KINGSOFT CLOUD (03896) fully upgrades its AI cloud infrastructure, accelerating the realization of AI value.

Currently, the industry has entered a new stage of scaling up Agent deployment. According to the latest data released by Sullivan and the Head Leopard Research Institute, it is estimated that the market size of AI Agents in China will increase from 2.6 trillion yuan in 2025 to 20.1 trillion yuan in 2030, with a compound annual growth rate of 50.4%. To seize this opportunity, KINGSOFT CLOUD (03896) is continuously refining its product technology and engineering capabilities around model services and intelligent agent applications. The KINGSOFT CLOUD Star Source AgentKit platform provides a one-stop development and governance foundation for enterprise-grade AI applications, covering the entire process from Agent development, tool integration, deployment and operation, to performance evaluation and security governance. The platform centers on Agents and integrates capabilities such as Agent Engine runtime, security sandbox, MCP tool square, Skills center, knowledge base, memory library, and Agent evaluation system, supporting enterprises in quickly building executable, reusable, and manageable intelligent agent applications. The KINGSOFT CLOUD Star Stream Model API service has upgraded its multi-model service and enterprise-level access capabilities, introducing Response API, online model debugging, backend model switching at access points, and refined API Key permission management to achieve high protocol compatibility. Through high-concurrency inference optimization, the core model throughput capacity has increased by approximately 150%, significantly reducing the costs for enterprises related to multi-model access and migration, while enabling fine-grained management across permissions, usage, and model dimensions. The KINGSOFT CLOUD Star Stream Training and Inference platform has completed upgrades in resource allocation, scheduling, training, and model fine-tuning capabilities, adding features such as physical queues, resource borrowing and returning, and flexible configuration of computing resources, enhancing resource scheduling flexibility in multi-task concurrent scenarios, assisting algorithm and R&D teams in efficiently completing resource allocation and task operations, and lowering the threshold for training environments and resource scheduling use. Currently, the KINGSOFT CLOUD Star Stream platform's MaaS services have been accessed by multiple commercial Internet clients, providing high-concurrency, low-latency model services.