Jin Rong Jie Zi Xun·2025-12-18 08:50Core Insights - JD Cloud ranks among the top three in product capability in the "2024 China AI Infra Platform Computing Power Management Layer Market Vendor Competitiveness Quadrant Analysis" due to its technological innovations in heterogeneous computing power scheduling and GPU pooling management [1][3] Market Overview - The AI Infra platform has established a three-layer capability system: computing power management layer, model management layer, and application management layer, with the computing power management layer dominating the market, expected to reach a market share of 64.6% in 2024 [3] JD Cloud's Strategy - JD Cloud focuses on "heterogeneous computing power + extreme inference" to build a new generation of AI infrastructure, leveraging JD Group's diverse business scenarios [3] - The core platform, JoyScale, optimizes hardware and software collaboration to support high-performance heterogeneous computing with domestic chips like Huawei Ascend, Cambricon, and Haiguang [3] JoyScale AI Computing Power Platform - JoyScale AI computing power platform has four core advantages: - Extreme Computing Performance: Features an industry-leading kernel pooling engine, enhancing overall inference performance by 50% [5] - Efficient Heterogeneous Computing Scheduling: Supports over ten domestic computing power cards and improves resource utilization by 70% [5] - Deep Domestic AI Ecosystem Collaboration: Collaborates with various domestic chip manufacturers to enhance AI computing efficiency [5] - Support for Over 20 AI Training and Inference Frameworks: Compatible with frameworks like PyTorch, TensorFlow, and others [5] Product Matrix - JD Cloud has developed a one-stop large model product matrix, supporting state-owned enterprises in rapidly deploying large models and AI applications, thereby reshaping AI productivity [6]