Horizon is preparing to launch a surprise attack.
```
Author | Chai Xuchen
Editor | Zhou Zhiyu
In the past few years, the industry debated who could achieve NOA, who could accomplish end-to-end, whose urban pilot is most like human driving. In the next few years, the critical question may become: Who can make users never want to live without it?
On April 11, 2026, Yu Kai, the founder and CEO of Horizon Robotics, gave a judgment at the high-level forum on the development of intelligent electric vehicles: As of the Spring Festival in 2026, the proportion of intelligent driving mileage of models equipped with the HSD system has exceeded 40%. Once it breaks through the 50% threshold, user dependence on intelligent driving will become irreversible.
In his view, this is similar to the tipping point when mobile phones moved from feature phones to smartphones. When users entrust most of their driving time to the system, intelligent driving is no longer merely a selling point, but will become standard. Whoever masters this capability will have a chance to become the core of the next round of the automotive industry chain.
Overtaking
Tesla once defined the first half of global intelligent driving.
The significance of FSD is not just to allow the car to change lanes automatically or park automatically, but to make consumers believe: cars can evolve continuously relying on software like mobile phones do. Tesla therefore gained a valuation premium far beyond traditional automakers.
But from the perspective of industry insiders, Tesla's lead was built on two premises: the largest-scale global data closed loop, and a long period without real rivals. Now, both premises are loosening.
In the Chinese market, Tesla still has brand appeal, but the local competitive environment has changed.
"Here, there is fierce price competition, complex road environments, rapid model iteration, and a group of users willing to try new things. The penetration rate of intelligent driving in China is already much faster than most global markets," an intelligent driving practitioner told Wallstreetcn.
In their view, the real beneficiaries might not be the automakers, but the "pickaxe sellers."
Yu Kai revealed that by August 2025, Horizon Robotics will have delivered 10 million sets of assisted driving systems, including 4 million delivered in just the past year. This shows that Horizon Robotics has crossed a common hurdle for Chinese tech companies: completing the closed loop of technical research, product, and large-scale commercial verification.
The financial report disclosed in March 2026 shows Horizon's 2025 revenue was 3.76 billion yuan, up 57.7% year on year; although all three expenses are also rising, Yu Kai is frank about burning cash, saying the company is continuously increasing R&D investment, with annual R&D expenses as high as 5.153 billion yuan, focusing on personnel, cloud computing power, and technology service procurement.
Calculated out, every 1 yuan of revenue burns 1.37 yuan of R&D expense. At this rate, there's not much time left for Horizon to verify its business model.
The most core asset for Horizon is HSD.
Yu Kai said HSD is the company's core product to win urban intelligent driving, and also the technical foundation toward higher-level autonomous driving. The system went into mass production in November 2025, landing first in mainstream models around 150,000 yuan. In just 8 weeks post-launch, shipments exceeded 25,000 units. By the end of last year, HSD had secured more than 20 designated models, and is expected to reach 400,000 shipments in 2026.
This proves high-level intelligent driving is breaking away from the exclusive label of high-end models and starting to sink to the mass consumption range.
An executive from a new energy auto startup told Wallstreetcn that in the past, industry consensus was that high-level assisted driving belonged to models above 300,000 yuan—a luxury feature. Now, it's appearing in the 150,000-yuan segment, and intelligent driving begins to meet conditions for large-scale popularization.
Yu Kai proposed that this year, the urban MPI (single takeover autonomous driving mileage) will be increased tenfold to reach several hundred kilometers. "If we achieve this goal, we'll surpass Tesla."
It's crucial to note that Tesla is a single-brand closed loop, but Horizon is a platform-type player. Its system can be installed on many automakers and price tiers. Every cooperative model sold helps to collect data, train models, and improve algorithms. The more automakers compete, the stronger the platform becomes.
This is why Horizon's software and service revenue already exceeds its hardware revenue.
In 2025, the company's automotive solutions business had product solution revenue of 1.62 billion yuan, while licensing and service revenue reached 1.93 billion yuan, accounting for more than half. Among them, software licensing gross margin was around 90%, far higher than hardware.
In other words, Horizon is evolving from selling chips, to selling systems, and then to selling ongoing services. Once intelligent driving mileage share breaks 50%, users form dependency, and subsequent software subscription fees, value-added services, and continuous OTA fees have a realistic basis.
Betting
After securing its footing in the intelligent driving track, Yu Kai plans to "switch lanes" and break through. He forecasts that cockpit-driving integration is the new trend, and Horizon will go ALL IN.
An informed source told Wallstreetcn that Horizon plans to launch two aces: Agentic CAR SoC (cockpit-driving integration whole car intelligence chip) and Agentic CAR OS (intelligent operating system). The former's code name is "Starry Sky," a chip supporting large models on the car side and planned to debut officially on April 22, 2026, with mass production within the year.
At this forum, Yu Kai “spoiled” that "Starry Sky" will support running an intelligent OS similar to openclaw "Lobster". According to the plan, they intend to debut this card at the Beijing Auto Show.
This means Horizon is not satisfied with making a smart driving chip, but aims to make the central brain of the whole car.
Currently, automotive electronic architecture is undergoing the biggest transition in the past decade.
An OEM R&D staffer told Wallstreetcn that traditionally, cockpit and intelligent driving are two separate systems in smart cars: two chips, two controllers, two sets of memory, two software teams. The advantage is independence; the disadvantage is high cost, poor synergy, and slow development. The core logic of cockpit-driving integration is merging the two systems into one central computing platform.
Yu Kai ran the numbers: through controller integration, 50% chip space can be saved; sharing parts and computing power can save 30% component costs; single vehicles can cut 1,500–4,000 yuan.
At current profit margins, this is a tangible attraction for carmakers.
The latest data from the China Passenger Car Association shows that in 2025, the Chinese automotive industry’s annual revenue is about 11.18 trillion yuan; but annual sales profit margin has dropped to 4.1%, and in December alone, the monthly profit margin fell to 1.8%.
The price war has pressed vehicle profit margins to extremely low levels; every thousand yuan saved is real money for carmakers. Especially in the 100,000–200,000 yuan mainstream market, cost-cutting directly determines product competitiveness.
Therefore, cockpit-driving integration may seem to be a technical route dispute, but its essence is first a profit dispute.
Hightech Intelligent Auto Research Institute data shows that in 2025, 1.5636 million units of pre-installed cockpit-driving integration computing units will be delivered in China, up nearly 50% year on year. In just January–September, new-car deliveries have approached one million units.
This means the market has passed the “concept verification period” and is entering the scale introduction period. By 2030, organizations expect cockpit-driving integration penetration to break 30%, with single-chip solutions becoming mainstream in the 100,000–200,000 yuan segment.
The window period for cockpit-driving integration SoC is opening. But things aren’t as easy as imagined; Qualcomm currently firmly controls the domestic cockpit chip market, with over 70% market share. Realistically, it is not simple for Chinese vendors to directly confront Qualcomm.
Yu Kai’s approach: don’t compete in mature markets, but target the next-generation market.
Yu Kai says some Chinese companies previously tried to directly challenge Qualcomm's cockpit position but did not see ideal outcomes. Horizon plans to connect chip architectures for intelligent driving and cockpit through foundational technological innovation and system-level optimization, link upper-layer software ecology, and create its own blue ocean market.
In other words, Horizon is going to circumvent Qualcomm’s strongest area and grab the market Qualcomm hasn’t fully occupied. This is Horizon’s unique advantage relative to pure cockpit vendors.
Yu Kai's logic is: it's hard to enter intelligent driving from the cockpit side, since intelligent driving needs safety redundancy and real-time decision-making; conversely, extending from intelligent driving to cockpit is a lower threshold. But Qualcomm is also moving towards cockpit-driving integration, and the cockpit ecosystem is far more mature than Horizon’s. Whether Horizon can build enough software ecosystem on the cockpit side remains an open question.
The industry chain is already restructuring around this trend. This year, Xpeng merged its autonomous driving center with its intelligent cockpit center to form a General Intelligence Center; Li Auto incorporated the autonomous driving team into its intelligent space system, reorganizing into three major teams: base model, software body, hardware body.
Organizational structure is often a precursor to trends. When carmakers no longer distinguish between cockpit and intelligent driving internally, it means they're defaulting to the idea: future automobiles will have only one system—an AI system. Horizon is betting on Starry Sky, and gambling on this future.
Tesla defined the first half of intelligent cars; Qualcomm ruled the era of cockpit chips. In the next phase, Chinese companies are fighting for the naming rights to the new order—which will also determine how long Horizon’s story can continue.
Risk Warning and DisclaimerThe market has risks, investment needs caution. This article does not constitute personal investment advice, nor does it take into account the unique investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions herein fit their specific situation. Investing accordingly is at your own risk. ```