Corporate Adoption Signals Growing Appeal of Chinese Technology
The increasing use of Chinese open-weight artificial intelligence (AI) models by United States companies demonstrates the growing appeal of Chinese technology among American developers. A prime example is Harvey, a legal AI firm based in San Francisco. The company recently developed its initial proprietary model, named Harvey Tenet, which was constructed using Moonshot AI’s Kimi K3 model. This move highlights the increasing interest in Chinese open-weight technology among US developers.
Industry experts suggest that Chinese open-weight models are gaining traction not solely because of their performance or cost-effectiveness, but also due to the flexibility and depth of their supporting ecosystems. Despite mounting US political pressure, these models are becoming difficult for global developers to ignore as they improve in efficiency, customization, and engineering capability.
Harvey reported that its first post-trained open-weight model, which utilized Kimi K3, delivered promising preliminary results in both legal AI performance and cost efficiency. The company characterized these initial results as achieving “state-of-the-art performance” when tackling complex legal tasks.
The Edge of Ecosystem Breadth and Customization
The observed industry pivot has prompted US professionals to take a closer look at Chinese open-weight models. Simon Hedlin, an AI policy researcher, commented on Harvey’s initiative, noting that it serves “a great example of the power of open-weight models,” which allows developers to fine-tune the model using specialized industry data or even proprietary company data. This process enables more accurate and less expensive inference.
We’re still only in the very earliest stages of exploring what’s possible to do with highly capable open-weight models. It’s unfortunate that America is lagging behind in developing frontier open-weight models.
Beyond mere performance or pricing, Chinese experts emphasize that the broader competitive advantage lies in the sheer variety of models available. Tian Feng, the former dean of SenseTime‘s Intelligence Industry Research Institute, explained that a key reason US firms are likely to adopt Chinese open-weight models is the vast scope of China’s model ecosystem. He noted that Chinese developers have established a continuous range of models, spanning from smaller versions like Qwen3-4B and DeepSeek V3.2, up to massive, trillion-parameter models such as Kimi K2 and K3. In contrast, the US open-weight ecosystem offers significantly fewer choices outside of the mid-sized category.
According to Tian, this diversity allows corporations to select models appropriate for various computing budgets within a single technology ecosystem, enabling them to deploy, customize, and fine-tune the technology for specific needs. He asserted, “That is where the real advantage in being customizable comes from – not simply from one model topping a benchmark, but from the broad and continuous coverage of China’s open-weight model ecosystem. In practice, professional market assessments tend to put performance, cost and business needs ahead of geopolitical rhetoric.”
Market Data and Policy Friction
The growing demand for Chinese AI models is evident in several recent commercial and research applications. For instance, US industry interest in Chinese AI models became publicly apparent when GLM-5.2 was utilized by Hugging Face to analyze a cybersecurity incident.
Furthermore, a report released by Hugging Face, the world’s largest AI open-source community, on August 14, indicated that some US model releases exceeding 100 billion parameters this year were built upon or leveraged artifacts from Chinese laboratories, such as Thinking Machines’ Inkling (952 billion). The report also highlighted that in nearly every month of 2026, the largest and most powerful open model released from a Chinese lab surpassed the size of any model released by an American lab. China’s monthly ceiling ranged between 754 billion and 2.78 trillion parameters, while US models generally remained below 130 billion in five of the seven months reported. The main exceptions were Nvidia’s Nemotron 3 Ultra at 561 billion in May and June, and Inkling from Thinking Machines Lab.
The business world is seeing tangible benefits. In a case study, the company Airbnb defended its use of Alibaba’s Qwen model for its customer service chatbot in May. The firm reported that since launching the agent, the average resolution time dropped from nearly three hours to just six seconds, according to a report by Forbes.
Despite these practical benefits, the adoption faces regulatory hurdles. In April, the US House committees on China and homeland security sent a letter to Airbnb, requesting clarification regarding the company’s utilization of Chinese AI models. The committee’s letter was linked to an investigation concerning a purported Chinese campaign aiming to “accelerate its AI capabilities by exploiting American innovation,” as reported by Forbes.
Tian Feng concluded that the conflict between US politicians’ national security stipulations and companies’ real-world operational needs is unlikely to resolve quickly, and US businesses are already bearing the financial costs. However, he added that “the adoption of technology will not simply wait for geopolitical concerns to fade,” suggesting that long-term competitiveness will rely more on sustained improvements in engineering efficiency than on isolated algorithmic breakthroughs.
