The Stanford University Human-Centered Artificial Intelligence Institute (HAI) released its 2026 AI Index Annual Report, highlighting a pivotal shift in the global AI race. The report reveals that the performance gap between China and the US in AI model capabilities has significantly narrowed, with both countries now entering a phase of "parallel competition." However, this progress comes alongside sharply rising capital and resource costs.
Global AI competition enters parallel stage
According to data from model evaluation platforms, top AI models from China and the US now alternate at the top of performance rankings, with the differences having shrunk to near statistical margins of error. While the US held a clear lead in generative AI in 2023, Chinese companies such as DeepSeek and Alibaba have steadily advanced their large-model capabilities between 2025 and 2026, reshaping the competitive landscape.
Looking further into model development, the report notes that the US maintains an edge in the number of cutting-edge models. In 2025, the US launched 50 notable models compared to China's 30. Historically, the US leads globally in total output volume, with China a close second.
Nevertheless, Chinese firms have rapidly closed the performance gap with leading models by employing efficient training strategies, optimizing open-source models, and controlling costs. Some Chinese models now rival mainstream US counterparts in specific tasks and inference efficiency, demonstrating strong engineering execution and a commercialization focus.
Additionally, Chinese players tend to achieve usable performance at lower computing costs, creating a differentiated competition model based on the performance-to-cost ratio.
Distinct advantages remain for each side
The report emphasizes ongoing differences in strengths. The US still leads in key model quantity, capital scale, and compute infrastructure, with 5,427 data centers worldwide. By contrast, China ranks first globally in AI paper publications, patent filings, and industrial robot deployments, reflecting its growing integration strength in industrial application and manufacturing.
China's AI sector has expanded rapidly in recent years. The report highlights Asia's notably higher acceptance of generative AI compared to Europe and the US, with China and several Southeast Asian countries ranking among the highest globally in AI adoption rates and willingness to use AI. This gives Chinese firms an edge in validating applications and commercializing solutions.
Capital investment surges amid resource-intensive growth
From a capital perspective, the AI industry is entering a highly resource-intensive phase. Global AI investments reached US$581 billion in 2025, more than doubling year over year, primarily flowing into data centers and advanced chips. Compute demand has grown approximately 3.3x annually since 2022, indicating that the competition extends beyond algorithms to foundational infrastructure.
However, rapid expansion entails significant cost and environmental impact. The report identifies the energy consumption and carbon emissions associated with training large models as critical concerns. For example, next-generation model training can emit tens of thousands of tons of CO2, while total data center power usage has hit 29.6GW. Against this backdrop, energy efficiency and green AI will be crucial drivers for the next phase of transformation.
Technical capabilities and future outlook
Technically, the report finds AI models approaching or surpassing human expert levels in various professional tests, including mathematical reasoning and programming challenges. Yet they still show clear deficiencies in basic cognition and real-world understanding, indicating that general AI remains in its early development stages.
Overall, the Stanford AI Index 2026 report concludes that the AI industry continues its rapid growth without signs of slowing. With China accelerating application deployment and industrialization, the China-US competitive dynamic is shifting from single dominance toward dual leadership under parallel competition.z
Article translated by Charlene Chen and edited by Jerry Chen
