Horizon’s Breakthrough in the “Unmanned Zone”

Author | Zhou Zhiyu

Editor | Zhang Xiaoling

In the Chinese automotive market of 2025, the "arms race" in intelligence has reached a fever pitch, and a concept has been upheld by all leading players: intelligent driving is the "soul" of the car and must be "fully self-developed," firmly in their own hands. From new forces to tech giants, countless companies have spent billions to gain a foothold in this "battle for the soul."

However, Yu Kai, the founder of Horizon Robotics and a scientist-turned-"intruder," has proposed a completely opposite "counter-consensus" prediction for this fervent gamble.

In a recent interview, he stated that what car companies today regard as the "soul" and core barrier of intelligent driving will ultimately evolve into a standardized "functional value" product, akin to the communication "baseband" that everyone uses but no one develops in smartphones. In his view, the vast majority of car companies will abandon self-developed intelligent driving in the future, opting instead to procure mature supplier solutions and focus their efforts on creating "emotional value" that truly reflects brand differentiation.

Since the beginning of this year, Horizon's latest generation of computing solutions, the "Journey 6" series, has been successively adopted by several new models from leading car manufacturers such as Li Auto, BYD, and Chery, with the high-end Journey 6P set to debut in Chery models within the year. This also serves as a key market validation of Yu Kai's "baseband theory."

This "intelligent driving baseband theory" not only directly challenges the current mainstream beliefs in the industry but also embodies Yu Kai's consistent "counter-consensus" philosophy.

Now, every participant in the automotive industry needs to rethink a fundamental question: In the second half of intelligent vehicles, what kind of industrial form and business model will be more viable? Is it a winner-takes-all "vertical empire" or a professionally divided "open alliance"? The path taken by Horizon and its "counter-consensus" allies will write a crucial chapter in answering this question.

From the Margins to the Table

In today's intelligent automotive industry, one word is revered as a guiding principle—"soul."

With Tesla as a lighthouse, leading players like Nio, XPeng, and Huawei regard "fully self-developed" as a core barrier. However, Yu Kai has proposed the highly controversial "intelligent driving baseband theory," believing that intelligent driving will ultimately become a standardized functional module.

"You can't say that autonomous driving has a style like Guo Degang or Lin Chiling; it's impossible," Yu Kai explained. "Whether it's men, women, the elderly, or children... the standards for intelligent driving experience are the same; it's all about getting from A to B, safely, comfortably, and efficiently." In his view, the future of car manufacturers should focus on the emotional value of products rather than spending huge sums to redundantly create a standardized "wheel."

This theory directly defines Horizon's strategic ecological position: at the table crowded with giants, it does not aim to compete for the "soul" as a complete vehicle player but is committed to becoming an indispensable "arms dealer" for the entire industry The survival rule of Horizon is to become "another choice": through the parallel development of software and hardware collaboration, providing a platform with higher cost-effectiveness and deeper services.

This "anti-consensus" theory did not come out of nowhere; its seeds have long been buried in the personal experiences of Richard Yu.

Richard Yu's "rebellion" was evident since his student days.

He originally wanted to be a painter but was accidentally "struck" by machine learning, and from then on, he could not stop. He plunged into the then extremely niche field of "deep learning," while shallow learning was the mainstream. The deep learning research group he was part of was one of only five in the world.

At the AI conference in 2002, which had only three hundred attendees, he witnessed the fierce debate between Yann LeCun and Geoffrey Hinton, and also saw Richard Sutton, the "father of reinforcement learning" who later won the Turing Award, eating alone due to his niche field being overlooked. This scene deeply impressed him: "This world is always created by a minority."

This determination for the "non-mainstream" led him to make the most critical "anti-consensus" decision when he founded Horizon in 2015. When everyone rushed towards software algorithms, he insisted on making chips because he believed that software and hardware are inseparable, and only dedicated hardware can maximize efficiency.

This path was exceptionally difficult; in the early days, Horizon was completely overlooked by China's semiconductor funds because they could not understand Horizon's model. Richard Yu recalled, "Most of the time, I was on my knees."

The entrepreneurial journey is far more brutal than financing. Richard Yu referred to the first five years of Horizon as the "darkest moment"—"dark and without light." Although he had just raised hundreds of millions of dollars in early 2019, he felt extremely uneasy inside. The company's strategy was scattered, simultaneously targeting multiple directions such as automotive and AIoT, making it impossible to focus resources, and the business was "tasteless to eat, but a pity to abandon." The organization was bloated, like a "brotherhood" maintained by emotions, "To allocate a person, I had to drink a bottle of Moutai with the head of that department."

The turning point occurred in 2019. At Lakeview University, Professor Zeng Ming's strategic class and the advice from Li Xiang, the founder of Li Auto, "You should focus on the automotive direction," made him determined to "give up nine for one," cutting off all businesses outside of automotive. He overturned HR's plan for "slow layoffs" and chose to reduce the company's size by half within a month. This "surgical" adjustment transformed Horizon and made him deeply understand the first principle of business: first, clarify who your customers are.

After focusing the strategy, Horizon began its difficult breakthrough in the automotive industry. "Every customer has to bow down," Richard Yu joked. The first breakthrough was with Changan Automobile. In 2018, when Changan was in a critical period of product breakthroughs, Horizon chose to cooperate deeply with them, with both teams "developing projects together in the scorching summer, exhausted to the point of sleeping outdoors at the site at night," forging a revolutionary friendship. In 2020, Horizon's first automotive-grade chip "Journey 2" achieved mass production on the Changan UNI-T, which became a hot-selling model that year If the cooperation with Changan is a "shared hardship," then the partnership with Li Auto is a "seizing of the window of opportunity." In 2020, Li Xiang decided to replace the foreign supplier Mobileye because it could not localize modifications for Chinese road conditions. This was a huge risk decision, but Li Xiang's "bold decision-making" met Horizon's "ability to handle challenges." The two sides completed the chip replacement and mass production of the Li ONE in just 8 months, once again creating a miracle of a blockbuster product.

From Changan and Li Auto to later securing BYD, Yu Kai attributes this to a core capability—"the essence of all business is empathy." Whether it's the joke about "deliberately losing a football match" or the alignment with Wang Chuanfu's "pragmatic" culture, it reflects the unique "street wisdom" of this scientist CEO.

Stars and Seas Beyond Automobiles

Horizon’s first battle was to become a reliable "baseband" supplier in the field of autonomous driving. Yu Kai has a clear roadmap for the future of this field.

He predicts that autonomous driving will progress in "three steps": achieving large-scale "hands-off" driving within 3 years, "eyes-off" driving in key scenarios within 5 years, and "minds-off" driving in limited scenarios within 10 years.

This prediction appears both forward-looking and relatively pragmatic, especially in light of the global industry's rational judgment that L4-level commercialization is generally postponed to around 2030.

However, the victory in autonomous driving is just the first step in Horizon's grand blueprint. Yu Kai's true ambition is to build a "CUDA of the robot era" or a "Wintel" alliance—an open, integrated computing platform to empower countless robots flying in the sky, swimming in the water, and working in factories and homes.

He believes that driving a new computing paradigm and defining hardware and software standards is more exciting than creating a specific product. This is the vast starry sea that is broader than the automotive market.

This ultimate goal is rooted in Yu Kai's deeper technological philosophy. He places his vision within the wave of the global robot era. This is not only the exploration direction of tech giants but has also been elevated to the level of national strategy.

Yu Kai remains somewhat cautious about the development of AI. He worries that in the digital world, humans are "farmed" by recommendation algorithms and information cocoons. Therefore, Horizon has chosen a more challenging path—doing AI in the physical world.

"Let machines be machines, and let humans be humans." This is the mission that Yu Kai repeatedly emphasizes. He hopes that Horizon's technology can liberate humans from boring, heavy, and dangerous physical labor, allowing them to engage in more creative and emotionally valuable work, rather than sinking into the virtual world.

To achieve this goal, Horizon must make disruptive innovations in computing architecture.

Yu Kai points out that the human brain consumes only 20 watts but has a computing power of up to 5000T. Yet today, even the highest-performing chips consume far more than 100 watts, while their computing power is far behind. The future challenge is to overturn the existing von Neumann architecture, integrating computing and storage to achieve an order of magnitude improvement in energy efficiency The landing of Journey 6 is a key battle in Horizon's "Baseband" strategy, proving that in the field of smart cars, the "Horizon model" of professional division of labor is a viable path with strong competitiveness.

However, the war is far from over. On one hand, top chip suppliers are continuously launching next-generation computing platforms with stronger computing power, further consolidating their technological barriers; on the other hand, technology giants with deep technological ecosystems are also crossing over, attempting to reshape the industry landscape through integrated "car-cloud-edge" capabilities. Meanwhile, those car manufacturers that insist on fully self-developing core technologies are also striving to transform hundreds of billions in R&D investment into their own, hard-to-replicate moats.

Horizon has successfully transformed itself from a marginal "anti-consensus" challenger into an undeniable force at the table. But in the future, whether it can promote the "Baseband" model as an industry standard in this ultimate war over the "soul" of cars depends not only on the speed of technological iteration and cost control capabilities but also on the final choices of the entire automotive industry regarding efficiency, value, and core competitiveness. This road remains long