The Dual-Loop Paradox: Capital Scaling versus Open Deployment in the China–US Artificial Intelligence Race

DOI: https://doi.org/10.62517/jbm.202609409

Author(s)

Shengchang Zhang1,*, Zongjun Song1, Fred Zhang2

Affiliation(s)

1School of Business Administration, Baise University, Baise, Guangxi, China 2Amador Valley High School, Pleasanton, California, USA *Corresponding Author

Abstract

This monograph frames the changing structural competition in the evolving generative artificial intelligence (AI) between the United States and China in a new analytical framework called the Dual-Loop Paradox. Although the more typical control mechanisms and export controls are tuned to a single large-scale digital code, namely, the compute-intensive scaling model that dominated the close-source giants in the United States, the Chinese ecosystem has strategically re-aligned to the open-weight, deployment-first approach. This paper formalizes the fact that open AI models in China are used to produce two feedback loops that are coherent and reinforcing: a digital diffusion loop that maximizes the efficiency of algorithms with limited compute resources, and a physical deployment loop that transforms its enormous industrial manufacturing base into a proprietary data-creation asset. Using large-scale Mixture-of-Experts (MoE) systems and post-training chain-of-thought systems, Chinese laboratories have shown the capability to reduce capabilities thresholds at the bottom and run at a fraction of Western inference and token licensing costs. We develop an analytic model of the dynamics of optimization of both loops and show that even strict constraints on the compute used in training can be unable to prevent the occurrence of competitive convergence when deployment-driven data and systemic industrial integration have reached extreme levels. The strategic implications to international technology governance, technological dependencies, and multi-pathway AI development are presented in detailed detail.

Keywords

Artificial Intelligence; Technology Scaling Laws; Open Source Diffusion; Mixture-of-Experts; Industrial Flywheels; Strategic Asymmetry; Technology Governance

References

[1] Kaplan, J., McCandlish, S., Henighan, T., et al. Scaling laws for neural language models. OpenAI, 2020. [2] Luong, N. Two loops: How China’s open AI strategy reinforces its industrial dominance. U.S.-China Economic and Security Review Commission Working Paper, 2026. [3] Ding, J. The diffusion deficit in scientific and technological power: Re-assessing China’s rise. Centre for the Governance of AI Working Paper, 2022. [4] OECD. OECD Digital Economy Outlook 2020. OECD Publishing, 2020. [5] Xue, F., Zheng, Z., Fu, Y., Ni, J., Zheng, Z., Zhou, W., You, Y. OpenMoE: An early effort on open mixture-of-experts language models. Proceedings of the 41st International Conference on Machine Learning (ICML), PMLR 235: 55625–55655, 2024. [6] Shazeer, N., Mirhoseini, A., Maziarz, K., et al. Outrageously large neural networks: The sparsity-gated mixture-of-experts layer. International Conference on Learning Representations (ICLR), 2017. [7] Fedus, W., Zoph, B., Shazeer, N. Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity. Journal of Machine Learning Research, 2022, 23(120): 1–39. [8] Krajewski, J., Ludziejewski, J., et al. Scaling laws for fine-grained mixture of experts. Proceedings of the 41st International Conference on Machine Learning (ICML), 2024: 33270–33288. [9] National Institute of Standards and Technology. Artificial Intelligence Risk Management Framework (AI RMF 1.0). U.S. Department of Commerce, 2023. [10] Bommasani, R., et al. On the opportunities and risks of foundation models. Stanford Center for Research on Foundation Models, 2023. [11] Wei, J., Wang, X., Schuurmans, D., et al. Chain-of-thought prompting elicits reasoning in large language models. Advances in Neural Information Processing Systems (NeurIPS), 2022. [12] Liu, J., Tang, P., Wang, W., et al. A Survey on Inference Optimization Techniques for Mixture of Experts Models. ACM Computing Surveys, 2026, 58(10): 1–37.