AI推理算力需求即将爆发，深圳云天励飞加注推理芯片

Core Insights - AI inference chips are emerging as a new focus in the artificial intelligence industry, with Shenzhen Yuntian Lifeng (688343.SH) announcing a comprehensive focus on this area during the World Artificial Intelligence Conference in 2025 [1][2] - The CEO of Yuntian Lifeng, Chen Ning, highlighted that 2025 will be a pivotal year for AI development, with significant reductions in model invocation costs and a shift from AI as an "expert tool" to a "universal infrastructure" [1][2] - The demand for inference computing power is expected to experience explosive growth as AI transitions from training to inference [1][3] Industry Trends - The report from CITIC Securities indicates that three main factors are accelerating the demand for inference computing power: the integration of AI with existing internet businesses, the combination of agents and deep reasoning, and the penetration of multimodal capabilities [2] - AI is anticipated to redefine various electronic products, including wearable devices and household appliances, enabling them to interact more naturally and respond to complex commands [2] Company Developments - Yuntian Lifeng is focusing on AI inference chips, which are categorized into training chips and inference chips, with the latter being crucial for utilizing neural network models for predictions [3] - The company has developed four models of chips: DeepEdge10C, DeepEdge10 Standard, DeepEdge10Max, and DeepEdge200, with the DeepEdge10 series specifically designed for edge AI applications [3][4] - The DeepEdge10 series employs a "computing power building block" architecture, allowing for scalable integration of computing units to meet varying power requirements [4][5] Financial Performance - Yuntian Lifeng reported an 81% revenue growth in 2024, with a further increase to 160% in the first quarter of this year [5] - The management expressed confidence in maintaining high growth rates in the second half of the year, driven by advancements in AI inference algorithms and increasing demand for computing power [5]