Author: Joint Research Team of Qiushi Economics Editorial Department and CCID Research Institute (求是杂志社经济编辑部、赛迪研究院联合课题组)
Writers: Guo Feiran (郭斐然), Jia Zijun (贾子君), Wang Yuxia (王宇霞), Shu Yu (舒予)
https://www.qstheory.cn/20260430/fcfeb89054fe453ebe265f7e270917e9/c.html
Translated by Claude, edited by Sinocism.com
────────────────────────────────────────────
In recent years, artificial intelligence, as the core engine of a new round of scientific and technological revolution, has been penetrating the fabric of human production and life at an unprecedented speed, breadth, and depth, profoundly reshaping the global economic structure, innovation paradigm, and logic of social governance. At present, China has entered the first tier of global AI development and is at a critical window for crossing from running alongside to leading the race. Facing increasingly fierce international competition and the intrinsic demand for high-quality development, what is the state of China's AI industry — what are its development trends, and where are its shortcomings and weaknesses? With these questions in mind, we conducted on-the-ground research.
近年来，人工智能作为新一轮科技革命的核心引擎，正以前所未有的速度、广度与深度渗透进人类生产生活的肌理，深刻重塑全球经济结构、创新范式以及社会治理逻辑。当前，我国已跻身全球人工智能发展第一梯队，正处在从并跑向领跑跨越的关键机遇期。面对日益激烈的国际竞争和高质量发展的内生需求，我国人工智能产业究竟"家底"如何，发展势头怎样，还有哪些短板弱项？带着这些疑问，我们进行了实地调研。
General Secretary Xi Jinping has profoundly noted: "Artificial intelligence is the strategic technology leading this round of scientific and technological revolution and industrial transformation, possessing a very strong 'leading goose' effect of radiating and driving others." AI is not a linear iteration of a single technology, nor a localized upgrade of one industry, but a global and disruptive reconstruction of the underlying logic governing economic and social operation. To measure its level and trends of development, one must break free of traditional frameworks of technology assessment and industry analysis, and conduct comprehensive judgment across dimensions including technological capability, industrial scale, factor support, and integrated application — only then can one glimpse the full picture and direction of this profound transformation.
习近平总书记深刻指出："人工智能是引领这一轮科技革命和产业变革的战略性技术，具有溢出带动性很强的'头雁'效应。"人工智能绝非单一技术的线性迭代，亦非某个产业的局部升级，而是对经济社会运行底层逻辑的全局性、颠覆性重构。衡量其发展水平和发展态势，必须破除传统的技术评估与产业分析框架，从技术能力、产业规模、要素支撑、融合应用等维度进行综合研判，方能窥见这场深刻变革的全貌与走向。
From the perspective of technological capability, AI technologies led by open-source development have achieved collective breakthroughs, forging new standards in the global developer network. During research at one laboratory, we observed that the research team had introduced an AI self-criticism mechanism: without human intervention, after multiple rounds of self-gaming, the model's accuracy rate on complex programming problems improved substantially. AI has progressed from "able to listen and see" to "able to think, reason, and plan" and then to "mastering how to learn." Overall, China's gap with the international cutting edge in key metrics such as model performance, training efficiency, and multimodal fusion has been continuously narrowing, with parity and leadership achieved in some fields, forming a distinctive technical path of open-source leadership and ecosystem flourishing. In 2025, Chinese open-source models accounted for 17.1% of global downloads. According to recent statistics, 8 of the top 10 open-source models globally come from China. DeepSeek-V4's large model performance is comparable to top international models, while its API price is less than 1% of that of GPT-5.5. Its deeper significance lies in breaking the technological monopoly of a small number of tech giants, allowing millions of developers worldwide to conduct secondary development based on Chinese open-source models. Open-source is not only a giving away of benefits, but also leveraging the strength of others; as knowledge accelerates in flowing and spilling over through open sharing, China's AI technology continuously forges its self-evolving capacity within the open ecosystem.
从技术能力看，以开源引领的人工智能技术实现群体突破，在全球开发者网络中锻造新标准。在一家实验室调研时我们看到，研究团队引入人工智能自我批评机制，无需人工干预，模型经过多轮自我博弈后，复杂编程解题正确率大幅提升。人工智能实现了从"能听会看"到"能思考、能推理、能规划"，再到"掌握如何学习"的进阶。整体上，我国在模型性能、训练效率、多模态融合等关键指标上与国际顶尖水平的差距持续收窄，部分领域已实现并跑、领跑，形成了一条开源引领、生态共荣的独特技术路线。2025年，我国开源模型全球下载量占比达17.1%。据近期统计，全球排名前10的开源模型中，有8款来自中国。DeepSeek—V4大模型性能比肩国际顶尖模型，而应用程序编程接口（API）价格低至GPT—5.5大模型的1%以下。其更深层意义在于打破了少数科技巨头的技术垄断，让全球数百万开发者可以基于中国开源模型进行二次开发。开源不仅是让利，也是借力聚力，随着知识在开放共享中加速流动与溢出，我国人工智能技术也在开放生态中持续锻造自我进化能力。
From the perspective of industrial scale, the AI industry has achieved non-linear explosive growth, with the value overflow effects behind the trillion-yuan blue ocean clearly apparent. In 2025, the global AI market reached $757.58 billion, and China's core AI industry scale surpassed 1.2 trillion yuan. The significance of this 1.2 trillion yuan lies not only in the number itself, but more in the logic of growth behind it. Traditional industries follow the iron law of linear factor inputs and diminishing marginal returns, while AI has broken this curse: technological breakthroughs and application diffusion mutually reinforce each other, forming a positive feedback loop of "the more it is used, the stronger it becomes." Research found that Beijing, as a center of innovation, had a core AI industry scale of 450 billion yuan in 2025, with a cohort of polished algorithmic models acting like "digital technology pumps," ceaselessly transmitting intellectual energy to factories in Hebei, ports in Tianjin, and pastures in Inner Mongolia; Shanghai, with "model-shaping the city" as its handle, is building an ecosystem gravitational field through "Model Speed Space"; Shenzhen is targeting industrial implementation, dedicated to building a highly concentrated enterprise ecosystem precisely serving the real economy. Fundamentally, the AI industry has a clear multiplier effect of "invest one yuan, leverage several yuan," and behind the trillion-yuan scale is a full industrial chain from bottom-level computing power to top-level applications, from core algorithms to intelligent terminals — catalyzing new services, new divisions of labor, and new markets.
从产业规模看，人工智能产业规模实现了非线性爆发，万亿蓝海背后的价值溢出效应明显。2025年，全球人工智能市场规模已达7575.8亿美元，我国人工智能核心产业规模突破1.2万亿元。这1.2万亿元的含金量，不光在于数字本身，更在于其背后的增长逻辑。传统产业遵循要素线性投入、边际收益递减的铁律，而人工智能打破了这一魔咒，技术突破与应用扩散相互强化，形成了"越用越强"的正反馈循环。调研发现，北京作为创新策源地，2025年人工智能核心产业规模达4500亿元，一批打磨成熟的算法模型，如同"数字技术泵"，向河北的工厂、天津的港口、内蒙古的牧场源源不断输送智力动能；上海以"模塑申城"为抓手，通过"模速空间"构建生态引力场；深圳瞄准产业落地，致力于构建高度集聚、精准服务实体经济的企业生态。归根结底，人工智能产业明显带有"投入一元，撬动数元"的乘数效应，万亿规模的背后是一条从底层算力到上层应用、从核心算法到智能终端的全产业链，它催生的是新服务、新分工与新市场。
From the perspective of factor support, China's core AI resources have achieved a strategic leap, with institutional innovation accelerating the release of factor vitality. The latter stage of AI competition depends not only on how fast models run, but also on how solidly the computing power base is built and how freely the data supply flows. In these two core resources, China has established a significant scale advantage. In computing power: 42 ten-thousand-card intelligent computing clusters have been built, and as of the first quarter of this year, intelligent computing power reached 1,882 ExaFLOPS (100 billion floating point operations per second), ranking among the global top tier. In data: there are more than 100,000 high-quality national datasets, with a total of more than 890 petabytes — equivalent to 310 times the total digital resources of the National Library of China. Moreover, institutional advantages are gradually emerging. At the Beijing Data Foundation System Pilot Zone, we observed that a "regulatory sandbox" mechanism has effectively broken the impasse of enterprises holding resources but being "unwilling to open, afraid to open, and not knowing how to open" — allowing enterprises to enter a protected "test field" for fusion training without transferring data ownership. A company's technical director said: "Before, training with our own small data made the model drift further and further; now with more than 10 sectors' real data aggregated in the sandbox, accuracy has significantly improved — the more data is used, the more valuable it becomes."
从要素支撑看，我国人工智能核心资源实现战略性跃迁，制度创新加速释放要素活性。人工智能竞争的后程，不只看模型跑得有多快，还取决于算力底座筑得有多实、数据活水流得有多畅。在这两大核心资源上，我国已建立起显著的规模优势。算力方面，建成万卡智算集群42个，截至今年一季度，智能算力规模达每秒1882百亿亿次浮点运算，位居全球前列；数据方面，全国高质量数据集超10万个，总量超890拍字节，相当于中国国家图书馆数字资源总量的310倍。而且，制度优势也在逐渐显现。我们在北京数据基础制度先行区看到，针对企业手握资源"不愿开放、不敢开放、不会开放"的问题，这里建立的"监管沙盒"机制有效打破了僵局，允许企业在不转移数据所有权的前提下，进入受保护的"试验场"进行融合训练。一位企业技术负责人说："以前用自己的小数据训练，模型越训越偏；现在沙盒里汇聚了10多个行业真实数据，准确率显著提升，数据越用越值钱。"
From the perspective of integrated applications, China's AI is accelerating its penetration into all walks of life, with the breadth of application and depth of integration building new competitive advantages globally. As of the end of 2025, the numerical control rate of key processes in major industry enterprises reached 68.6%, and AI integrated applications are crossing from "scattered blooms" to "full-chain intelligence." First, the range of penetration continues to broaden, with applications covering the vast majority of major industry categories in the national economy, and benchmark applications having formed in manufacturing, healthcare, transportation, finance, energy, and other fields. Second, the level of empowerment has significantly risen, advancing rapidly from auxiliary processes toward core processes such as R&D and design, production and manufacturing, and operations management. At a heavy equipment manufacturer in Shandong, we observed an industrial large-model system that has taken comprehensive charge of the entire chain of processes from blueprint parsing and process planning to quality inspection: new product process design that used to take several senior engineers weeks has now been compressed to within 72 hours, with the rate of good products rising by 5 percentage points. Third, new business forms and models are emerging at an accelerating pace, with intelligent connected vehicles, AI pharmaceutical development, and embodied intelligent robots and other integrated new business forms flourishing vigorously, continuously forming new industrial tracks at the trillion-yuan scale. What we felt deeply in our research is that in this global intelligence competition, whoever has the richest application scenarios, the tightest integration, and the densest industrial feedback holds the standards system and application paradigm that define "how AI is used, where it is used, and how deeply it is used" — and holds the initiative in the intelligent era.
从融合应用看，我国人工智能加速向千行百业渗透，应用广度与融合深度构筑起全球竞争新优势。截至2025年底，我国重点行业企业关键工序数控化率达68.6%，人工智能融合应用正从"点状开花"向"全链智能"跨越。一是渗透领域持续拓宽，应用覆盖国民经济绝大部分大类行业，并且在制造、医疗、交通、金融、能源等领域形成一批标杆应用。二是赋能能级显著提升，从辅助环节加速向研发设计、生产制造、运维管理等核心环节纵深推进。我们在山东一家重型装备制造企业看到，一套工业大模型系统全面接管从图纸解析、工艺规划到质量检测的全链条流程，过去多名资深工程师耗时数周的新品工艺设计，如今压缩至72小时以内，良品率提升5个百分点。三是新业态新模式加速涌现，智能网联汽车、人工智能制药、具身智能机器人等融合新业态蓬勃生长，不断形成万亿级产业新赛道。调研中能够深切感受到，在这场全球智能竞赛中，谁的应用场景最丰富、融合程度最紧密、产业反馈最密集，谁就掌握了定义人工智能"怎么用、用在哪、用多深"的标准体系和应用范式，谁就掌握了智能时代的主动权。