财通证券：模型密集发布 国产算力需求有望加速

Group 1 - The core viewpoint of the report indicates that the domestic model iteration is accelerating, leading to a significant increase in inference-side demand, which will benefit the domestic computing power industry chain [1] - Major domestic models are entering a dense release period, with significant models such as DeepSeek's open-source OCR2, Kimi's K2.5, Alibaba's Qwen3-MaxThinking, and Baidu's Wenxin 5.0 being launched [1] - ByteDance plans to introduce three new AI models in February, including Doubao 2.0, Seedream 5.0, and SeedDance 2.0, while Alibaba is also set to release its next-generation flagship AI model Qwen 3.5 during the Spring Festival [1] Group 2 - Domestic cloud vendors are maintaining high levels of capital expenditure, with ByteDance planning a capital expenditure of 160 billion yuan for 2026, up from approximately 150 billion yuan in 2025 [2] - Alibaba announced a three-year plan for AI infrastructure construction with a budget of 380 billion yuan during the 2025 Cloud Habitat Conference, indicating sustained investment in cloud computing [2] Group 3 - The rapid iteration of domestic models is expected to capture user AI interaction entry points, leading to a corresponding increase in inference-side computing power demand [3] - 2026 is projected to be a pivotal year for the deployment of domestic supernodes, with numerous domestic manufacturers releasing next-generation supernode solutions, including Huawei's Atlas 950/960 and others [3] - The industry is poised for a significant growth phase as both supply and demand sides are actively engaged in the market [3]