计算机行业快报：UltraMem架构为推理降本，AI应用全面落地可期

Investment Rating - The report maintains an investment rating of "Leading the Market - A" for the computer industry [1][11]. Core Insights - The introduction of the UltraMem architecture by ByteDance's model team significantly enhances inference efficiency, achieving a speed increase of 2-6 times compared to the MoE architecture and reducing inference costs by up to 83% [2][4]. - The report highlights that the continuous decline in inference costs will accelerate the proliferation of AI applications, with the cost of model inference expected to drop by 1200 times by the end of 2024, making it a critical factor for users when selecting models [5][6]. Summary by Sections Market Performance - The computer industry has shown strong market performance over the past year, with a focus on AI applications and the impact of new architectures like UltraMem [1]. Preferred Stocks - The report identifies preferred stocks in the sector, including Kingsoft Office (688111.SH) and Haiguang Information (688041.SH), both rated as "Buy - A" [2]. Technological Developments - The UltraMem architecture optimizes model structure and retrieval methods, significantly improving inference efficiency while maintaining model performance [4]. - The report notes that the UltraMem model, with a training scale of 20 million values, achieves industry-leading inference speed and performance under the same computational resources [4]. Investment Recommendations - The report suggests focusing on AI application-related stocks across various sectors, including enterprise services, office software, multi-modal fields, finance, education, and healthcare, as well as domestic computing chip manufacturers and AI server companies [6][7].