 5.15 billion yuan, illustrating the company's commitment to innovation, even as it faces significant operating losses and a competitive market.

Horizon wants to compete with Tesla, countdown to mass production of Journey 7