A potential U.S. ban on Chinese open-weight artificial intelligence (AI) models could impose significant costs on American businesses, potentially reaching up to $12 billion annually, according to an analysis by a U.S.-based academic. This estimate underscores concerns over the increasing reliance of U.S. technology firms on cost-efficient AI models developed in China.

Daniel Yue, an assistant professor at the Georgia Institute of Technology’s Scheller College of Business, based his calculations on usage data from OpenRouter, a New York-based large language model (LLM) aggregator that enables developers to switch between different AI models via a unified API. Yue estimated that if users were forced to replace Chinese open-weight models with proprietary alternatives, their combined annual expenses could rise by about $2 billion, with the broader economic impact potentially ranging from $3 billion to $12 billion depending on the extent of reliance on Chinese AI models. He emphasized that these figures represent rough approximations rather than precise forecasts, partly due to the difficulty in tracking AI model usage outside centralized platforms.

Tensions around Chinese AI models intensified in late July following the release of Moonshot AI’s Kimi K3 model, which boasts performance on par with leading U.S. proprietary models from companies such as Anthropic and OpenAI. This development reportedly spurred efforts by the previous U.S. administration to restrict foreign open-source AI models. U.S. officials have also accused Moonshot AI of intellectual property infringement.

However, this potential crackdown has met resistance from major American technology companies. Industry leaders including Nvidia, Palantir, and Meta Platforms have jointly urged the U.S. government to avoid restrictions on open-source AI models, warning that premature limitations could stifle competition and drive innovation overseas.

The broader economic impact of such a ban remains uncertain. Yue noted it is unclear whether U.S. firms would migrate to closed models or cease certain AI workflows. Jaya Gupta, a partner at Foundation Capital, predicted in a recent essay that many critical U.S. systems rely on open-weight models for local AI operations. Gupta suggested that an outright ban could quickly erode AI demand, potentially destabilizing the AI infrastructure market, which has seen substantial investment in data centers based on anticipated growth.

For some U.S. startups, Chinese open-weight models have become crucial to managing costs. Ben Cera, founder of the AI agency Polsia, shared that switching to Chinese open-source options cut his company’s monthly AI expenses from $1.2 million to $100,000.

Contrastingly, some experts argue the economic fallout may be limited. Steve Hou, head of research at AI analytics firm Silicon Data, said most U.S. enterprises do not use Chinese open-weight models at a scale that would cause significant cost increases if removed. He suggested that the main impact of these models has been indirect, contributing to lower prices for proprietary models. For example, OpenAI recently announced substantial price reductions on its GPT-5.6 Luna and Terra models.

According to a Goldman Sachs report citing Silicon Data, global spending on LLM inference decreased from $2.07 per million tokens in early June to $1.67 in early July, a trend possibly linked to the adoption of affordable Chinese open-source models. However, Hou noted that pricing advantages for Chinese models are narrowing, with offerings like Moonshot’s Kimi K3 priced comparably to mid-tier U.S. models. Ultimately, overall AI spending will depend on additional factors such as enterprise discounts and the complexity of AI tasks.