EN / 中文

Can Chinese EV Makers’ In-House Chips Compete in the Open Semiconductor Market?

by bandaotichanyezongheng·November 13, 2026

Author: Xiao Meng

In 2026, chip manufacturing has become a key keyword in the intelligent competition among automakers.

From NIO's Shenji and XPeng's Turing to Li Auto's Mach M100, automakers have begun launching self-developed chips and integrating them into mass-produced models. The self-developed chip businesses of some companies are also starting to extend outward, with goals no longer limited to meeting the needs of their own vehicle models.

Recently, Shanghai Mach Xin Technology Co., Ltd., a chip subsidiary under Li Auto, completed its first round of equity financing worth CNY 1.625 billion. In addition to continuing investment in chip and domain controller R&D, Li Auto explicitly proposed pushing related products toward external customers.

From self-use to attempting external sales, why are automakers turning their attention to the external market for chips? Can these self-developed chips become a new force in automotive semiconductor supply?

01 Why Are Automakers Starting to Develop Their Own Chips?

As the saying goes, different industries are like mountains apart. For automobile enterprises whose main business is vehicle R&D and manufacturing, chip design involves multiple specialized processes such as architecture development, tape-out, and test verification, which differ significantly from traditional vehicle R&D. Why are automakers rushing into the field of chip design?

First, the development of smart cars places higher demands on computing power. As intelligent driving evolves from traditional driver assistance to advanced intelligent driving, vehicles need to process more sensor data from cameras, LiDARs, etc., and run larger-scale AI models, placing higher requirements on chip computing power, memory bandwidth, and data processing efficiency. At the same time, vehicle intelligence is shifting from single functions to coordinated whole-vehicle systems, making chips an increasingly critical component affecting intelligent driving performance and overall smart vehicle experience.

For automakers, directly purchasing general-purpose chips on the market can meet certain computing needs, but such products must accommodate various customer application scenarios, making it difficult to deeply optimize chip architecture, computing unit configuration, and data processing methods for a specific automaker's algorithms. Especially in intelligent driving systems, differences exist among companies in algorithm models and sensor configurations, leading to varying requirements for chip resource allocation, memory bandwidth, and data transmission efficiency. By designing chips independently, automakers can adjust hardware architectures based on their own algorithms and actual application scenarios, allowing chips to better adapt to software systems.

Second, software-hardware synergy helps improve system development efficiency. Intelligent driving involves multiple links including chips, algorithms, operating systems, and sensors. If each part is provided by different suppliers, extensive adaptation and optimization are often required between systems. By participating in chip design, automakers can integrate hardware architecture with their own algorithm development, considering model execution, operator scheduling, and data transmission needs during the chip design phase, thereby reducing software-hardware adaptation work and improving computing resource utilization.

Furthermore, cost is also an important factor for automakers considering self-developed chips. As intelligent driving features become widespread across more models, the cost of onboard computing platforms increases accordingly. Long-term reliance on external supplier procurement forms continuous expenditure. NIO CEO William Li previously revealed that NIO's annual procurement cost for NVIDIA Orin X chips reached USD 300 million at peak business periods. For automakers with high sales volumes, if they can meet their own model needs through independent R&D, they have the opportunity to amortize initial R&D investments over subsequent large-scale installations, lowering long-term chip procurement costs.

Finally, mastering the design capability of key chips relates to automakers' initiative over supply chains and technology iteration. The update speed of automotive products is accelerating, requiring continuous upgrades of intelligent driving algorithms and vehicle functions. If core computing platforms rely entirely on external suppliers, the iteration cycle, supply price, and capacity arrangements of chip products will affect automakers' own R&D plans. By participating in chip design, automakers can arrange chip iterations according to vehicle platform development plans and gain more autonomy in product definition, technical roadmaps, and supply chain management.

However, although the momentum of self-developed chips is intensifying, automakers have not completely abandoned external procurement in the chip sector. Chip R&D requires high capital investment and long development cycles, and after design completion, it still involves tape-out, verification, software adaptation, and mass production. For automakers, self-developed chips require comprehensive consideration of sales scale, technical accumulation, and actual application needs; therefore, the coexistence of self-development and external procurement remains prevalent in the current automotive industry.

02 What Different Routes Do Automakers Take in Self-Developed Chips?

Different automakers have varying technical accumulations and product needs, resulting in distinct characteristics in their self-developed chip layouts.

Intelligent driving computing chips are currently one of the most popular directions for automakers' self-development.

Li Auto's Mach M100 adopts a 5nm automotive-grade process, achieving a single-chip computing power of 1280 TOPS and utilizing an independently developed dynamic dataflow architecture. This architecture emphasizes coordination between data transfer and computing tasks, triggering corresponding calculations upon data readiness to reduce overhead caused by centralized scheduling and data movement.

In terms of chip structure, the Mach M100's NPU contains 56 computing units, the CPU part adopts a 24-core Arm Cortex-A78AE architecture, and it is equipped with an 8-channel LPDDR5X memory subsystem with a peak bandwidth of 273GB/s. Li Auto hopes this architectural design will better adapt to the parallel computing needs of in-vehicle AI models.

NIO's Shenji NX9031 adopts a 5nm automotive-grade process, featuring over 50 billion transistors, a 32-core CPU architecture, and integrated high-performance ISP and other computing units, with pixel processing capabilities reaching 6.5 G Pixel/s and processing latency less than 5ms.

The ISP is primarily responsible for processing image data collected by cameras, providing a data foundation for subsequent object recognition and environmental perception through noise reduction, image enhancement, and other processing. NIO integrates high-performance computing and image processing capabilities into a single chip to meet the requirements of intelligent driving systems for computing performance and image processing efficiency.

XPeng's Turing AI chip focuses more on large model computing capabilities and multi-scenario applications. The chip adopts a 40-core processor architecture, integrates two self-developed NPUs, and introduces a dedicated Domain-Specific Architecture (DSA) optimized for large model inference. XPeng positions it as a multi-terminal general-purpose AI chip, covering applications in smart cars, robots, and flying cars.

Inside vehicles, the Turing chip can undertake intelligent driving computing tasks, and some high-end models are equipped with independent Turing AI chips for cockpit large model-related computations. According to XPeng's 2026 model configuration data, the effective computing power of the Turing AI chip on certain models is 750 TOPS, and high-spec versions can further enhance computing capabilities through multi-chip collaboration.

BYD has also launched self-developed products in the field of intelligent driving computing. In May this year, BYD released the 4nm automotive-grade intelligent driving chip Xuanji A3, supporting L3 and L4 autonomous driving, incorporating triple-core NPUs, and combining self-developed algorithms for computing optimization. According to BYD's published data, when three chips work collaboratively, the total computing power exceeds 2100 TOPS, and the power consumption per unit of computing power is reduced by 20% compared to similar solutions. BYD also stated that through the synergy between chips and self-developed algorithms, computing resource utilization can increase by 100%, reflecting its approach of deeply integrating self-developed chips with its own intelligent driving system.

Power semiconductors are also an important direction for automakers' self-development layout.

BYD is one of the earlier domestic complete vehicle enterprises to enter the automotive-grade power semiconductor field. It began laying out IGBT R&D in 2005, gradually establishing industrial chain capabilities covering chip design, wafer manufacturing, module packaging, and testing. In the IGBT field, BYD has launched multiple generations of self-developed products. Among them, IGBT 4.0 was released in 2018, adopting fine-tuned planar gate design, reducing comprehensive loss by approximately 20% compared to mainstream market products under the same conditions, while improving current output capability and device reliability. Subsequently, BYD launched IGBT 5.0, adopting micro-trench structures and composite field-stop technology to further reduce conduction and switching losses and improve device turn-off characteristics.

For plug-in hybrid electric vehicles, BYD also developed dedicated DM4.0 IGBT integrated modules. These modules carry self-developed high-performance IGBT chips, integrating functions such as drive, generation, inversion, and boost/buck, consolidating originally dispersed power control links and applying them to the DM-i hybrid system. Data published by BYD Semiconductor shows that these modules have been mass-produced since 2021 and have been applied to BYD DM-i models.

Beyond IGBTs, BYD started relatively early in promoting the R&D and mass production of silicon carbide (SiC) power devices. In 2020, it launched self-developed automotive-grade SiC three-phase full-bridge power modules, applied to the main drive motor controllers of models like the BYD Han. These modules mainly handle energy conversion tasks in main drive inverters, converting DC electricity from power batteries into AC electricity required for motor drive. BYD's initially launched 1200V, 840A three-phase full-bridge SiC modules adopt packaging technologies such as nano-silver sintering and AMB ceramic substrates, with module output power reaching 200kW. BYD also stated that the copper clip interconnection process used enhances the chip's overcurrent capability, doubling power density compared to previous solutions and reducing volume by about half.

In 2025, BYD launched the world's first mass-produced 1500V high-voltage resistant, high-power SiC chip, applied to its Super e Platform. This platform adopts a full-domain kilovolt-level high-voltage architecture, upgrading battery, motor, power supply, and air conditioning systems to the 1000V voltage level, placing higher demands on the voltage resistance and energy conversion efficiency of power devices. The 1500V SiC chip can adapt to this high-voltage system, undertaking power control tasks in high-voltage energy conversion and motor drive, providing power device support for the Super e Platform to achieve high-power output and megawatt flash charging.

Automotive control chips such as MCUs are also important entry points for automakers' self-development.

Hunan Zijing Semiconductor Co., Ltd., incubated by Great Wall Motor, lists automotive-grade MCUs as a key R&D direction. Its first product, Zijing M100, is a RISC-V automotive-grade MCU led by Great Wall Motor in joint R&D with multiple parties, adopting open-source RISC-V kernel design, achieving mass production and installation in December 2025, and mainly applied to body control links such as combination headlights, ambient lights, and combination switches.

Dongfeng Motor focuses its self-development on high-performance automotive-grade MCUs. The DF30, developed under the leadership of its R&D General Institute, adopts an autonomous open-source RISC-V multi-core architecture, based on domestic 40nm automotive process, with a functional safety level reaching ASIL-D. This chip mainly targets key automotive control scenarios such as engine control and has undergone verification on models like Dongfeng Yipai 007, Mengshi M817, and Dongfeng Fengshen Haohan. Disclosed by Dongfeng in 2026, the DF30 automotive MCU and high-side driver chips have achieved mass production installation, further expanding the application of self-developed chips in vehicle control systems.

03 Commercialization Layouts Vary, External Market Extends Beyond Other Automakers

Although all are manufacturing chips, the commercialization layouts of each automaker's self-developed chips vary, and target customers are gradually extending from other complete vehicle enterprises to fields such as robotics, unmanned logistics, and edge AI computing.

NIO's Shenji has already begun generating external business revenue. In April 2026, William Li revealed in an interview that many automotive industry clients, Robotaxi companies, unmanned logistics companies, and embodied intelligence companies are contacting Shenji, hoping to cooperate. In October, Li further stated that Shenji is expected to generate hundreds of millions of yuan in external revenue this year, and related businesses have already started generating actual income.

XPeng's Turing chip has secured a mass production designation from Volkswagen Group, initiating external commercial cooperation. Besides other automakers, XPeng is actively expanding customers such as embodied intelligence companies and Tier 1 suppliers. In October 2026, XPeng further launched the Smart Vehicle Solution Business XPENG AI ENGINEERING, planning to open up vehicle platforms, E/E architectures, intelligent driving technology, AI chips, and related toolchains to global partners, expanding the external market through combined output of chips, software, and vehicle technology.

Li Auto's current measures focus mainly on Mach Xin Technology's financing and subsequent commercialization layout. Related funds will be used for chip and domain controller R&D, and to push products toward external supply. However, Mach Xin Technology has not yet generated operational revenue, and related businesses are still in the stage of customer expansion and commercial preparation. Besides intelligent computing chips, Li Auto is also planning external supply business for silicon carbide modules, with the chip business expansion scope including power semiconductors.

Looking at the entire power semiconductor market, IGBTs and SiCs have broad application space beyond automobiles. IGBT modules can be used in energy conversion links such as industrial motor drives, PV and wind power inverters, and EV charging equipment. SiC modules have already been applied in fields such as PV inverters, fast charging equipment, energy storage systems, industrial drives, and data center power supplies. For enterprises with power device R&D and production capabilities, potential external supply opportunities exist in these industries.

04 Where Is the Appeal of Automakers' Self-Developed Chips to External Customers?

First, some chips possess performance advantages in specific computing tasks. Taking NIO's Shenji NX9031 as an example, its high bandwidth and ISP image processing capabilities can support the operation of world models and large-parameter AI models. For Robotaxi, unmanned logistics, and embodied intelligence companies, such hardware capabilities can meet computing needs such as visual data processing and AI model inference, providing them with new chip options.

Second, some automakers can provide software and hardware technical solutions beyond chips. For external customers, purchasing a chip is only part of product development; they also need to complete underlying software adaptation, hardware design, and system integration. Automakers with vehicle platform R&D experience can combine chips with relevant software, toolchains, and system solutions to help customers reduce some underlying development work and accelerate product landing. XPeng opening up vehicle platforms, E/E architectures, and intelligent driving technology to partners also reflects this combined output approach.

Additionally, chip technology licensing and joint development provide another cooperation mode for external enterprises. NIO's cooperation with Axera Semiconductor to develop the M97 adopted the core modules and architecture of Shenji NX9031, with both parties jointly developing new chip products. For enterprises with chip R&D capabilities, such cooperation can utilize existing core design resources to accelerate new product development and adjust product specifications according to target markets.

05 What Thresholds Must Automakers' Self-Developed Chips Cross to Enter the External Market?

For other automakers, competitive relationships and technical trust may affect willingness to cooperate.

Automakers' self-developed chips are usually deeply integrated with their own algorithms, software, and vehicle platforms. During chip adaptation, both parties may need to exchange technical information such as algorithm characteristics, sensor configurations, and computing resource allocation. If the chip developer simultaneously operates an automotive brand, potential buyers may consider whether core data and technical information can be effectively protected, and whether the supplier can maintain a relatively independent commercial stance.

Previously, regarding the external supply issue of Li Auto's Mach chips, industry experts mentioned that some automakers might worry about the competitive relationship with Li Auto. Whether other companies are willing to build their intelligent systems on top of a chip company controlled by a competitor involves factors such as data security, product roadmap, supply stability, and commercial trust. This is a special issue automakers' self-developed chips face when selling to other vehicle manufacturers.

For non-automotive customers such as robots and embodied intelligence, technical adaptation may become a more direct consideration.

Differences exist between automobiles and robots in sensor configurations, algorithm models, and software architectures. Even if automotive-grade chips possess strong AI computing capabilities, they need software adaptation completed according to specific applications. Taking Li Auto's Mach M100 as an example, reports indicate that embodied intelligence companies have contacted Li Auto; whether they ultimately adopt this chip may depend on the difficulty of algorithm migration. For these customers, whether the chip can run their own algorithms and how much software adaptation and system development resources are needed will influence procurement decisions.

Professional chip manufacturers already occupy significant market share and have established mature product and customer ecosystems.

For automakers' self-developed chips, there is a group of professional suppliers who have been deeply cultivating the market for years in the external market. They have accumulated considerable time in areas such as chip architecture iteration, software toolchains, mass production verification, and customer service, forming high market competition barriers.

Taking Horizon Robotics and NVIDIA as examples, statistics show that in September 2026, NVIDIA and Horizon Robotics held shares of 28.5% and 20.3% respectively in the domestic ADAS domain controller chip installation market, ranking first and second. Together, these two manufacturers occupy nearly half of the market share and already possess a large customer base and mass production scale.

Horizon Robotics' advantage lies not only in shipment volume. In September 2026, its Journey series intelligent driving chips exceeded 15 million sets in cumulative mass production shipments, established cooperation with over 40 automotive brands, and obtained designations for 500 mass production models. The accompanying Tiangong Kaiwu toolchain covers development links such as model quantization, compilation optimization, simulation, deployment, and debugging, cumulatively empowering over 200 ecosystem partners. After multiple generations of product iteration and validation in numerous model projects, Horizon Robotics has formed a complete system from chip hardware to development tools, reference designs, and mass production services.

In the robotics field, professional suppliers have also established corresponding development ecosystems. Taking NVIDIA as an example, its Jetson computing platform, combined with the Isaac robot development platform, provides simulation frameworks, robot learning tools, CUDA acceleration libraries, AI models, and reference workflows, covering links such as robot development, training, simulation, and deployment. For external developers, such mature platforms can provide a relatively complete development environment, reducing the work of building computing platforms and software systems from the ground up.

In comparison, although automakers' self-developed chips can form certain technical advantages relying on their own product and algorithm accumulation, they still need to establish cross-customer product adaptation capabilities and development ecosystems when facing external customers. The market share, mass production records, and customer cooperation relationships accumulated by professional chip manufacturers will also increase the difficulty for new entrants to obtain orders.

External customer onboarding cycles are long, and shipment scale must support continuous R&D.

Chips entering the automotive supply chain typically undergo lengthy verification and onboarding processes. Audit documents state that automotive chips are closely related to driving safety, and downstream component manufacturers and vehicle enterprises have high requirements for chip reliability; introducing new products requires investing significant manpower, time, and testing costs. If the comprehensive benefits brought by replacing new chips cannot cover verification investments, customers have little willingness to proactively switch original suppliers.

The document also mentions that automotive customers tend to maintain relatively streamlined supplier systems and do not easily introduce multiple suppliers for the same product line. For newly introduced chips, even if they pass preliminary verification, customers often arrange small-batch, non-key model production first, and gradually expand application scope after actual operation and reliability performance checks. This means new suppliers often need to experience a long ramp-up period from obtaining designations to forming scaled shipments.

For chip R&D enterprises, initial architecture design, tape-out, and verification require continuous investment, followed by costs for next-generation product R&D, software tool updates, and customer service. If the number of external customers is limited and the shipment scale of individual projects is small, it is difficult to form a revenue and order base sufficient to support continuous R&D.

06 Conclusion

In summary, automakers' self-developed chips can indeed open up a certain amount of external business space, with customer ranges not limited to other automakers; potential application opportunities exist in fields such as Robotaxi, unmanned logistics, robotics, industry, and energy. However, moving from internal matching to external supply requires chips to possess stronger product versatility, software adaptation capabilities, and customer service capabilities.

For most automakers that have tried or are preparing to try self-developed chips, self-developed chips will still primarily serve their own vehicle needs in the short term; external cooperation can increase product application scenarios and revenue sources, but it is difficult to universally develop into independent, continuously profitable chip businesses. Only a few enterprises with large internal installation scales, sustained R&D capabilities, and the ability to provide standardized products and long-term technical support to different customers are more likely to develop self-developed chips into stable external businesses.