OEM Data Capability Research: 24 OEMs × 9 Capability Domains, Who Takes the Lead?

An L4 autonomous driving R&D and test vehicle can generate approximately 10TB of raw data per single day — data volume for just one day is enough to fill two to three hundred Blu-ray high-definition movies. For production intelligent connected vehicles, daily data volume generally ranges from GB to 1-2TB. Faced with daily TB data per vehicle, automakers are no longer challenged by “how to store data”, but “how to monetize data”. The sharp surge in data volume forces automakers to shift focus from “data collection” to “data operation”. Data is no longer merely an auxiliary tool for R&D and marketing. Similar to engines and chassis, it has become a core asset that defines automakers’ competitive edges.    
  
Based on the Data Management Capability Maturity Assessment Model (DCMM), the China OEM Data Capability Analysis Report, 2026 conducts systematic evaluation on 24 mainstream automakers from nine capability domains and 36 sub-domains. Combined with Technology Maturity Business Value Delivery (TUV) analysis, it presents a panorama of data capabilities of mainstream Chinese automakers. Evaluation results show that FAW Group obtains a composite score of 91.2, making it the sole automaker reaching L5 Optimization Level. Geely, Xpeng, NIO, Li Auto, AVATR and BYD form the leading camp. The overall landscape is characterized by “traditional giants sprinting for top positions, emerging automakers taking the lead, and joint venture enterprises feeling stressed”. Meanwhile, the report analyzes how data infrastructure and governance systems such as CDO system and data asset accounting are driving data to evolve from production factors to value-measurable assets. On compliance side, it sorts out the latest regulatory frameworks including the Guidelines for Cross-Border Automotive Data Transfer Security (2026). It summarizes 17 evolutionary trends for automotive data capabilities in the coming 2-3 years, including lake-warehouse integration and millisecond-level architecture, DaaS and AI Agent, trusted data space and cross-border data ecosystem.

Two noteworthy signals emerge. First, L3 Robust Level represents the largest group in the industry, indicating most automakers have built basic systems yet lack quantitative management capabilities. Second, a generational gap has existed between leading emerging automakers and transformed traditional automakers in terms of data architecture and data application. Take the top five players as examples. FAW Group achieves balanced performance across all nine domains without weak links. Xpeng leads in data architecture and data application. Data standards constitute a common weak spot for multiple emerging automakers.  

I. Data Awakening: From “Vehicle Manufacturing” to “Data Nurturing”, The All-Or-Nothing Move of Automakers Changes

Automakers’ competitive edge is shifting from “vehicle manufacturing capability” toward “data operation capability”. Massive in-service vehicles continuously generate real-world scenario data to feed iteration of intelligent driving algorithms, proactive after-sales services, and optimization of production and supply chains, forming a data flywheel. 

Intelligent driving data infrastructure: Nurturing data through scale. BYD builds an industry’s leading data closed loop with an ownership of over 3 million intelligent vehicles. As of May 2026, the ownership of its vehicles with assisted driving has exceeded 3.15 million units. The God's Eye platform generates more than 200 million kilometers of new assisted driving training data each day. A 5,000-person intelligent driving R&D team drives iteration of cloud world models, generative AI and end-to-end algorithms. On May 28, 2026, BYD launched XuanJi A3, a 4nm automotive-grade intelligent driving chip, establishing a positive cycle of “expanded vehicle scale-data accumulation-algorithm iteration-experience improvement higher utilization rate”. This serves as a typical model of leveraging vehicle sales volume to “nurture” intelligent driving data.    

Additionally, Xpeng’s second-generation VLA is trained on a cloud computing cluster of approximately 30,000 cards. Cumulative R&D investment exceeds RMB 2 billion. It leverages 50PB training data to enhance spatial reasoning capability. Supported by such massive data, the model enables adaptive responses to complex scenarios such as bypassing accident sites and decelerating in advance on bumpy poor-quality roads. More importantly, the full-link iteration cycle from cloud to vehicle is compressed to about five days, allowing algorithms to rapidly respond to real-world driving feedback and form an agile closed loop of “data collection-training iteration deployment and verification”.

Manufacturing and supply chain: Reshaping production with data. FAW-Volkswagen, with its “cloud” distributed automobile factory, partners with Huawei Cloud to build China’s first “cloud” distributed automobile factory in Foshan. Adopting the architecture of “Changchun headquarters dedicated cloud + branch factory CloudPond”, it realizes full data fusion and unified scheduling across remote factories and headquarters, facilitating efficient collaboration in R&D, production and supply chain. For smart factory scenarios, FAW-Volkswagen has rolled out more than 100 intelligent scenarios, generating an annual benefit of RMB 46 million.

The “industrial brain” at Seres Super Factory connects more than 300 business partners and links directly with over 50 core suppliers. It delivers a 20% improvement in R&D efficiency and a 13% rise in on-time order delivery rate. Production lines can be adjusted within 20 minutes upon receipt of data instructions.

Automated Production Site at Seres Super Factory

After-sales service: From “passive maintenance” to “data-driven proactive service”. NIO Service applies data for “proactive service”. Its Service Guardian system remotely monitors and captures abnormal signals via vehicle network, predicts vehicle faults and triggers remote warnings, transforming “passive maintenance” into “proactive service”. The system supports second-level remote diagnosis. Diagnostic experts can follow up when necessary. Its comprehensive diagnostic accuracy exceeds 90%. It has delivered proactive services to approximately 120,000 vehicles under warranty, cutting average repair duration for similar faults by roughly 30%.

Changes also take place in marketing and R&D. Changan establishes a group-wide lead center for “digital-intelligent lead operation”, also known as full-process “digital funnel” operation. It completes full-process digitalization covering lead aggregation, cleansing, rating, nurturing, distribution and reuse, lifting lead conversion rate by about 50% and generating incremental sales of 82,000 vehicles. IM Motors deploys more than 30 “digital engineer” agents for R&D. Multi-agent collaboration supports solution selection, requirement distribution and auxiliary consultation. The agent for body sheet-metal structure modeling compresses 3D design for single reinforcing plates from around three hours to minutes. For 50 reinforcing-plate projects of four doors and two covers across ten iterations, it saves 1,500 working hours.

II. Comprehensive Survey on Data Capabilities of 24 Automakers: FAW Takes Top Spot, and Emerging Players Lead - Who Falls Behind

Among 24 representative automakers, why does FAW Group achieve the top-ranked L5 Optimization Level with a clear margin? The answer lies in “institution-driven development”. In 2021, FAW issued the industry’s first enterprise-level Basic Data Act, stipulating that “data sharing constitutes an obligation, and departments generating data shall not reject reasonable cross-domain data sharing demands”. It sorts out more than 38000 business units and achieves “one-time entry, multi-point invocation”. Combined with the “1174” all-employee digital intelligent transformation strategy and Seven-Star Cloud Workbench, FAW cuts R&D cycles by 50% and reduces manufacturing costs by 40%. Besides, mainstream emerging automakers including Voyah and Seres have built self-developed or deeply customized data platform architectures. They generally adopt technical routes of lake-warehouse integration, storage-compute separation and real-time data warehouses. Their data collection- processing -application links feature low latency at millisecond-to-second level and data scale reaches PB level.

On data governance paths, automaker camps adopt distinctly different approaches. Emerging players set up explicit data or AI-related organizational departments. Different from independent data governance departments in traditional automakers, they emphasize embedded data functions and AI-native features. They elevate data from “one function or department” to the “underlying operating system of the whole organization”, and restructure organizations with AI-native, flat and data-driven middle platforms. In early 2026, Xpeng merged its Autonomous Driving Center and Intelligent Cockpit Center into the General Intelligence Center. It establishes secondary organizations around foundation models and infrastructure.  It has a unified AI middle platform supporting intelligent driving, intelligent cockpit and robot businesses simultaneously. Li Auto disbands its autonomous driving R&D department, and following AI company structure, reorganizes it into sections of Infra (infrastructure responsible for computing power and data), foundation models, software ontology and hardware ontology, and builds an AI-native middle platform similar to those of technology companies.

Traditional independent automakers realize value through scale. Take BYD as an example. As of May 28, 2026, BYD’s vehicles equipped with assisted driving systems have exceeded 3.15 million units. Its self-developed God’s Eye intelligent driving system generates more than 200 million kilometers of data on a daily basis. Real-world scenario data underpins its technical foundation, and its 5,000-member intelligent driving team translates strategies into deliverable outcomes. Tesla China is the sole foreign-funded automaker that realizes full-link domestic closed loop of “collection-storage-training-deployment-iteration”. Its Shanghai Data Center and Lingang AI Training Center form a dual-center setup, accumulating over 3 billion kilometers of Chinese road data. Constrained by global architectures and lagged local platforms, joint-venture automakers show relative backwardness in data governance.

Notably, data assets are emerging as a new dividing line. In May 2026, AVATR Technology obtained DCMM Quantitative Management Level 4 certification, a high-level certification under China’s national standard for data management. It marks that an enterprise has enabled "quantifiable, measurable, and predictable" refined operations. Leveraging this standardized control system, AVATR implements the industry’s first closed loop for automaker data asset accounting. Massive vehicle data previously deemed un-bookable and hard to value are converted into measurable and recordable assets in corporate financial statements. A six-dimensional data quality scoring mechanism eliminates cross-department silos, compressing cross-domain data retrieval time for R&D personnel from days to minutes.

By the end of 2025, automaker data assetization has advanced from “financial accounting” to “capital operation”, gaining real visibility in capital markets. BAIC BJEV secured RMB 200 million financing from Bank of Beijing by pledging data assets covering after-sales services, quality claims and power batteries, setting the record for the largest single data asset financing deal in the automotive industry. Wuhu KUMQUAT Technology under Chery completed the Yangtze River Delta’s first data asset-pledged financing with automotive after-market datasets. At exchange level, Shenzhen Data Exchange has registered 126 data assets for 32 automakers with transaction volume exceeding RMB 1.5 billion. Automotive industry data zones are launched intensively at Wuhu Yangtze River Delta Data Exchange and Western Data Exchange. BYD, SAIC and Geely rank among the first batch of automakers completing data asset-related layout.  

The capability of “data operation” begins to exert influence over automaker valuations in capital markets.

III. Policy Red Lines and Security Bottom Lines - Mandatory Compliance Courses for Automakers

2026 is an important year for data element-related policies. The Guidelines for Data Property Right Registration was issued in July, and the national registration system went live for trial operation in September. Vehicle data obtain unified national “legal identity certificates” for the first time. The National Data Administration launches the high-quality industry dataset construction action with 106 chain-leading enterprises as pilot participants, and automotive industry serves as a key vertical sector. The Industrial Data Foundation Action by the Ministry of Industry and Information Technology (MIIT) lists automotive and steel industries as priority sectors. The automotive industry consortium jointly led by China Automotive Engineering Research Institute (CAERI) and China Mobile Shanghai Branch is established in Chongqing. The Implementation Plan for Digital Transformation of Automotive Industry released by four ministries sets hard targets: the numerical control rate of key processes shall surpass 70% by 2027. 

Beyond policies lie red lines. Several Provisions on the Administration of Automotive Data Security establishes “three default principles”: default non-collection, default non-desensitization, and default non-cross-border provision of in-vehicle processed data. In February 2026, eight ministries including the MIIT jointly issued the Guidelines for Cross-Border Automotive Data Transfer Security (2026). It specifies 27 categories and 51 items of important automotive data, and lays out nine exemptions and three cross-border data transfer paths. Its definition that “remote access counts as cross-border data transfer” directly impacts global IT architectures of joint-venture automakers. Meanwhile, the automotive industry trusted data space undertaken by the CAERI covers the full research- production -supply-sales-service value chain under the “four-horizontal and four-vertical” framework, supporting scenarios including intelligent driving insurance, battery passport and carbon footprint. BAIC BJEV obtained RMB 200 million financing via data asset pledge, making data evolve from “cost” to “collateral”.     

Supervision also “takes off the brake”. In December 2025, the MIIT announced the access licenses for the first batch of L3 autonomous models, with models from Changan and ARCFOX approved. In July 2026, six cities including Beijing, Chongqing, Shanghai, Guangzhou, Shenzhen and Wuhan launched city-wide L3 on-road use pilot programs. Together with the expansion of vehicle-road-cloud integration pilots from 20 cities to 50 cities, automotive data is under proper supervision while gaining room for utilization.       

Security forms the bottom line, and compliance serves as the admission ticket. In January 2026, QAX released the Intelligent Connected Vehicle Cloud Platform Vulnerability Analysis Report. Among cloud platforms of 30 mainstream manufacturers, 28 are detected with security vulnerabilities, accounting for 93.3%. 23 manufacturers carry critical or high-risk vulnerabilities, representing 76.7%. A total of 207 vulnerabilities are discovered, among which 135 or 65.2% stem from low-level mistakes such as unverified identities and unauthorized interfaces. 22 manufacturers have vulnerabilities (73.3%) related to identity authentication and access control, and 60% suffer from failed access control. Risks including invalid digital key management and excessively exposed interfaces make cloud security a major hidden obstacle for large-scale deployment of intelligent driving. 

Automakers have rolled out countermeasures. Xpeng signs a post-quantum encryption cooperation agreement with Alibaba Cloud, and the all-new P7 is the first model to be equipped with post-quantum security technologies. NIO independently develops the NPCC end-cloud converged AI agent framework and builds a five-layer data privacy protection system. Geely releases Full-Domain Security 2.0 system and introduces quantum-security technologies into vehicle networks. BYD builds defense lines supported by the iDiDog intelligent security platform and UN R155/R156 and ISO 27001 certifications. Seres puts forward the proposition that “security represents the ultimate luxury”. Automakers start competing by treating data security as a “new luxury configuration”.

Compliance enters the “field level” era. The Guidelines for Cross-Border Automotive Data Transfer Security (2026) clarifies three cross-border data transfer scenarios. Traditional models for joint-venture automakers granting overseas headquarters database access through global IT systems must be reconstructed. “Local Chinese data centers plus domestic AI training” becomes the standard configuration for foreign-funded automakers operating in China. The Tesla model is being replicated by Volkswagen, BMW and Mercedes-Benz. Joint Venture 2.0 reconstructs data rights and obligations. SAIC-GM renews its contract until 2047 with product definition authority transferred to Chinese teams. FAW-Volkswagen restructures joint-venture patterns through China-led digital-intelligent transformation. CDO (Chief Data Officer) and data governance committees are expanding from standard configurations at leading OEMs to industry-wide requirements. 

IV. Battle for Infrastructure: From 10,000-Card Clusters to Real-Time Analysis of Trillion-Level Vehicle Network Signals

If evaluation scores serve as physical examination reports, data platforms function as blood-producing systems.

Xpeng builds its data base into a “cloud model factory”. Starting with a 10,000-card computing cluster, it has deployed a 30,000-card computing cluster for its second-generation VLA, delivering 10 EFLOPS computing power. Cluster utilization rate stays above 90% on average and reaches 98% at peak hours. R&D investment for training exceeds RMB 2 billion. It adopts 50PB training data. The full-link iteration cycle from cloud to vehicle takes roughly five days. Supported by the vehicle-cloud integrated data closed loop covering “collection, encrypted upload, cleansing, storage, labeling, training, simulation, OTA and attribution”, it horizontally connects seven business domains: vehicle, user, sales, after-sales, manufacturing, supply chain and R&D.

Li Auto achieves state-of-the-art real-time analysis. Based on Alibaba Cloud Hologres and Flink, it constructs a real-time analysis platform for trillion-level vehicle network signals with millisecond response. With unchanged overall costs, write performance improves by 200%. It supports second-level write operations for trillion-scale data and cuts computing costs by 40%. It processes real-time signals at tens of thousands per vehicle per second for millions of in-service vehicles.

Geely adopts “organization + computing power” dual drives. In 2022, it established a data governance committee chaired by CDO, forming echelons of level-one and level-two data Owners and data managers. It stands as one of the most standardized governance samples among Chinese automakers in terms of organizational setup. Its Xingrui Intelligent Computing Center 2.0 delivers 23.5 EFLOPS computing power, ranking first among Chinese automakers. The GDMP big-data platform and MirrorShip LakeHouse fusion system support connections for over 3 million vehicle network units. Geely also co-founds the Intelligent Vehicle Computing Power Alliance with enterprises including StepFun and Afari Technology.

Joint-venture automakers are catching up. SAIC-Volkswagen cooperates with Alibaba Cloud Datapilot Dataphin to build a full-domain data asset system. More than 10,000 data objects including tables, fields, indicators, dimensions and tags are sorted out. Field-level data lineage tracking runs through the full chain of research, production, supply and sales, and a standardized corporate-wide data asset inventory is completed. The implementation of Dataphin at SAIC-Volkswagen not only builds clear, credible and easy-to-use enterprise-level data asset systems, but also reshapes work paradigms for data development and governance. It acts as the core engine for SAIC-Volkswagen to realize full-life cycle management of data covering “collection-construction-governance-utilization”, laying a solid foundation for digital transformation.

Gaps in data infrastructure are shifting from “whether infrastructure exists” to “how well infrastructure performs”.

Looking ahead to future trends, three certainties deserve key attention:
① Real-time capability: Data architectures will evolve comprehensively from T+1 batch processing to second-level and even millisecond-level processing. Stream-batch integration (Flink + Paimon/Iceberg) will become the default choice for new platforms.
② Training closed loop: Competition in intelligent driving is moving from “data closed loop” toward “training closed loop”. World models and simulation enable training data to shift from passive collection to active synthesis. Data quality and simulation capability replace sheer data volume as core barriers.
③ Compliant localization: The Guidelines for Cross-Border Automotive Data Transfer Security (2026) pushes regulation down to field-level practical operations. The rule that “remote access counts as cross-border transfer” compels joint-venture automakers to complete “logical isolation + physical separation” between China regional data platforms and global systems. 

In the future, judging an automaker’s data quality will not need to merely focus on the technical terms it releases. Instead, four things will suffice: whether CDO and data governance committees are established within organizations; whether platforms are frequently adopted in actual business scenarios; whether data completes right confirmation and accounting as assets; whether data capabilities translate into verifiable revenue or efficiency indicators. Simply put, data capabilities are a watershed for the next-round industry reshuffle.