CITIC Securities analysis report: Domestic AI models accelerating iteration (GLM-5.2, Kimi K2.7, etc.) driving expansion of enterprise workflow and Agentic Coding demand; AI cloud vendors' backlogs continue to grow, GPU rental prices rising, CPU/DRAM/NAND supply constrained, computing power rental services expected to maintain favorable momentum.

Domestic AI model development continues accelerating, with CITIC analysts noting that Chinese models including GLM-5.2 and Kimi K2.7 Code are enhancing capabilities in extended context windows (1M tokens), long-horizon agents, and Agentic Coding. These advances reflect a strategic shift by domestic players away from general-purpose question-answering toward developer tools and enterprise-level workflows. Integration of AI into WeChat's ecosystem—currently in beta testing—signals a broader move from standalone applications to super-app platforms, which analysts expect will increase model invocation frequency and inference token consumption across the market.

Internationally, Anthropic continues expanding its footprint with growing revenue, an expanding customer base, and increasing fundraising rounds. Enterprise workflow and Agentic Coding remain in their expansion phase globally, with sustained demand signals. Domestically, competitors including Kimi are strengthening international operations while DeepSeek's fundraising activities reinforce expectations around commercialization of leading-edge models in China.

On the infrastructure side, pressures are mounting across compute and memory resources. AI cloud vendors report growing backlogs and rising GPU rental rates, while broader supply constraints in CPUs, DRAM, and NAND are widening. These dynamics underscore sustained demand for compute capacity and rental services, with CITIC analysts projecting the current prosperity in AI infrastructure services to persist.