•  
  •  
 

Bulletin of Chinese Academy of Sciences (Chinese Version)

Keywords

foundation model, innovation ecosystem, bottleneck segment, breakthrough strategy, artificial intelligence

Abstract

Foundation models are a key vehicle driving artificial intelligence toward general intelligence, and their industrialization urgently requires support from a systematic and collaborative innovation ecosystem. This study focuses on the construction of the foundation model industry innovation ecosystem. It first reviews the frontier progress and identifies its essence as a complex innovation network featuring the three-dimensional synergy of technological, organizational, and industrial architectures, and then analyzes the architecture along the upstream, midstream, and downstream of the industrial chain: the upstream supports computing power and data, the midstream undertakes algorithmic innovation and platform services, and the downstream realizes multi-scenario value transformation. On this basis, it identifies four bottlenecks in data (low mobility, quality, and diversity, and high compliance costs), computing power (expensive chips, energy, and maintenance, and insufficient elasticity), algorithms (high hallucination rates, complexity, and homogeneity, and weak interpretability), and applications (lack of standards, talent, and sustainable business models, and slow implementation), and proposes a four-in-one breakthrough pathway centering on “data–computing power–algorithms–applications”, covering data sharing, computing services, algorithm collaboration, and application ecosystems, so as to advance foundation models toward deep integration, open sharing, and improved governance, thereby achieving inclusive intelligence and sustainable development.

First page

1416

Last Page

1428

Language

Chinese

Publisher

Bulletin of Chinese Academy of Sciences

References

[1] Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need// Proceedings of the 31st International Conference on Neural Information Processing Systems. Long Beach: Curran Associates, 2017: 5998-6008.

[2] Moor M, Banerjee O, Abad Z S H, et al. Foundation models for generalist medical artificial intelligence. Nature, 2023, 616: 259-265.

[3] 施锦诚, 王迎春. 大模型产业创新生态系统竞争力评价研究. 科学学研究, 2025, 43(2): 382-393. Shi J C, Wang Y C. Evaluation of competitiveness of foundation model industry innovation ecosystem. Studies in Science of Science, 2025, 43(2): 382-393. (in Chinese)

[4] 唐露源, 谢士尧, 徐源. 大模型产业创新生态系统构建及风险识别研究. 科学学研究, 2025, 43(9): 1972-1981. Tang L Y, Xie S Y, Xu Y. Research on the construction and risk identification of large language model industrial innovation ecosystem. Studies in Science of Science, 2025, 43(9): 1972-1981. (in Chinese)

[5] 施锦诚, 王国豫, 王迎春. ESG视角下人工智能大模型风险识别与治理模型. 中国科学院院刊, 2024, 39(11): 1845-1859. Shi J C, Wang G Y, Wang Y C. Artificial intelligence foundation model risk identification and governance model from the ESG perspective. Bulletin of Chinese Academy of Sciences, 2024, 39(11): 1845-1859. (in Chinese)

[6] Noy S, Zhang W. Experimental evidence on the productivity effects of generative artificial intelligence. Science, 2023, 381: 187-192.

[7] Hager P, Jungmann F, Holland R, et al. Evaluation and mitigation of the limitations of large language models in clinical decision-making. Nature Medicine, 2024, 30(9): 2613-2622.

[8] Bommasani R, Kapoor S, Klyman K, et al. Considerations for governing open foundation models. Science, 2024, 386: 151-153.

[9] Adner R. Ecosystem as structure: An actionable construct for strategy. Journal of Management, 2017, 43(1): 39-58.

[10] 陈晓红, 刘浏, 牛雅娟, 等. 数智病理平台构建及服务模式研究. 中国工程科学, 2025, 27(2): 304-314. Chen X H, Liu L, Niu Y J, et al. Digital intelligence pathology platform and its service pattern. Strategic Study of CAE, 2025, 27(2): 304-314. (in Chinese)

[11] 陈晓红, 刘浏, 袁依格, 等. 医疗大模型技术及应用发展研究. 中国工程科学, 2024, 26(6): 77-88. Chen X H, Liu L, Yuan Y G, et al. Technology and application development of medical foundation model. Strategic Study of CAE, 2024, 26(6): 77-88. (in Chinese)

[12] Singhal K, Azizi S, Tu T, et al. Large language models encode clinical knowledge. Nature, 2023, 620: 172-180.

[13] Vorontsov E, Bozkurt A, Casson A, et al. A foundation model for clinical-grade computational pathology and rare cancers detection. Nature Medicine, 2024, 30(10): 2924-2935.

[14] Fei N Y, Lu Z W, Gao Y Z, et al. Towards artificial general intelligence via a multimodal foundation model. Nature Communications, 2022, 13(1): 3094.

[15] Jacobides M G, Cennamo C, Gawer A. Towards a theory of ecosystems. Strategic Management Journal, 2018, 39(8): 2255-2276.

[16] 任磊, 王海腾, 董家宝, 等. 工业大模型：体系架构、关键技术与典型应用. 中国科学：信息科学, 2024, 54(11): 2606-2622. Ren L, Wang H T, Dong J B, et al. Industrial foundation model: Architecture, key technologies, and typical applications. Scientia Sinica (Informationis), 2024, 54(11): 2606-2622. (in Chinese)

[17] Vinuesa R, Azizpour H, Leite I, et al. The role of artificial intelligence in achieving the Sustainable Development Goals. Nature Communications, 2020, 11: 233.

[18] 郭华源, 刘盼, 卢若谷, 等. 人工智能大模型医学应用研究. 中国科学：生命科学, 2024, 54(3): 482-506. Guo H Y, Liu P, Lu R G, et al. Research on a massively large artificial intelligence model and its application in medicine. Scientia Sinica (Vitae), 2024, 54(3): 482-506. (in Chinese)

[19] Cheng H R, Zhang M, Shi J Q. A survey on deep neural network pruning: Taxonomy, comparison, analysis, and recommendations. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024, 46(12): 10558-10578.*Corresponding author

[20] Costa-Jussà M R, Cross J, Çelebi O, et al. Scaling neural machine translation to 200 languages. Nature, 2024, 630: 841-846.

[21] 刘浏, 张馨, 张琪琪, 等. 医疗领域大模型伦理风险识别、治理及前瞻研究. 中国工程科学, 2025, 27(6): 54-67. Liu L, Zhang X, Zhang Q Q, et al. Ethical risk identification, governance, and foresight of medical foundation models. Strategic Study of CAE, 2025, 27(6): 54-67. (in Chinese)

[22] 李兴腾, 冯锋, 黄鹂强. 突破人工智能大模型的“数据瓶颈”——构建国家级语料库运营平台的思考. 中国科学院院刊, 2025, 40(3): 522-529. Li X T, Feng F, Huang L Q. Breaking through “data bottleneck” of AI large models-Reflections on building a national corpus operation platform. Bulletin of Chinese Academy of Sciences, 2025, 40(3): 522-529. (in Chinese)

[23] Zhou L X, Schellaert W, Martínez-Plumed F, et al. Larger and more instructable language models become less reliable. Nature, 2024, 634: 61-68.

[24] Farquhar S, Kossen J, Kuhn L, et al. Detecting hallucinations in large language models using semantic entropy. Nature, 2024, 630: 625-630.

[25] Jacobides M G, Tae C J. Kingpins, bottlenecks, and value dynamics along a sector. Organization Science, 2015, 26(3): 889-907.

[26] Engelmann J, Bernabeu M O. Training a high-performance retinal foundation model with half-the-data and 400 times less compute. Nature Communications, 2025, 16: 6862.

[27] Bi K F, Xie L X, Zhang H H, et al. Accurate medium-range global weather forecasting with 3D neural networks. Nature, 2023, 619: 533-538.

[28] 温馨, 张超, 郭锐, 等. 推动我国大模型开源创新生态建设的挑战与建议. 中国科学院院刊, 2024, 39(8): 1313-1326. Wen X, Zhang C, Guo R, et al. Challenges and recommendations for building open source innovation ecosystem for large-models in China. Bulletin of Chinese Academy of Sciences, 2024, 39(8): 1313-1326. (in Chinese)

Download
Request a Copy

Share

COinS