Author | Shi Jie, Managing Editor | Shi Jie, Editor | Wang Yue
"Still waters run deep" is the only way out. Who can "ignore the hype and just do the homework"?
As both the industry and consumer markets reach a stage where "AI is on everyone's lips," the true test for enterprises has arrived.
In Lingang, Shanghai, the afternoon sunlight of September slants into the conference room on the third floor of the Horizon Building.
"Horizon chose this location for its events and industrial park because it offers a view of the first glimpse of Shanghai's 'horizon,'"
Before Yu Kai, founder of Horizon, lay a set of figures: 15 million Journey series chips, 20 million units of overseas orders, and one in every three intelligent driving vehicles in China equipped with Horizon's solutions. The 15 millionth chip was just deployed in the FAW-Volkswagen ID. AURA T6.
Yet, sitting across the table, Yu Kai's face did not show much of a "winner's" expression.
He said calmly, "After the mass production and shipment of 15 million chips, I was happy for two seconds before turning the page. Greater challenges must lie ahead, and harder things will always come one after another. After all, we are climbing such a steep slope."
Indeed, if intelligent advanced driver-assistance systems (ADAS) were the most favored track across all industries two or three years ago, the flavor of the entire environment has changed by 2026.
Capital and public opinion are increasingly focusing on more talk-worthy fields such as "embodied intelligence" and "robotics." Even Horizon itself was initially named "Horizon Robotics," and has now launched D-Robotics products.
However, behind the wave of AI and robotics lies another unequal formula: before 2030, global annual AI investment will exceed USD 1.5 trillion, while the market size of generative AI by then will be only USD 150 billion. A trillion-dollar gap lies between input and output.
"Victor Gao predicts that the US AI bubble will burst within 10 months. Will this affect Horizon's business decisions?" I posed this question to Yu Kai.
How will Horizon choose?
Will it follow the name "Horizon Robotics," shift its future focus to robotics while downplaying intelligent driving, to escape the fate of being "a thing of the past"? Or will it stick to its automotive intelligent driving business and criticize the AI bubble?
In fact, this is no longer just an inquiry about a single company's strategic direction, but a soul-searching question for the entire manufacturing and technology sectors.
01. "Enduring Solitude" is a Survival Strategy
"We have always chosen the relatively unconventional path in our strategic choices," Yu Kai said in a flat tone, as if stating a long-verified common sense.
He repeatedly emphasized that Horizon's values are encapsulated in eight Chinese characters: "Empower Customers, Endure Solitude." The first four represent business common sense, while the latter four constitute the true threshold.
But if it is just talk, it would be too much like a superficial slogan. As industry observers, we must not only "listen to their words" but also "observe their actions."
In 2015, during the first wave of AI hype, the facial recognition track was crowded with star companies and top-tier capital. Horizon chose chips—a direction that was unconventional, capital-intensive, long-cycle, and unpromising at the time.
After the intelligent driving track took off, the most dazzling narrative was Robotaxi, "disrupting mobility," and valuation stories often reaching tens of billions of dollars. Horizon again chose ADAS chips—which didn't sound sexy enough, nor "world-changing" enough.
As Wang Huai, one of Horizon's early investors from Linear Capital, evaluated Yu Kai: "Having the foresight to keep warm in winter during summer," Yu Kai's forward-looking vision and sense of crisis are the prerequisites for today's success. Therefore, choosing what is right rather than what is popular naturally became Horizon's strategic direction.
Time has proven that being unconventional does not equal being wrong, and being popular does not equal being right. Where are the leading players in the facial recognition track back then? And how much of the Robotaxi valuation myth remains?
Therefore, the four characters "Endure Solitude" are not just promotional rhetoric for investors and the general public; they are lessons on survival strategies summarized by Yu Kai and Horizon after fierce battles.
"Today, we are still not the brightest or the most popular, but we enjoy this relatively unconventional state. Because it keeps us more sober and enables us to think more clearly about our product rhythm and core strategy, rather than being in the vortex of the times." Yu Kai paused, "Staying sober while in the vortex is actually very difficult."
These words carry weight. Over the past two decades, China's technology industry has experienced too many "vortex moments"—the thousand-group war of group buying, the O2O bubble, the craze for the sharing economy, the metaverse concept, and the large model arms race.
Every wave has swept up countless smart people and hot money, and every ebb has left a mess behind. Those companies that have truly crossed the cycle are often not the most dazzling ones on the crest of the wave, but the ones still at the table after the tide has receded.
"Nothing in this world can be so hot that it turns purple and remains eternal; there is always a cycle. A field may be cold now, but it could be hot tomorrow; hot today, cold tomorrow," Yu Kai said, "Just adapt to it."
02. Talent Outflow = Maturity of Intelligent Driving
Another intriguing phenomenon in the industry is that executives and talent in the intelligent driving circle are flowing into the field of embodied intelligence.
To the outside world, this seems like a signal of "blood loss" in the intelligent driving industry, and some even worry it will affect the sustainable development of intelligent driving businesses.
Yu Kai's judgment is exactly the opposite.
"This is inevitable. Horizon's English name has always been Horizon Robotics. We believe that towards the era of robotics, the first stop must be to do a good job in intelligent driving. For talents who have been fully trained in intelligent driving, the next stop being embodied intelligence is very natural."
He further broke it down: the outflow of talent precisely indicates that the intelligent driving industry has highly converged. "It is no longer the early state where a large number of new companies emerged. In the future, whether it is OEMs' in-house R&D or third-party technology companies, the number of players will only decrease, and talent will inevitably overflow to their next rightful destination."
This is not merely a contrarian perspective: the flow of talent to embodied intelligence is precisely a sign of the maturity of intelligent driving.
Data supports this judgment. In the first half of 2026, Horizon's market share in ADAS chips for domestic independent brands exceeded 50% for the first time, about twice that of the second place. Who is the second? Yu Kai smiled and asked around the scene, but no one could answer immediately. "Hasn't it highly converged?" he countered.
A highly converged industry will no longer attract hot money, nor will new companies rush in. For Horizon, this is not bad news, but good news.
"Horizon is in a relatively good state. It is not an emerging industry, but one that is beginning to show an inflection point, moving towards natural and healthy growth," Yu Kai said. "We focus on our own business, do a good job with each generation of products, and deliver every car model well. There is no need to tell new stories; just doing our business well and singing our own song is enough."
Yu Kai repeatedly mentioned "still waters run deep." This phrase seems out of place in the impetuous AI era, yet it precisely summarizes Horizon's posture for survival.
Still waters run deep does not mean standing still, but rather turbulent undercurrents beneath a calm surface. Behind the shipment of 15 million chips are eleven years of stumbling tape-outs, validations, and mass production ramp-ups.
"Most of the time is not a highlight moment; most of the time is full of mud," Yu Kai admitted. Back then, when taping out a chip, he was anxious, not knowing if the tape-out would be successful. "It was really bitter."
But it is precisely this "bitterness" that has built the moat. There are no shortcuts in the chip industry; the iteration of each generation of products requires real money for R&D investment and validation on real roads.
Over 8 million cars equipped with Horizon's solutions drive on the roads daily, accumulating over 100 billion kilometers of real road mileage, covering the globe, all-weather, and all-working conditions—these figures cannot be achieved overnight, nor can they be replaced by PPTs.
In terms of business model, Horizon has chosen "the hardest technology" combined with "the most flexible posture."
The Wintel model directly sells chips and software, while the Arm+Android model provides IP licensing and technology empowerment. The two models run in parallel, serving the 80% of mainstream OEMs that need integrated solutions, and also empowering the top 20% of automakers that wish to pursue vertical in-house R&D.
"95% of Horizon's shipments are completed through partners. We already have orders for over 20 million vehicles in the global ADAS market, in many markets we haven't even been to."
03. Ignore the Hype, Just Do the Homework
Back to the issue of the AI bubble. Yu Kai's sobriety lies in his clear understanding of what Horizon can and cannot do.
"Since Horizon holds high the banner of being down-to-earth and seeking stability, have you ever thought about using the reliable label of manufacturing to 'pull back' the impetuous trend of AI and intelligence?"
I asked this deliberately to test Yu Kai's mindset. If someone is only superficially down-to-earth, they would often follow the question and talk at length about "how we seek stability and hope to change the industry's ethos."
"We cannot have such delusions. Who are we at Horizon to pull back the AI wave? That is an illusion. It is already extraordinary enough for us to do our own homework well." As expected, Yu Kai once again demonstrated a higher level of eloquence through humble words.
Then, he continued to elaborate on Horizon's philosophy.
"It is better to go home and weave a net than to stand by the water envying the fish. I need to do my own homework well, get the scriptures Horizon needs to obtain, and lead humanity towards ubiquitous, free, and safe advanced driver assistance and even autonomous driving—this alone makes me feel that it requires Herculean efforts. I don't care about other things."
This is not passive escapism, but strategic determination. When the capital market's attention is absorbed by data centers and large models, when the embodied intelligence track is crowded with dozens of unicorns, and when the automotive industry launches over 580 new car models in the first half of the year, with involution leading to "astonishingly low gross and net profit margins"—Horizon has chosen another path.
"We don't need to tell new stories," Yu Kai said. "The field of intelligent driving does not belong to the noisy and lively domain like embodied intelligence; it is now in a normal domain of 'still waters running deep.'"
But still waters running deep does not mean a lack of ambition. Horizon's goal is clear and grand: to become the definer of computing platforms in the automotive era, just as Wintel defined the PC era and Arm+Android defined the mobile era. First, establish the computing standards for the automotive industry, and then move towards robotics—because automotive computing is the inevitable path to future ubiquitous robotics computing.
"For an enterprise, being healthy and living long is much more important than living flashily." When Yu Kai said this, there was no impassioned tone in his voice, only a kind of certainty after eleven years of muddy trekking.
Will the AI bubble burst? When will it burst? What impact will it have on Horizon if it bursts? These questions may be important, but in Horizon's coordinate system, none of them are the most core variables. The most core variables are always: whether the products are good enough, whether the delivery is stable enough, whether customers trust enough, and whether the technology moat is deep enough.
Hot money will recede, trends will shift, and bubbles will burst. But the trend of human mobility towards autonomous driving will not change, the industrial law of underlying computing platforms converging to two or three players will not change, and the value of safe, reliable, and smooth intelligent driving experiences to consumers will not change.
Horizon bets its strategy on these "unchanging" things. Because only unchanging things can generate the compound interest effect.
"We enjoy the current state of 'still waters running deep,'" Yu Kai said.
The successful foothold over the past eleven years is precisely the result of still waters running deep. Whether Horizon can deliver another eye-catching answer sheet in the next stage also depends on whether it can continue to "ignore the hype and just do the homework."