China’s Ministry of Industry and Information Technology has put its mandatory national standard for L3 and L4 autonomous driving out for public comment. The draft, Safety Requirements for Intelligent and Connected Vehicle Automated Driving Systems, would replace the old voluntary standard and for the first time set a legally binding floor for L3 and L4. Products that fail it cannot be built or sold, and breaching the rule is a legal violation. It is slated to take effect on 1 July 2027.
The move from “recommended” to “mandatory” marks a full turn in the industry’s winds. Autonomous driving now enters a compulsory regulatory cycle. Several regions have already begun issuing L3 road-test and demonstration licences, and compliant mass production is on a countdown.
The competitive logic has changed. Makers used to race on how fast they could ship features. Now the game is lifetime safety, stability and iterability. Architecture choice decides who stays at the table. With several carmakers already fitting high-compute chips, one signal is clear: large-compute architectures are inevitable for L3 and L4.
Horizon Robotics founder Yu Kai laid out a timeline: 100 per cent hands-off driving by 2028, L4 eyes-closed driving by 2030, and sleeping-through-it by 2035. XPeng chief He Xiaopeng is more aggressive, forecasting L4 or even L5 within three to five years.
On hardware, 2000 TOPS is the entry ticket for AI-native L3, according to Desay SV principal engineer Peng Xueming. From L3 to L4, compute demand roughly doubles. Li Auto’s L9 Livis carries two self-developed Mach M100 chips at 2560 TOPS; XPeng’s GX tops out at 3000 TOPS across four Turing chips. None are yet L3-capable, but every one is pre-positioned for it.
For European supply chains, the takeaway is direct: a mandatory L3 and L4 regime in China accelerates the compute, sensor and software stack that EU OEMs will have to match to stay competitive.
Translated and adapted from CheDongXi (chedongxi.com).