China's three state-owned automakers – FAW, Changan and Dongfeng – are doubling down on self-developed intelligent driving systems, a strategy that could face its toughest test in a market shakeout expected around 2028, according to a report by Chinese tech outlet Huxiu.
The push follows a wave of price hikes for third-party ADAS packages. Huawei's ADS premium feature bundle returned to a 36,000 yuan (about $5,070) one-time fee on July 1, BYD's 'God's Eye B' rose from 9,900 to 12,000 yuan, Xpeng's XNGP went up 2,000 yuan, and Li Auto's urban NOA (city advanced driver assistance) climbed 2,500 yuan. Automotive-grade memory chip prices have surged as much as 180% in the short term.
But the core question, Huxiu argues, is not whether the automakers can afford to buy these systems, but whether they can afford to build their own. The answer hinges on sales volume – and the math is not straightforward.
The cost of control
In September, Changan executive vice president Yang Dayong said the brand's vehicles would be equipped with – and only with – its self-developed Tian Shu Linghang ADAS system, with no plans to bring in other suppliers. Changan vice president He Gang revealed that the broader Beidou Tianshu system, which includes the ADAS, has received total investment exceeding 60 billion yuan (about $8.45 billion) and employs 7,500 people in R&D.
However, Huxiu notes that this figure covers intelligent products, management and manufacturing – not just driver assistance. The dedicated ADAS team is estimated at 1,500 to 2,500 people, with annual operating costs of 5 to 10 billion yuan (about $704 million to $1.41 billion).
A sensitivity analysis in the report shows that at an annual fixed investment of 5 billion yuan and a per-vehicle third-party cost of 8,000 yuan, the breakeven point is about 625,000 vehicles. At 8 billion yuan fixed investment and 5,000 yuan per-vehicle cost, it jumps to 1.6 million vehicles.
The three automakers sold a combined 3.659 million vehicles in the first half of 2026 – FAW 1.518 million, Changan 1.119 million and Dongfeng 1.022 million. But they do not share R&D teams, data pools, cloud platforms or vehicle architectures, so the combined volume cannot be used to amortize a single system.
Data is the real bottleneck
Changan's He Gang pointed out that many rivals cannot access their own data because it belongs to suppliers. He said Changan's AI-assisted labeling has exceeded 50%, but that does not equal a data asset.
Changan's chief ADAS technology officer Tao Ji said the biggest challenge is data quality and consistency. 'Different people drive very differently in the same scenario, causing a lot of causal confusion in the data,' he said. 'We need a lot of data engineering in the cloud to make the data understandable to the model.'
Huxiu contrasts this with Huawei's ADS5, which had accumulated 1 billion kilometers of data by 2025. Dongfeng has released 1.25 million sets of end-to-end autonomous driving open-source data from L4 demonstrations in 15 provinces and 35 cities, but the report notes that metrics like total mileage are not directly comparable.
The stakes are high. He Gang predicts the ADAS industry will enter a 'Spring and Autumn period' of consolidation, with only five or six competitive solution providers left in China by around 2028. Yang Dayong sees a similar shakeout among automakers, with only 20-odd active brands out of 120 surviving.
For the three state-owned giants, the clock is ticking. With less than two years until the predicted ADAS shakeout, they must convert their sales volume into algorithmic capability – or risk being left behind.
