Cursor considers phasing out Chinese AI models over geopolitical risk
The SpaceX-owned coding platform is reviewing its use of Chinese models as U.S. lawmakers express concern over technological dependence and security.
The short version
- Cursor plans to explore phasing out PRC-developed open-weight models due to geopolitical risks.
- The coding platform's Composer 2 model was built on Moonshot AI's Kimi K2.5.
- U.S. lawmakers are concerned about China's efforts to create long-term global dependence on its AI.
- The House Committee on Homeland Security is recommending a ban on Chinese AI in federal systems.
- The U.S. is also cracking down on Chinese companies accessing Nvidia chips via remote data centers.
Why it matters
This shift signals a potential cooling of the global AI market where Chinese models are currently favored for their price and performance. For developers and companies, it may mean higher costs or different technical requirements as U.S. firms move toward domestic alternatives.
The shift away from Chinese AI
Cursor, the artificial intelligence coding platform owned by SpaceX, is currently evaluating a significant change to its underlying technology. The company informed U.S. lawmakers that it is considering moving away from using AI models developed in the People's Republic of China (PRC). This decision stems from concerns regarding the geopolitical risks associated with integrating foreign technology into American infrastructure.
The move comes after a report was shared exclusively with CNBC by the House Committee on Homeland Security and the House Select Committee on China. The report followed a months-long investigation into how American technology firms are adopting Chinese AI models. Cursor's specific model, Composer 2, was built using the Kimi K2.5 model from the Chinese laboratory Moonshot AI.
While Chinese models have seen a spike in global usage this year, the move by Cursor suggests a growing hesitation among major U.S. tech players. The company was acquired by SpaceX in August at an implied equity value of $60 billion. As part of this transition, Cursor is also involved in training the Grok family of models for SpaceX.
Drivers of Chinese AI adoption
Despite the growing geopolitical tension, Chinese AI models have become increasingly popular among international developers. The House investigation found that several factors are driving this trend, most notably the cost and customization offered by Chinese labs. Many firms find these models provide near-frontier performance that is difficult to match with current American alternatives.
The report highlights a specific gap in the current market, noting that the U.S. currently lacks a competitive open-weight alternative at the price and performance levels that the market demands. Open-weight models are versions of AI where the underlying parameters are available for others to download and run on their own hardware. This accessibility makes them highly attractive for developers who want to maintain control over their systems.
Lawmakers expressed concern that Beijing is attempting to turn these technological gains into a form of global market share. The goal, according to the report, is to create a long-term dependence on Chinese AI infrastructure. By offering high-quality tools at lower prices, Chinese firms may be positioning themselves to become the backbone of international tech stacks.
Federal policy and security
The U.S. government is responding to these concerns with a series of new recommendations and restrictions. One major proposal includes the creation of a national American open-weight initiative. This program would combine public computing power with shared evaluation infrastructure to provide lawful access to American frontier models for training purposes.
The House committees also suggested that the government should build on an existing order from October that banned the DeepSeek model from federal civilian systems. The proposed new rule would prohibit the use of any PRC-developed AI models in federal information systems. It would also extend this ban to any company performing federal contracts, creating a significant barrier for firms using Chinese technology.
Beyond software, the U.S. is also looking at hardware security and how Chinese companies might bypass export controls. The Bureau of Industry and Security (BIS) recently announced an interim final rule to prevent Chinese companies from accessing Nvidia chips remotely. This is intended to stop them from using overseas data centers to gain computing power for their own AI development.
Technical hurdles and distillation
A major concern for U.S. officials is the method of distillation, which is a process where new models are trained using older, more established models to help them learn faster. This technique allows smaller or newer models to achieve high performance by mimicking the outputs of larger, more complex systems. Washington is worried that Chinese companies are using this method to gain ground quickly in the AI space.
The investigation into the American tech stack revealed that the adoption of these models is not just a matter of preference but a matter of practical utility. Because Chinese models are currently seen as high-performing and low-cost, many companies have integrated them into their daily workflows. Moving away from these models will require U.S. firms to find viable replacements that meet the same technical standards.
The report suggests that the current environment is a high-stakes battle for AI supremacy between the U.S. and China. As Chinese models continue to improve, the pressure on U.S. companies to find domestic alternatives increases. Cursor's move to explore phasing out these models reflects the growing pressure from both the private sector and the federal government to prioritize national security over immediate cost savings.
What to watch
- Whether Cursor officially announces a timeline for removing Chinese models.
- The progress of the proposed 'national American open-weight initiative'.
- How other major firms like Airbnb and DoorDash respond to the House investigation.
Try it
Check the current capabilities of open-weight models on platforms like Hugging Face to see how they compare to closed-source alternatives.
Sources and links
- Governmentinterim final rule bis.gov
- ReportingCNBC: the article we sourced
First reported by CNBC. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.