国产大模型告别百模大战，头部领先优势明显，摩根大通：首次覆盖智谱与MiniMax

Core Insights - The competitive landscape of China's AI market has shifted from over 200 developers to less than 10, creating a dual-track competition between tech giants and independent pioneers [1] - JPMorgan has initiated coverage on two key players, Zhipu and MiniMax, both of which are positioned to capitalize on the next wave of global AI value creation [1] Industry Changes - The transformation in the AI industry is characterized by a shift from parameter scale to model performance, commercialization efficiency, and global adaptability as key survival factors [1] - The current competition is marked by tech giants like Tencent and Alibaba leveraging scale and data advantages, while independent firms like Zhipu and MiniMax focus on technological agility and innovation [1] Zhipu: Industry Benchmark Driven by Technology - Founded in 2019, Zhipu has established a technological barrier centered around its GLM series models, focusing on an enterprise-level MaaS (Model as a Service) model [3] - By the first half of 2025, 85% of Zhipu's revenue is expected to come from localized deployments, with a gross margin of 59.1% [3] - Zhipu's financial forecast indicates a compound annual growth rate (CAGR) of 127% from 2026 to 2030, with profitability expected by 2029 and a target price of HKD 400, representing a potential upside of 44% [3] MiniMax: Builder of a Global Multimodal AI Ecosystem - Established in late 2021, MiniMax has a "born global" strategy, creating a comprehensive multimodal model covering text, voice, video, and music [4] - By the first three quarters of 2025, overseas revenue is projected to account for 73.1% of MiniMax's total, with significant contributions from the U.S. and Singapore [4] - MiniMax's financial outlook suggests a CAGR of 138% from 2026 to 2030, with profitability anticipated by 2029 and a target price of HKD 700, indicating a potential upside of 36% [5] Competitive Dimensions: Technology, Commercialization, and Cost Structure - Zhipu focuses on text models with iterative upgrades, while MiniMax has a comprehensive multimodal approach [6] - Zhipu's revenue structure is heavily weighted towards domestic markets (88%), whereas MiniMax has a balanced international presence [6] - Both companies rely on leased computing power, with a significant shift in cost structure expected post-2025, as training costs decrease and inference costs increase [6] API Business as a Core Growth Engine - The profitability of the API business is influenced by unit inference pricing, efficiency, GPU utilization, and regional structure, with MiniMax achieving a higher gross margin of 69.4% in its API business compared to Zhipu [7] - The rise of Zhipu and MiniMax signifies a transformative moment in the Chinese AI market, showcasing the innovative capabilities of Chinese enterprises on a global scale [7]