# Why Chinese AI Models Now Command Nearly Half of US Enterprise Tokens > Source: > Published: 2026-07-16 20:53:45+00:00 # Why Chinese AI Models Now Command Nearly Half of US Enterprise Tokens Chinese AI models have surged to 46% of US enterprise token usage, driven by economics and performance. This shift challenges the dominance of models like GPT-5.6. Chinese AI models have reached a significant milestone by capturing 46% of US enterprise [token](/glossary/token) usage in just one week this summer, according to an analysis of OpenRouter traffic published on July 7. This is a stark increase from only 4.5% eighteen months ago. On average, over the past year, these models have accounted for 11% of usage, maintaining at least 30% each week since February 8, 2026. ## Economic Efficiency Drives Adoption At the heart of this shift is a simple factor: cost. Chinese open-[weight](/glossary/weight) models offer substantially lower input and output costs. For many enterprises, the value proposition lies in their ability to perform tasks such as extraction, summarization, and retrieval-augmented drafting at a fraction of the price of their counterparts like [GPT](/glossary/gpt)-5.6. The specification is as follows: these models are deemed 'good enough' for numerous real-world applications, challenging the notion that advanced performance is always necessary. ## Overcoming Security Concerns Despite their rising popularity, security and compliance remain primary concerns for enterprises considering Chinese models. However, companies can mitigate these risks by deploying open weights on US-hosted managed APIs, hyperscalers, or even their own infrastructure. This ensures that sensitive tokens don't travel to servers overseas, addressing the most significant barrier to adoption. ## A Practical Migration Strategy So, how can enterprises transition to these cost-effective models? A straightforward 5-minute migration strategy is available. Start by A/B testing with your own prompts via OpenRouter. Employ routing with fallback options to safeguard against complex cases where flagship quality might be essential. Adjust routing percentages based on [evaluation](/glossary/evaluation) results, selecting model options that align with workload and regulatory needs. The upgrade introduces three modifications to the execution layer, allowing for easy integration. What does this mean for the future of AI in the enterprise sector? The dominance of US models is being challenged, and if Chinese models can continue to deliver near-frontier performance at lower costs, this trend is likely to persist. Will enterprises prioritize cost over top-tier performance? This remains an open question, but the current data suggests a shift is underway. Get AI news in your inbox Daily digest of what matters in AI.