On May 26, news emerged that Shenzhen has unveiled the outline of its '15th Five-Year Plan', advocating for a synchronized progression in the research, development, and production of intelligent computing chips, the erection of computing power infrastructure, and the advancement of model algorithms, all aimed at laying a robust foundation for AI technology. The plan is designed to fortify the industrial bedrock for computing power chips, propelling the refinement, iteration, and application-specific customization of training chips such as Ascend, as well as edge-side inference chips, thereby nurturing a vibrant domestic AI computing power ecosystem.

Concurrently, it underscores the importance of reinforcing a multi-layered intelligent computing power foundation, constructing high-caliber training infrastructure, and establishing a sustainable computing power supply system that embodies ultra-intelligent collaboration, heterogeneous integration, and universal inclusivity. The plan further urges the enhancement of fundamental AI theoretical research and innovation in model architecture, fostering the development of internationally recognized domestic general-purpose large models and industry-specific application models, and facilitating the deployment of model algorithms through high-value application scenarios.

By 2030, the city anticipates that its real-time available computing power will surpass 150 EFLOPS, with select metrics for domestic chips and computing clusters attaining international cutting-edge standards, thereby laying a solid large model foundation for widespread application.