In January 2025, the Chinese app, DeepSeek, became the number one downloaded app on Apple. Marc Andressen declared DeepSeek’s release “AI’s Sputnik moment.” Fast forward 18 months, and a Chinese company has yet again sparked global shockwaves. In mid-July, Chinese startup Moonshot AI released Kimi K3, a model viewed as on par with the most sophisticated models from U.S. frontier companies. The demand was so high they had to halt subscriptions and raised a funding round at $35B valuation.
The global AI race is heating up, with U.S. companies garnering most of the attention while Chinese models gradually expand their global market share. This is important, as the AI models with the largest footprint shape the information landscape. From framing progress in wartime to erasing historical events, AI models reflect the preferences, policies, and narratives of their designers. AI diffusion is not only a technical strategy but also a geopolitical instrument of power and influence.
AI with Chinese Characteristics
Since DeepSeek’s initial release, global usage of Chinese models has accelerated. Most recently, the steady drumbeat of new releases from DeepSeek in April and August, Z.ai in June, Moonshot last week, and Alibaba’s release on August 3rd of the Qwen3.8 Max AI model, highlight growing diversity and capabilities of Chinese AI companies.
Driven by increased capability coupled with lower cost and compute requirements, Chinese models have become the go-to choice for many around the world. For instance, on Hugging Face, usage of Chinese AI models has surpassed those made in the U.S. with 41% of the downloads. In fact, when the breach first occurred, Hugging Face resorted to a Chinese model to battle the AI attacker agent. Moreover, 63% of U.S. OpenRouter usage is now driven by Chinese models. In Silicon Valley, some U.S. tech executives are pushing for greater use of Chinese models by U.S. companies as they are cheaper and require less compute.
The global diffusion of Chinese models is not accidental, but part of a concerted strategy by the Chinese government. When speaking at the ninth annual World AI Summit in Shanghai, Chinese President Xi Jinping declared, “The development of artificial intelligence should not be a solo performance by any single country but rather a symphony of global cooperation.” This statement both implicitly takes aim at the U.S., while simultaneously illustrating how AI serves as a geopolitical instrument.
A Familiar Playbook
In many ways, China is pursuing a similar playbook as it has with other emerging technologies. For instance, from solar panels to steel, China has forced prices down on critical technologies and products by making their products significantly cheaper than competitors. Chinese dumping has distorted markets for decades. In 2012, the U.S. Commerce Department ruled that Chinese dumping of solar panels put U.S. firms out of business. More recently, this strategy is reflected in China’s near monopoly on some critical minerals and rare earth elements.
It’s not just the AI models but the hardware as well. Growing Chinese competition on memory chips is shaking up stock markets around the world. On July 28, China’s leading chipmaker, ChangXin Memory Technologies, saw an initial public offering that has shaken other markets, soaring nearly 500% in the first day. From South Korea to Japan to the U.S. to Germany, tech stocks – and specifically those of chipmakers – took significant hits. Some U.S. tech stocks experienced their worst monthly drop in over a decade.
Security Risks & IP Theft
As in the past, Chinese technological advances often coincide with accusations of IP theft and regulatory circumvention, in addition to anti-dumping lawsuits. In fact, former NSA Director General Keith Alexander declared Chinese IP theft as “the greatest transfer of wealth in history.” There are growing signs that this familiar playbook is underway again.
A U.S. official accused Moonshot AI of IP theft, claiming Moonshot AI used Fable to create its groundbreaking release. Anthropic recently sent a letter to Senators detailing China’s ‘distillation’ approach, wherein frontier AI models are repeatedly queried to produce input-output pairs that then train a smaller and cheaper model for replication. They have similarly made distillation accusations at Alibaba’s Qwen lab, DeepSeek, and MiniMax.
The security risks and IP theft accusations are sparking growing discourse in the U.S. around export-control violations surrounding Chinese AI models. The White House has accused Moonshot AI of circumventing the Nvidia export ban. There have been numerous reports of smuggling Nvidia chips, as well as arrests well beyond the latest accusations. In light of this, the U.S. Department of Commerce’s Bureau of Industry and Security has formally launched an investigation into whether Chinese companies accessed banned, U.S. advanced chips, in addition to distillation and IP theft accusations. Chinese officials have rejected the accusations, and remarked, “China will take all necessary measures to firmly safeguard its legitimate rights and interests against any actions that substantially harm China’s interests.”
In addition, cybersecurity researchers have identified several security risks in China-based AI models. DeepSeek generates 50% more vulnerable code, for instance, and is triggered by reference to politically-sensitive topics to the government, such as Tibet or Uyghurs. Researchers also found code hidden within DeepSeek that has the capability to send user data to the Chinese government. A study conducted by the UK Artificial Intelligence Security Institute (AISI) evaluated Kimi K3’s security capabilities and found them well below those of U.S. frontier models. There also is growing evidence of Chinese AI companies’ outputs reflecting CCP propaganda and censoring current events.
Looking Ahead
The U.S. is not alone in concerns over the security risks of foreign-made AI models. As Chinese models gain market share, China is increasingly wary of security risks of non-Chinese AI models. While China has been a strong proponent of open models, President Xi has also warned of the risks, with a specific need “to constantly refine measures to forestall loss of control.” From this perspective, the Chinese government is most concerned about the output of non-Chinese models that may present answers or information that run counter to government objectives and propaganda.
For China, AI is not just a means for domestic influence, but also to shape global attitudes and perspectives in favorable ways for China. As Microsoft’s report on Global AI diffusion notes, “open source AI can function as a geopolitical instrument, extending Chinese influence in areas where Western platforms cannot easily operate.” The open versus closed weight approach is embedded with a broader regulatory transformation, driven by a confluence of revenue maximization and geopolitics. The third post in this AI series will explore the open versus closed weight approach, shifts in global AI regulations, and how geopolitics and supply chain warfare are shaping the regulatory environment.