Source: Fortune
Summary
U.S. officials accused six Chinese AI companies of using American AI models to train their own systems, claiming they gained “capabilities worth billions.” The FBI, NSA, and CISA said DeepSeek and others underreported training costs by using U.S. models. China’s foreign affairs ministry called the allegations “groundless,” citing self-reliance. A Stanford report noted Chinese models are nearly as strong as U.S. ones, with a 2.7% gap. Analysts said Chinese labs improved efficiency by optimizing algorithms, allowing them to do more with less compute. U.S. restrictions on chips pushed China to develop its own alternatives. Chinese models are now gaining traction in U.S. enterprises due to cost and flexibility.
Our Reading
The numbers tell one story.
Chinese AI companies allegedly used U.S. models to train their own, cutting costs and improving efficiency.
U.S. labs initially relied on brute force compute, while China optimized algorithms to do more with less.
Chinese models are now being adopted by U.S. companies for cost and flexibility.
The U.S. still leads in performance, but China’s efficiency is making it a viable alternative.
Author: Evan Null
More value for fewer tokens
Chinese AI labs have found a way to make their models more efficient by optimizing the attention mechanism, a key part of how AI understands context. This allows them to process information faster and with less computational power. U.S. labs initially focused on using more powerful hardware, but Chinese companies adapted to limited resources by improving their algorithms. This shift was partly driven by U.S. restrictions on access to high-end chips, which forced China to develop its own alternatives. As a result, Chinese models can handle many tasks at a fraction of the cost of their U.S. counterparts.
U.S. enterprises warming up to Chinese models
Chinese AI models are gaining popularity in U.S. enterprises due to their cost-effectiveness and flexibility. Companies like DoorDash and Airbnb are using models from DeepSeek and Alibaba, citing lower costs and better performance. Open-source models from China, such as DeepSeek’s R1 and Alibaba’s Qwen, are being downloaded more frequently than U.S. models on platforms like Hugging Face. This trend is growing as companies look to cut AI costs, with 20% of business leaders reporting that AI spending is becoming a constraint. Chinese models are now being used for tasks like document review and coding, even replacing some U.S. models in specialized work.
Chinese models are starting to replace U.S. ones
Some U.S. companies are beginning to use Chinese AI models for specific tasks, especially where cost is a concern. For example, Thomson Reuters built its own model, Thomson-1, using Alibaba’s Qwen to handle document reviews that were previously done by U.S. models like Claude. This shift is becoming more visible, with data showing that the share of businesses using Chinese models has increased. While U.S. models still lead in performance, Chinese models are proving to be capable for most enterprise tasks. This trend suggests that efficiency and cost are becoming key factors in AI adoption, even as U.S. labs continue to push the boundaries of AI capabilities.
Chinese models are not replacing U.S. ones entirely
Despite the growing use of Chinese models, U.S. AI companies still maintain a performance advantage. Many Chinese labs rely on the work of U.S. frontier models to innovate and improve their own systems. This relationship suggests that the U.S. still sets the standard for AI development, even as China finds ways to compete more effectively. Companies using Chinese models often do so alongside U.S. ones, using the best of both worlds. While the gap is narrowing, the U.S. remains the leader in cutting-edge AI, with Chinese models offering a more cost-effective alternative for many businesses.
The future of AI is becoming more global
The competition between U.S. and Chinese AI models is reshaping how companies approach AI adoption. As Chinese models become more efficient and cost-effective, they are gaining traction in the U.S. market. This shift is not just about cost, but also about flexibility and the ability to adapt models to specific needs. While U.S. labs continue to lead in innovation, Chinese companies are proving that efficiency and resourcefulness can make a significant impact. This dynamic suggests that the future of AI will be shaped by a combination of global competition and collaboration, with both regions contributing to the advancement of the field.
