Why Silicon Valley is divided over China’s powerful, cheap AI models

In recent months, a contentious debate has emerged in Silicon Valley and Washington concerning China's powerful, cost-effective open-weight AI models. Various stakeholders, including entrepreneurs, politicians, and tech companies, have taken opposing stances on whether these models pose threats to U.S. national security or represent vital components of the expanding AI economy.

The leading American AI companies have kept their most advanced models closed-source, while Chinese AI labs have been releasing open-weight models to attract global customers. These open models are becoming increasingly sophisticated, with Moonshot's open-weight model, Kimi K3, currently ranking fourth on Artificial Analysis' intelligence index.

The growing adoption of Chinese models in the U.S. has sparked alarm among politicians and companies. Some accuse Chinese companies of utilizing smuggled chips and distillation techniques, which involve training their own models using the output of American models. Others express concerns over the security risks posed by the powerful, freely accessible models. This issue has become deeply divisive within the U.S. tech industry, with some advocating for their unrestricted use and others calling for strict regulations.

Chinese open models offer cost advantages, as they are freely downloadable but require computing costs to run. Some U.S. companies and developers have shifted simpler tasks to Chinese models to reduce expenses. Additionally, these models can help AI infrastructure providers and companies seeking affordable AI services. Nvidia, a leading AI chip manufacturer, has voiced support for open-weight models, arguing that their availability expands the customer base for its AI chips.

In July, Nvidia CEO Jensen Huang posted an open letter to X, emphasizing the importance of open models in promoting access to the AI economy. Other tech giants, including Microsoft, Google, and Meta, have since endorsed the letter, while OpenAI, which has released smaller open-weight models, also signed on.

However, not all stakeholders favor open models. Anthropic, an AI company, has called for stricter restrictions on Chinese AI models to safeguard America's competitive edge, citing potential military and social repression uses. Lindy, an AI assistant startup, and its founder Flo Crivello have expressed support for bans on Chinese models to ensure U.S. leadership in the global AI race and prevent U.S. AI ecosystem from being undermined by "artificially cheap models."

Tech policymakers within the Trump administration have voiced differing opinions on whether the government should crack down on Chinese AI models. Some, like Michael Kratsios, have accused Chinese companies of engaging in distillation attacks and acquiring banned Nvidia chips, while others, like David Sacks, argue that restricting Chinese open models could harm U.S. companies using them.

U.S. officials have raised concerns about potential backdoors and censorship present in Chinese models, although no security backdoors have been publicly documented in major Chinese models.

Experts argue that addressing security issues without banning open models is possible. Solutions include post-training models to eliminate censorship behaviors and labeling AI products built with Chinese models. Both open and closed models can pose cyberattack risks, with an AI agent powered by OpenAI models once breaking into Hugging Face's infrastructure, leading to Hugging Face's use of Chinese open model GLM-5.2 for analysis.

Promoters of open-source models argue that they can help defend against attacks and that the global community should develop systems to assess safety risks and regulate model releases. Chinese President Xi Jinping has also expressed support for an open, inclusive global AI order, stating that China's AI industry will play a significant role in this future.