Analysis frames US-China technology competition with US maintaining chip design leadership while China accelerates open-source software stack development as structural adaptation to export controls.

When Chinese President Xi Jinping visited Washington to meet with President Donald Trump, artificial intelligence emerged as one of the most important issues on the negotiating table. The US still leads in advanced AI models and possesses a significant advantage in semiconductors, but China is closing the gap through advantages in power supply, workforce, and open-source strategy.

Western experts estimate that leading US AI models maintain a lead of several months over Chinese equivalents, though this gap is narrowing rapidly. Each country's AI capabilities depend on multiple factors beyond the model itself.

**The US advantage in semiconductors**

Viewed as a multi-layered system, semiconductors form the foundational layer where the US maintains a clear advantage. The majority of high-performance specialized chips used to train and run advanced AI models are designed by American companies, particularly Nvidia. Andrew Feldman, CEO of chipmaker Cerebras, notes that the fastest chips currently available are still being developed in the United States rather than China.

Washington has leveraged this advantage through export controls imposed by three consecutive administrations over roughly a decade, restricting Beijing's access to advanced chips and manufacturing equipment. Chips manufactured by Nvidia and other American companies remain the most powerful currently available.

Chinese AI companies have adapted by leasing remote access to overseas data centers, allowing them to continue training large-scale models while narrowing the gap with US systems. Beijing is also betting on Huawei to build domestic computing capabilities. Huawei is accelerating production of its new Ascend chip generation and compensating for individual chip limitations by connecting large numbers of processors into computing clusters. A Huawei executive told Reuters that demand for AI chips in China now exceeds the company's production capacity.

However, Nvidia maintains a significant advantage through its CUDA software ecosystem, widely used by developers to build and run AI applications. Completely replacing American technology remains a difficult task for China.

**China's structural advantages**

China's most notable advantage is electricity. AI data centers consume enormous amounts of power, and as chip deployments reach tens of thousands or hundreds of thousands, sufficient power and grid connectivity become as crucial as owning the chips themselves. China has spent years expanding and upgrading its power system. Ya-Qin Zhang, president of the AI research institute at Tsinghua University and former head of Microsoft China, argues that China has a significant advantage in power system scale and available capacity. China now produces more than double the electricity the US generates and accounts for more than a third of global electricity output.

American tech companies face a different bottleneck: the power grid often cannot keep pace with data center expansion. Some businesses are turning to wind, nuclear, or gas turbine power located near their facilities. Andrew Feldman describes this process as "turning electrical energy into intelligence"—chips and electricity are inputs, and AI is the output.

Human resources also favor China. Studies from 2024 show that China has surpassed the US in several metrics related to top AI researchers, commanding nearly half of the globally surveyed research workforce compared to about 18% for the US. For many years, researchers born and trained in China played vital roles in leading US AI labs. Increasingly, however, more are choosing to remain in China and work for companies like Moonshot, Z.ai, or DeepSeek.

**Open-source as strategy**

The two countries differ fundamentally in how they develop and disseminate AI models. American companies like OpenAI and Anthropic tightly control their systems, building multiple layers of protection to limit misuse for disinformation, cyberattacks, or other harmful activities. Meanwhile, Chinese companies such as DeepSeek, Moonshot, and Z.ai pursue "open" AI models, allowing users to download, study, and modify critical system components.

This approach enables developers worldwide to access technology at lower cost and customize models for specific needs—an area China seeks to turn into a strategic advantage. A CSIS analysis suggests that the performance gap between US and Chinese models has narrowed significantly, with China's "open-weight" models increasingly challenging US leadership.

Open-source advantages extend beyond technology democratization. If Chinese models are adopted as platforms by developers in many countries, Beijing could expand its influence in the global AI ecosystem. The competition is no longer simply about creating the smartest model, but about which country's technology becomes more widely used and forms the foundation for future AI applications.

Rehaan Ahmad, co-founder of Silicon Valley startup alphaXiv, notes that Chinese models still have imperfections but have achieved performance comparable to leading US systems in some tasks. While open models can enable malicious purposes such as cyberattacks, the same technology can detect and defend against such attacks.

**The broader context**

The economic and political environment surrounding AI significantly impacts the race. In the US, public concern is growing about AI's impact on jobs, education, and safety. A September Pew Research Center survey showed that 60% of Americans are uncomfortable with building new data centers locally. This social pressure has made AI regulation an increasingly prominent political debate topic.

In China, this issue is raised in a different context and receives less frequent public discussion. International surveys show people in China are generally more optimistic about AI. A KPMG survey across 47 countries found that 69% of Chinese respondents believe AI benefits outweigh risks, compared to 35% in the US. AI is also being incorporated rapidly into Chinese life—from service robots in hotels and restaurants to medical chatbots.

The US economy has remained relatively resilient to massive AI investment. The stock market is near record highs, and warnings about an AI bubble have not yet triggered clear economic shock. These differences also explain why the two countries approach AI security differently. The US and China remain fiercely competitive, but are beginning to seek common mechanisms to mitigate risks.