The image generated by AIYXL using AI
When automakers use AI, they often need to solve a series of concrete problems: how design proposals enter subsequent engineering workflows, how quality inspection models adapt to different production lines, and where the data and compute needed for training come from.
On October 23, at the 2026 World Intelligent Connected Vehicles Conference, the National Artificial Intelligence Application Pilot Base (Automotive Direction, Manufacturing Sector), led by China Mobile, released an automotive industry AI foundation and its construction results. According to Cailian Press, citing Capital Construction News, the capabilities are open and shared to the entire industry starting today, and enterprises can use them on demand.
Four Major Domains, Six Types of Agents
The disclosed capabilities cover four major domains: R&D, manufacturing, intelligent driving, and testing and evaluation, forming six types of agents: automotive styling design, process planning, visual quality inspection, intelligent operations and maintenance, intelligent driving data generation, and intelligent driving model training.
This list reflects how an industry foundation is organized: the bottom layer integrates “compute—data—models,” while the upper layer provides capability access points around enterprise tasks. For example, visual quality inspection corresponds to recognition needs on the production floor, process planning connects to manufacturing preparation, and intelligent driving data generation and model training serve the development process.
According to the report, the foundation has brought 112.7 EFLOPS of compute under management, and related results have been deployed at more than 10 central and state-owned automakers. However, the report did not specify the computing precision, test conditions, or specific application results used in the compute statistics, so it is not yet possible to compare processing speed for individual tasks or enterprise benefits on this basis.
After They Can Be Called, Integration into Business Is the Next Test
From an engineering application perspective, shared capabilities offer an alternative to building in-house. Enterprises can first carry out validation around specific tasks, then decide whether to expand the scope of use.
Actual evaluation also needs to focus on the tasks themselves: quality inspection focuses on missed detections, false positives, and production line takt time; process planning focuses on whether a plan can meet equipment and process constraints; and design tools need to be examined on whether outputs can continue to be modified and enter engineering workflows. These are evaluation dimensions, not results already announced in this announcement.
Therefore, the follow-up information worth watching about this foundation is how various capabilities are integrated, what data they require, how they are billed, and whether reviewable case metrics can be provided. Opening resources gives enterprises a chance to start experimenting; stable delivery will determine how far it can go in production processes.