On March 13, OpenAI released a proposal for the U.S. AI Action Plan. The report asserts that while the United States currently holds a leading position in the field of artificial intelligence, the success of China-based DeepSeek indicates that this advantage is not as significant as it appears and is gradually narrowing. The AI Action Plan is intended to ensure that AI innovation in the United States continues to outpace that of China, thereby securing U.S. leadership in the AI domain.
However, reducing the rivalry to a simplistic “who leads in AI” frame overlooks its complexity. The competition between the U.S. and China in the AI domain is not a zero-sum game. Rather, it is a multifaceted and complex rivalry, shaped by numerous factors such as geopolitical considerations, access to data, talent, regulatory environments, and technological infrastructure.
The competition between China and the United States in the field of artificial intelligence has driven the development of AI technologies to a more diversified and differentiated contest. The development of large artificial intelligence models exemplifies the evolving nature of this competition. OpenAI’s GPT-4.5, for instance, is specifically designed for complex, high-performance tasks, excelling at intricate text generation and understanding through massive computational resources. This specialization allows it to handle tasks that require a deep grasp of language, context, and nuance. On the other hand, DeepMind’s Perceiver takes a different approach, offering a Transformer variant that can process multimodal data – such as images, sounds, and video – making it versatile across a variety of input types.
The AI model development competition between the two countries has led to the exploration of diverse architectures optimized for distinct use cases, rather than just pushing computational boundaries. The contrasting national approaches highlight the dynamic and evolving nature of AI research, where innovative methods and applications are emerging. This suggests the future of AI will be defined by multiple specialized and adaptable systems, not a single dominant architecture.
Diversified Development of Large AI Models
The technical landscape of large AI models is becoming increasingly diversified, making it impractical to draw comparisons based solely on a single-dimensional perspective. An AI model encompasses a variety of domains, such as generative AI, machine learning, deep learning, computer vision, and hybrid AI models, each employing different architectural frameworks. While most leading generative AI models in the United States, like OpenAI’s ChatGPT-O3 and xAI’s Grok-3, rely on Transformer architectures, this does not necessarily mean that these models will continue to dominate the future of AI development. The rapid evolution of AI technologies, coupled with shifting demands across various sectors, makes it clear that technological leadership cannot be determined solely by the current supremacy of any single architecture. Instead, the future of AI will be shaped by an array of specialized models that each address unique tasks and requirements.
Tech companies in both China and the U.S. are continually innovating and refining AI models, releasing large-scale products with varying applications and distinct advantages. For instance, Anthropic’s Claude 3.5 Sonnet enhances visual reasoning capabilities by improving the AI’s ability to transcribe text from imperfect or noisy images, marking a significant advancement in multimodal AI. Similarly, DeepSeek’s use of the open-source MoE (Mixture of Experts) model exemplifies how efficiency can be boosted by dynamically allocating resources to experts specialized in particular tasks, improving both resource utilization and task performance. On the other hand, Tencent’s Hunyuan Turbo S model represents a different approach, leveraging the Hybrid Mamba Transformer architecture to balance fast, shallow reasoning with slower, more deliberate thinking, achieving a unique flexibility in decision-making.
These examples illustrate the breadth of approaches being taken by AI developers, with each model representing a different frontier in artificial intelligence technology. Though these models excel in different areas, they are not mutually exclusive; rather, they reflect a rapidly evolving landscape where innovation is being driven by multiple, differentiated architectural directions.
Collaboration Opportunities Between China and the U.S.
To maintain its dominance in AI, the United States has tightened export controls and restricted collaboration with China. In January 2025, the Bureau of Industry and Security (BIS) introduced the Framework for Artificial Intelligence Diffusion, limiting China’s access to advanced U.S. AI technologies, including high-end chips, model weights, and cloud computing. However, this approach ignores the potential benefits of cooperation. Both nations have unique strengths, and collaboration could drive innovation in research, security, and global AI standards, ultimately benefiting the broader AI ecosystem.
On the one hand, there is a significant two-way flow of talent in the field of artificial intelligence between China and the U.S., with a high degree of cross-border integration in AI research. Bedoor AlShebli and others, by analyzing datasets of over 350,000 AI scientists and 5,000,000 AI papers, found that most AI scientists migrating to China come from the U.S., while most migrating to the U.S. come from China, highlighting the obvious two-way flow of talent. Additionally, although there is a declining trend, AI research papers resulting from China-U.S. collaboration still significantly outperform papers resulting from U.S. collaboration with other countries. According to a database created by Georgetown University, the number of China-U.S. co-authored articles reached 47,715, significantly higher than the second-largest collaborator, the United Kingdom, with 18,400. Studies have also shown that research papers involving collaboration between the U.S. and China tend to have greater impact than those led by a single country alone.
However, the U.S. government has taken steps to limit the exchange of talent and collaboration between the two countries, particularly in fields like AI. A recent bill called the Stop CCP VISAs Act has been proposed, which aims to exclude Chinese students from participating in U.S. academic projects. Such measures risk damaging the collaborative environment between the U.S. and China across various fields of research and innovation.
On the other hand, there is significant potential for cooperation between Chinese and U.S. AI companies, particularly in areas such as security, governance, and international technology standards. As China’s large AI models rapidly advance, they are increasingly narrowing the gap with those led by U.S. companies. For example, Baidu’s recent release of Ernie 4.5 has demonstrated impressive multimodal understanding and language processing capabilities, while Tencent’s open-source Hunyuan 3D-2.0, launched on March 18, enables the transformation of text and images into 3D models, marking a notable advancement in AI-generated content (AIGC).
Moreover, the ongoing collaboration between Chinese and U.S. AI companies highlights the potential for technological synergy. Recently, Apple partnered with Alibaba to develop advanced AI features, while Ford is exploring the integration of AI models into automotive design, including models from OpenAI, Anthropic, and China’s DeepSeek. These AI models assist in automating tasks such as creating 3D models from sketches and conducting stress analyses on vehicle components, significantly reducing simulation and testing times.
In conclusion, while the AI rivalry between China and the U.S. is often framed as a competition for global dominance, it is, in fact, a multi-dimensional and collaborative challenge. Both nations are driving forward technological innovation in AI, and rather than simply competing, they have numerous opportunities to collaborate. By focusing on shared goals in AI safety, governance, and innovation, the United States and China can not only advance their own technological agendas but also contribute to shaping a responsible and innovative global AI landscape.