During AI project delivery, the company realized that public cloud fell short on cost, control, and security for core industries, officially establishing the 'AI computing infrastructure + in-house chip' strategy.

The company formed chip architecture, server systems, and low-level software teams, initiated in-house AI chip and heterogeneous computing research, while building internal compute clusters and exploring platformized deployment.

Introduced a 'three-layer DI architecture' development roadmap: 1) DI Foundation: AI chips, compute servers, infrastructure; 2) DI Applications: training/inference platforms, industry solutions; 3) DI Services: compute operations, model customization, continuous optimization.