Written by | Xu Zhi, Edited by | Wang Pan
Yu Kai, the CEO of Horizon Robotics, who is typically low-key, has rarely made public statements subtly criticizing competitors.
When Momenta went public in July this year, Yu Kai posted on Weibo self-deprecatingly describing Horizon as "not very good at navigating social circles" and "a rather boring company." Many interpreted this as a jab at Momenta's capital narrative of being the "first physical AI stock."
Recently, XHEART, a chip company deeply tied to Momenta, launched its new product, the X7 Ultra, claiming it features 700 TOPS of large model computing power, with the next-generation X9 targeting 3000 TOPS.
In response, Yu Kai posted on Weibo again, stating, "For such a serious matter as intelligent driving chips, China's tech industry is engaging in an endless race for PPT-defined computing power," implying that competitors' chips overly pursue paper specifications.
This break in silence likely stems from XHEART's unusually rapid R&D and mass production speed.
Founded only in 2023, XHEART's first product, the X7, was already installed in vehicles by March of this year. By the recent launch event, it had shipped approximately 200,000 units in just a few months, equipped in 10 mass-produced models, and secured orders for dozens more.
It can be said that XHEART completed the journey from chip R&D to mass production in just over two years, covering what similar companies usually take five to eight years. The core reasons include not only the collective migration of the experienced team from Zhiku (Zeku), but also significantly, the support from Momenta: its algorithms and mass production delivery experience quickly filled XHEART's gaps in understanding requirements and validating scenarios.
In simple terms, XHEART designed its chips from the outset based on end-user application needs, which is why they proposed the concept of "large models defining chips."
Of course, the cost, performance, and stability of XHEART's products still require longer-term validation, and it cannot be said that they pose a sufficient threat to Horizon Robotics yet.
However, there is no doubt that Horizon Robotics and Momenta, two former rivals who competed more in distinct lanes, are now engaging in close combat on multiple fronts. The focus of competition has escalated beyond just customer numbers or market share to the solidity of the integrated hardware-software "foundation" and influence in the next phase.
Multi-Front Confrontation
"People who really care about software should make their own hardware," a viewpoint proposed by Turing Award winner Alan Kay, later championed by Steve Jobs, has long been a key creed for Horizon Robotics.
Now, Momenta appears to be following the same path, with XHEART as its solution.
Although there is no direct equity relationship between XHEART and Momenta on the surface, their ties are deep. XHEART itself originated from Momenta's chip project; Chairman Li Jun is the former head of Momenta's chip division. According to multiple media reports, Momenta founders Cao Xudong and Xia Yan indirectly hold shares in XHEART through controlling shareholders.
While Momenta did not enter the field directly, it retains a certain kinship with XHEART. An industry insider who attended XHEART's launch event told us, "I have never seen a software company and a chip company so tightly bound."
One consideration is certainly cost. Given Momenta's massive scale, there is limited room for upward expansion; reducing costs and increasing efficiency is a better evolution.
The internal competition in the current intelligent driving industry is nearly as fierce as among automakers. Previously, industry sources revealed that in bidding for projects across multiple vehicle models, Momenta often lost out to competitors "more willing to compete on lower prices" due to pricing issues.
The other ambition is raising the ceiling of capabilities. XHEART's products allow algorithm models and chips to iterate jointly from the initial architectural design stage, aiming to ensure the chips perfectly carry and adapt to Momenta's world model.
Previously, Yu Kai mentioned that chip development is a marathon, often requiring five to eight years from R&D and certification to mass production. High barriers and long cycles are core moats for Horizon Robotics. Therefore, XHEART has merely opened the局面 (situation); the challenge ahead is maintaining acceleration.
Compared to Momenta's "software-to-hardware" approach, Horizon Robotics follows the standard "hardware-to-software" path, with its HSD solution, based on its own chips, serving as the core carrier.
Anxiety stems from the stalemate in the main battlefield. Horizon Robotics' chip business faces a "three-front squeeze": first, overseas giants represented by NVIDIA; second, the increasingly obvious trend of automakers developing their own chips; and third, the catch-up efforts from second-tier rivals like Black Sesame Technologies and Axera Semiconductor.
Therefore, although HSD started late and disrupted some of Horizon's original ecosystem balance—beginning to "compete for meals" with ecosystem partners—it remains a step Horizon must take.
Similar to XHEART, Horizon's HSD, which did not start from scratch, has also demonstrated astonishing development speed, including extremely short mass production cycles and rapid scalable expansion. Currently, HSD covers several mainstream domestic automakers, and Horizon has set a new goal of achieving tens of millions in mass production within the next few years.
These two companies, whose main battlefields were previously algorithm solutions and chip hardware respectively, are now extending their capabilities and engaging in direct confrontation on both fronts.
This is a match between equals. Both are leaders in their respective fields, with deep team, financial, and technical barriers. In terms of corporate temperament, both exhibit highly pragmatic styles.
Momenta's current advantage lies mainly in the customer scale of its software solutions, especially among overseas manufacturers; while Horizon's advantages lie more in engineering experience in the hardware sector and penetration rates among domestic manufacturers.
Battle for the Endgame
The early landscape of "Di-Da-Hua-Mo" (Horizon, DJI, Huawei, Momenta) belongs to the past.
Huawei, due to its irreplaceability, largely operates independently; while DJI Automotive, now Zhuoyu Technology, competes more against emerging players like DeepRoute.ai and QCraft, focusing heavily on software algorithms and striving to capture niche markets outside of Momenta and Huawei.
Meanwhile, Horizon Robotics and Momenta are forming direct competition in the field of integrated hardware and software.
Of course, whether it is Momenta's chips or Horizon's advanced intelligent driving solutions, neither can strictly be called a "second curve." The common goal of both sides is to position themselves for the next stage of competition.
First, the next generation of chips will see more direct close-quarters combat.
XHEART plans to release the X9 chip in 2027, targeting 3000 TOPS of computing power and 700 GB/s memory bandwidth, aiming to bypass the current cost war and plant its flag directly at the high ground of world models.
Similarly, Horizon's next-generation flagship Journey 7 series is also planned for release in 2027, with target computing power "significantly exceeding NVIDIA Thor-X" and featuring the fourth-generation BPU. Its core purpose is not solely to meet automotive intelligent driving needs.
It can be said that the focus of next-generation chip competition is the right to define toolchains and ecosystems, as well as industry influence.
A further manifestation is the migration of capabilities toward embodied intelligence, or physical AI. This may be the true new curve for both Horizon and Momenta, and the same antidote available to both companies beyond the intensely competitive intelligent driving industry.
The ticket supporting their path to physical AI is precisely the foundational capability formed through native binding of hardware and software.
In fact, Horizon and Momenta have already made significant deployments, albeit with differences in paths and tactics. The former incubated D-Robotics and built an ecosystem through investments; the latter more directly reuses passenger car intelligent driving capabilities, personally entering L4 businesses such as Robotaxi and Robovan.
This is driven not only by the need to escape intense competition but also by talent retention pressures, as both companies have seen many top talents leave in recent years.
Regarding Momenta, besides the well-known Ren Shaoqing, individuals like Gao Jiyang and Chen Jianxing have left to found Starry Sea and Simple Intelligence Robots, respectively.
Recent media reports indicate that Cai Na, Momenta's former VP of Product, has also joined the embodied intelligence sector, becoming the actual controller of Yangu Intelligent.
Horizon, meanwhile, is undoubtedly one of the "Whampoa Military Academies" for embodied intelligence. Many departed executives are founders of renowned embodied intelligence enterprises, such as Yu Yinan of Vita Power, Zhang Yufeng of Wujie Power, Du Dalong of Octopus Power, Niu Jianwei of Dingdang Power, and Huang Guan of Jiaji Vision.
If viewed solely from the perspective of intelligent driving, the conflict between Horizon and Momenta will undoubtedly intensify; however, if the perspective is widened to the entire field of physical AI, both are still in the early stages. Neither side has a compelling reason for a "must-win fight." They are more like advancing toward the same goal from different starting points, with occasional collisions arising from crossing paths.
For Horizon, the threat does not come primarily from the chip products themselves, but rather from Momenta's determination to complete its foundational capabilities. The essence of this race is whose foundation is deeper and more stable, which determines who holds greater and prioritized influence in the next stage.