Group 1 - The core viewpoint is that the application of AI in enterprises is still in its early stages, with a significant gap between consumer and enterprise adoption rates. In 2024, the penetration rate of generative AI among U.S. residents is projected to reach 39.6%, while the adoption rate among U.S. enterprises is only 5.4% [2][4] - The number of A-share listed companies mentioning AI in their financial reports has rapidly increased from 172 in 2020 to over 1200 in 2023, yet this still represents less than 20% of all A-share companies [2][4] - The EU's AI enterprise adoption rate varies between 3.1% and 27.6%, with an overall average of 13.5% as of 2024, indicating that AI enterprise applications are still in the nascent stage across different regions [2][4] Group 2 - AI application in enterprises shows significant industry differences, primarily influenced by information density. Industries with higher information density, such as computing, telecommunications, and media, are more likely to adopt AI [4][6] - In 2023, over 250 A-share listed companies in the computing sector mentioned AI, accounting for more than 70% of mentions, while industries like food and beverage, agriculture, and coal have very low or no mentions [4][6] - The highest AI adoption rate in the U.S. is found in the information sector at 18.1%, while agriculture has the lowest at 1.4% [6][8] Group 3 - High-density information fields such as programming, advertising, and customer service are leading in AI application. For instance, programming is significantly influenced by AI, with companies like Google and Microsoft reporting that a substantial percentage of their new code is AI-generated [9][11] - In advertising, AI has improved click-through rates significantly, with some ads achieving a 3.0% click rate compared to the historical average of 0.1% for banner ads [11][13] - Customer service applications of AI have shown efficiency improvements, such as Klarna's AI assistant handling 230 million conversations in one month, equating to the workload of 700 full-time agents [11][13] Group 4 - Traditional industries face challenges in digital transformation, including poor data infrastructure, low accuracy of AI models, and organizational resistance. These issues hinder the integration of AI into broader business processes [14][15] - The average hallucination rate of large language models is 6.7%, with some models reaching as high as 29.9%, which poses a challenge for industries requiring high accuracy [15][16] - The disparity between software and hardware investment in China, where IaaS dominates, contrasts with global trends, leading to inefficiencies in AI project implementations [16][17] Group 5 - AI is considered a general-purpose technology (GPT) that requires time to impact productivity significantly. Historical examples show that the benefits of GPTs often manifest only after a considerable delay [18][20] - The productivity paradox, where significant technological advancements do not immediately translate into productivity gains, is evident in the current AI landscape, as U.S. labor productivity growth remains low [20][22] - The expectation is that AI will follow a similar trajectory as past GPTs, with a potential future turning point for productivity improvements yet to be identified [20][22]

一、AI的企业应用仍处于初期阶段 人工智能的2C应用进展很快，2024年美国居民生成式AI的渗透率已达39.6% （来源：圣路易斯联储） 。然 而，当前的模型厂商还热衷于评分打榜、技术炫耀，企业应用尚处于早期阶段。迫切需要找到丰富落地 场景，加快推进AI和各行各业的深度融合。 国联证券对A股上市公司财报中提及AI的情况进行了梳理，近年提及数量迅速增加，从2020年的172家 上升至2023年的超过1200家，然而在所有A股上市公司的占比仍然不高，2023年还不到20%。根据国家 经济研究局 （NBER） 数据，截止2024年2月，美国AI企业采用率仅有5.4%。根据欧盟统计局数据，2024 年欧盟各国AI企业普及率在3.1%-27.6%之间，总体为13.5%。如下图所示。各国对问题的定义和调研方 法有所不同，以上数据不能简单横向比较，但都反映出AI的企业应用还处于初期阶段。 图 2024年欧盟的AI企业普及率 来源：根据欧盟统计局数据整理，2025 闫德利 腾讯研究院资深专家 二、信息密度越高，AI应用越易越深 AI的企业应用具有明显的行业差异，它与信息密度有关。大体是信息密度越高，AI应用越容易越深入； 信 ...