Meet Tesla FSD's Fiercest Competitors in China
There are two types of ADAS players in China: self-developed OEMs like Li Auto and XPeng, and Tier-1 suppliers such as Horizon Robotics(地), DJI(大), Huawei(华), and Momenta(魔).
Starting December 1, 2025, Tesla has launched another 30-day trial of its latest advanced driver-assistance system (ADAS), Full Self-Driving (FSD) version 14.2, which can navigate through urban, residential, and highway environments, respond to traffic signals, switch between lanes, and park automatically.
FSD is classified as Level 2 partial automation, meaning drivers must keep their hands on the steering wheel, remain attentive, and most importantly bear responsibility for any accident. But the system’s ultimate goal is to achieve fully autonomous driving in any situation, at any time.
The U.S. electric vehicle giant is also eyeing FSD expansion into China, its second-largest market, where data sovereignty and regulatory constraints restricted its entry until February 2025. Yet even after approval, mainland Chinese owners are limited to a less advanced FSD option poorly adapted to local road conditions.
In November, CEO Elon Musk told the company’s annual general meeting he expects full approval in China “around February or March (2026) or so.”
Meanwhile Tesla faces strong domestic competition. As the home of the world’s largest EV makers such as BYD and Geely, China has nurtured a batch of companies developing rival ADAS systems that threaten Tesla’s technological edge.
Two major headlines over the past week for Chinese ADAS companies:
First, Zhuoyu Technology, a 2023 spin-off from DJI, raised RMB 3.6 billion ($500 million) in strategic investment from state-owned FAW Group.
Second, in an unusual turn, Reuters retracted its earlier report claiming that Momenta, an autonomous driving startup, had filed for a Hong Kong IPO. Momenta received approval from China’s securities regulator to go public in the U.S. in June 2024, but hasn’t made it yet.
Why ADAS Became the Battlefield
ADAS isn’t a single technology. It’s more like a spectrum. Basic Level 2 systems can accelerate and brake, handle adaptive cruise control, and lane keeping. Level 2+ adds features like automatic parking and traffic sign recognition.
Then there’s Level 2.99, a uniquely Chinese marketing term. These systems support Navigation on Autopilot (NOA), meaning the car can technically drive itself from point A to point B, handling highway merges, exit ramps, and urban driving. However, drivers remain legally responsible because Chinese regulators haven’t approved Level 3 ADAS, which would allow drivers to truly disengage under specific conditions.
China’s EV market is notorious for cutthroat competition and homogenization. With an overly mature supply chain, building an EV is no longer difficult. Carmakers have responded by slashing prices while piling on features—capabilities, interior space, range. Intelligent driving has emerged as one of the few areas that can differentiate.
The numbers tell the story. According to China’s Ministry of Industry and Information Technology, L2-capable new passenger vehicles reached 55.7% market penetration in the first half of 2024. Projections suggest this will exceed 65% in 2025.
Consumer demand is accelerating the trend. A 2024 report by the China EV100 found that 90% of Chinese consumers are willing to pay extra for advanced intelligent driving services, with 30% willing to spend over 10,000 yuan ($1,400). For younger buyers especially, intelligent driving experiences have become more attractive than traditional vehicle metrics.
For years, intelligent driving remained exclusive to premium vehicles above $40,000. Entering 2025, that’s also changed dramatically. Vehicles priced above $20,000 also offer intelligent driving features. For example, China’s largest EV maker, BYD, unveiled its God’s Eye ADAS this year, offering it at no extra cost on its vehicles.
The race to dominate China’s ADAS market features two categories:
Self-developing OEMs like XPeng, Li Auto, and NIO invest heavily in proprietary systems, betting vertical integration will deliver competitive advantage. Other domestic carmakers like Geely and Changan choose to back startups to develop ADAS.
Tier-1 suppliers including Huawei, Momenta, Horizon Robotics, Zhuoyu Technology (DJI spinoff) provide turnkey solutions to automakers lacking in-house capabilities. Huawei occupies a unique position. It’s not a traditional tier-1 supplier but deeply embedded in partnerships with select OEMs through its Harmony Intelligent Mobility Alliance, engaging in nearly every aspect of vehicle development except manufacturing and final assembly.
Huawei’s Automotive Pivot
Since being sanctioned by the U.S., Huawei has lost ground in its global telecommunications and smartphone business. But over the past decade, the company has successfully forged a new path: automotive technology.
Huawei doesn’t manufacture cars like traditional OEMs. Instead, it partners with automakers through the Harmony Intelligent Mobility Alliance (HIMA), engaging deeply in nearly every aspect of vehicle development except actual manufacturing and final assembly. This model has made Huawei a strong player in ADAS.
Huawei ADS, called Qiankun乾崑, offers automakers a comprehensive solution spanning the entire ADAS stack, from sensors including homemade LiDAR, to onboard chips, AI software, and vehicle control systems.
Huawei employs multi-sensor fusion, combining LiDAR, radar, and cameras. On the software side, ADS drives just like a human with capabilities including path planning, route optimization, lane switching, navigation assistance on highways and urban roads, autonomous and valet parking.
As of 2025, Huawei offers ADS 4.0 as its latest generation, with four sub-versions. The flagship ADS 4.0 Ultra supports Level 3 highway driving when regulation permit, allowing drivers to take their hands off the wheel and divert attention from road conditions, intervening only when the system requests.
ADS mixes rule-based system, high-definition maps and neural networks, built by tens of thousands of Huawei engineers who hand-code responses for specific driving scenarios. Version 4.0 contains reportedly 20 million lines of code, a stark contrast to Tesla’s FSD, which reportedly uses only 2,000–3,000 lines after adopting a neural network architecture.
This code-intensive approach delivers exceptional performance in well-defined road conditions with predictable, safe behavior. The trade-off is reduced flexibility. The system may struggle with edge cases that haven’t been explicitly programmed. Neural network approaches, conversely, generalize better to novel scenarios. For version 4.0, Huawei is also exploring the adoption of World Action Model.
Huawei currently offers the best on-road ADAS among mass-production systems, according to multiple trusted sources. While better solutions may exist in demonstration, Huawei maintains its lead in pre-installed, customer-ready ADAS.
According to Jin Yuzhi, CEO of Huawei’s Intelligent Automotive Solution BU, 28 vehicle models now feature Qiankun ADS. Key models include Stelato S9, Maextro S800, AITO M9, and AITO M5.
Beyond the HIMA partnership, Huawei has confirmed or planned ADS 4.0 deployment across multiple automakers, including Dongfeng, Audi China, GAC Group (Trumpchi), Chery (Luxeed), VOYAH, and Changan (Deepal Automotive), among others.
Huawei’s biggest challenge comes not from an established giant but from Momenta, a rising startup gaining momentum in autonomous driving.
Momenta: Foreign Carmakers’ Favorite
Calling Momenta a startup undersells its history. Founded a decade ago by Cao Xudong, a computer vision scientist who worked at Microsoft Research Asia and SenseTime, the company was among the earliest robotaxi developers before pivoting strategically to Level 2 ADAS.
That shift has paid off. According to ZoZo Auto Research data, Momenta dominated the ADAS NOA market between January 2023 and October 2024 with a 60.1% share. Its customers include BMW, BYD, Toyota, SAIC, Mercedes‑Benz, Audi, and Nissan. As of November 2025, Momenta has secured cooperation on more than 160 vehicle models, and the number of vehicles equipped with its ADAS has exceeded 500,000.
The number (of vehicles) will reach 10 million in 2028, said Cao.
It has also gained strong backing and partnerships from major foreign carmakers include Toyota, General Motors (GM), and Mercedes‑Benz. Its latest valuation reportedly reaches $6 billion.
Momenta’s production-ready ADAS, Mpilot, targets mostly premium vehicles. The Pro version can operate without LiDAR or HD maps, relying instead on end-to-end neural networks to deliver capabilities comparable to Tesla FSD or Huawei ADS such as urban navigation, highway cruise, and parking automation.
The company is also exploring reinforcement learning in ADAS. Its R6 Flywheel model, based on reinforcement learning, is now deployed on vehicle models from GAC Toyota, Chery, and Buick.
What distinguishes Mpilot is its compatibility. While ADAS functions are becoming standardized, different automakers may use various components from their own supply chain. Mpilot as an open solution, provides extensive support for OEMs and customize products based different chips, sensors, data, and costs.
For instance, Mpilot can be built on top of both NVIDIA Orin chips and Qualcomm’s less powerful Snapdragon Ride chips. In contrast, Haomo.AI, another ADAS startup backed by Great Wall Motor, recently ceased operations due to its inability to deliver a reliable ADAS solution based on Snapdragon Ride, according to a person familiar with the matter.
Data collected from Mpilot also feeds into Momenta’s Level 4 autonomy solution, MSD (Momenta Self-Driving). This year, the company announced a strategic partnership with Uber to test Level 4 robotaxis in Munich, Germany, starting in 2026.
Despite impressive model counts, actual sales tell a more modest story. Most vehicles models averaged fewer than 2,000 monthly units in the first nine months of 2025. Unlike Huawei, Momenta lacks consumer brand awareness, a disadvantage when automakers are competing not just on technology but on marketing appeal that helps sell more cars.
Li Auto and XPeng: In-House ADAS Believers
Both Li Auto and XPeng are vertically integrated OEMs developing ADAS in-house, betting heavily on end-to-end neural networks and vision-language-action (VLA) models. But they arrive at this moment from very different positions.
Li Auto has endured a bruising 2025. The company that rose to prominence making large extended-range SUVs has struggled transitioning to pure EVs while facing fierce competition from Huawei-powered brands. Third-quarter revenue plummeted 36%, the steepest decline since its 2020 U.S. listing.
The good news is, despite being late to ADAS development, Li Auto has rapidly ascended to China’s top tier. In December 2023, its OTA (over-the-air) 5.0 software update upgraded its ADAS to AD Max 3.0, a version that pivots toward end-to-end neural network architecture from traditional rule-based systems.
Li Auto uses high-definition cameras as its primary sensors, with the Max version adding LiDAR, more cameras, and radars. For computing, entry-level AD Pro uses Horizon Robotics chips while premium AD Max use NVIDIA Orin or Thor platforms. The 2025-version AD Max V13 reportedly increased mileage per intervention (MPI) by 30%, meaning fewer hand-overs to drivers.
The company is also moving aggressively toward a new generation of ADAS technology based on VLA models. CEO Li Xiang claims VLA can help vehicles make better decision making in complex traffic scenarios and remember user preferences.
Unlike Li Auto, a latecomer to intelligent driving, XPeng has positioned itself from the outset as a smart EV and mobility technology company. Beyond manufacturing EVs, it invests heavily in autonomous driving software, proprietary AI chips, and harbors ambitions spanning robotaxis, flying cars, and robotics.
XPeng’s ADAS, branded XNGP, represents one of the industry’s most aggressive pivots toward end-to-end neural networks. While early models like the P5 featured multi-sensor fusion suites including LiDAR, newer models like the Mona—with a retail price starting at just $16,800—have adopted vision-only ADAS, replacing LiDAR with cameras and 4D millimeter-wave radars in a Tesla-like approach.
XPeng’s software stack comprises three components: XNet for visual perception, XPlanner for planning and control, and XBrain for voice assistance and cockpit AI. For computing, XPeng this year demonstrated its proprietary Turing AI chip to reduce dependence on external suppliers like NVIDIA.
Like Li Auto, XPeng is also doubling down on VLA models. Its next-generation VLA 2.0 system targets more complex driving scenarios. To accelerate AI development, the company recently announced a leadership shift: Li Liyun is stepping down as head of XPeng’s autonomous driving center, with Liu Xianming, head of the World Model division, taking over the role.
Whether these vision-first, end-to-end bets pay off will determine not just XPeng and Li Auto’s ADAS competitiveness, but their survival in China’s brutally competitive EV market.
Horizon Robotics: China’s Mobileye Bets on HSD
When Yu Kai, the extroverted, charismatic, former dean of Baidu’s deep learning research lab, founded Horizon Robotics with ambitions to build China’s Mobileye, skeptics abounded. Yu wasn’t a chip expert. How could he pull it off?
A decade later, the answer is clear. In October 2024, Horizon Robotics raised $696 million in Hong Kong’s biggest IPO of the year. The company has become a major supplier to automakers, offering low-to-mid tier chips, software stacks, and full-stack driving solutions that directly challenge Mobileye and Texas Instruments.
The backbone of Horizon’s offering is its Journey family of automotive AI chips. The Journey 6 series, launched last year and produced earlier this year, cover low, mid, and high-end use cases from 18 TOPS to 560 TOPS. So its low-end 6B chip can support basic ADAS featuring only lane-keeping and emergency braking, while high-end 6P chip can power urban navigation on autopilot.
Journey 6 uses Horizon’s proprietary BPU (Brain Processing Unit), integrated with CPU, GPU, and MCU into a highly optimized automotive SoC. This integration enables sensor fusion across cameras, radar, and LiDAR, supporting heavy AI inference for end-to-end neural network-based perception, planning, and control.
Beyond chips, Horizon offers SuperDrive (HSD), its flagship ADAS covering urban streets, highways, and parking. The company poached Su Jing, former head of Huawei’s self-driving division, to lead its ADAS development. Unlike traditional ADAS that rigidly separates perception, planning, and control, HSD employs an end-to-end neural network to output driving trajectories directly from sensor inputs.
If we can’t even handle autonomous driving, what qualifies us to make robots, said Su.
Horizon has secured partnerships with major automakers. In 2025, the company deepened cooperation with Volkswagen Group through their joint venture CARIZON to develop smart-driving solutions for the Chinese market. As of 2025, over 10 million vehicles have been equipped with Horizon’s Journey chips. HSD has also been adopted by Chery Automobile’s EXEED brand in September 2025, a step toward international exposure.
Zhuoyu Technology (DJI): The ‘DeepSeek’ in ADAS
Calling Zhuoyu Technology the DeepSeek of ADAS may be click-bait, but the similarities are obvious: both companies operated under the radar for a long period of time; both excel in engineering; and both build high-performing systems—whether AI models or ADAS—on significantly less powerful chips.
Zhuoyu’s origins trace back to DJI Automotive, the autonomous driving division of the renowned drone maker. Both ADAS and drones fundamentally rely on computer vision technology, an area where DJI has long been strong.
DJI Automotive first gained widespread recognition in 2022 when it partnered with SAIC-GM-Wuling (SGMW) to bring ADAS to a budget mini EV, Baojun KiWi EV, priced at just RMB 102,800 ($14,500). Powered by only two TI TDA4 chips with a combined 16 TOPS, the vehicle could park itself, avoid collisions, and perform functions like lane-change, ACC, stop-and-go traffic following, lane keeping, etc.
In 2020, DJI was added to the U.S. Commerce Department’s Entity List—commonly known as the economic blacklist. Meanwhile, DJI Automotive’s business diverged from DJI’s core operations. These two reasons led to the spinoff.
Spun off as an independent company in 2023 and formally branded as Zhuoyu in June 2024, the company recently raised RMB 3.6 billion ($500 million) in a strategic investment from state-owned FAW Group, which is now its largest shareholder with a 35.8% stake.
Zhuoyu democratizes intelligent driving through cost-effective solutions. The company’s core offering, the Chengxing Platform, supports ADAS across highway, urban, and parking scenarios. What sets it apart is a radical approach to computing efficiency. Zhuoyu claims its highway navigation pilot can run on as little as 32 TOPS of computing power with seven cameras. Urban navigation requires only around 100 TOPS, and reportedly costs under RMB 7,000 ($1,400).
Zhuoyu is also hedging its bets upmarket. For flagship ADAS, the company has partnered with NVIDIA to develop solutions running on Thor, enabling higher-level autonomy and neural network-based driving for premium vehicles.
I think we are among the top three (ADAS Tier-1 suppliers) in China, said Shen Shaojie, CEO of Zhuoyu.
More significantly, Zhuoyu’s solutions aren’t limited to EVs. FAW-Volkswagen deploys Zhuoyu systems in fuel-powered Volkswagen models for the Chinese market—demonstrating that intelligent driving is expanding beyond EVs
Last Takeaway
China’s ADAS market is fiercely competitive and technology evolves rapidly. Leading players can stall if they choose the wrong tech stack, while dark horses can surge ahead in just three to six months. For example, WeRide, known for its L4 robotaxis and minibuses, has recently made significant progress in ADAS through its joint development with Bosch.
One clear trend is that nearly every ADAS player is adopting data-driven, end-to-end systems, inspired by Tesla FSD’s impressive performance. Some companies have implemented two-module end-to-end architectures—one for perception, the other for planning. A select few claim to have achieved fully end-to-end systems that go directly from video inputs to vehicle controls. There’s also fierce debate over whether ADAS should be built on World Models or VLA models, with no consensus yet.
However, to handle China’s road conditions, which is far more complex and unpredictable than U.S. and European roads, incorporating rule-based guardrails into ADAS is inevitable. I don’t believe there’s any purely end-to-end ADAS deployed in China today.
Unlike FSD, which is purely vision-based, Chinese automakers favor LiDAR to provide additional safety when visibility is poor. According to the latest data, in May 2024, LiDAR penetration in over 500,000 new energy vehicles sold above RMB 150,000 ($20,000) reached 20.5%. U.S.-listed Hesai, the world’s largest LiDAR manufacturer, reports shipping 185 million units and partnering with nearly all major EV makers including BYD, Xiaomi, Li Auto, and Geely.
ADAS research and development costs are also substantial. For example, Horizon Robotics spent RMB 2 billion ($280 million) on R&D in the first half of 2024 alone, up over 60% year-over-year. As competition intensifies, only a handful of ADAS players will survive to truly challenge Tesla FSD.