U.S. Tech Firms Shift to Chinese AI Models as Open-Weight Systems Capture Nearly 50% of Traffic
Updated
Updated · The Washington Post · Jul 15
U.S. Tech Firms Shift to Chinese AI Models as Open-Weight Systems Capture Nearly 50% of Traffic
3 articles · Updated · The Washington Post · Jul 15
Summary
American tech companies are increasingly adopting Chinese open-source AI families including Alibaba’s Qwen, Z.ai’s GLM and Moonshot AI’s Kimi, marking enterprise-scale U.S. uptake.
Rising AI bills are driving the shift, as companies look for cheaper alternatives to closed U.S. models from OpenAI and Anthropic.
Chinese open-weight models have gained influence after U.S. restrictions on Anthropic’s flagship Fable model were later rescinded, widening interest in lower-cost options.
Nearly half of U.S. AI traffic by tokens went to Chinese models in late June, up from 16% earlier this year, underscoring how quickly the competitive balance is changing.
As Chinese AI models slash costs and match performance, is loyalty to US providers becoming a business liability?
With companies prioritizing data control, are open-weight models making closed-source AI giants like OpenAI obsolete?
Beyond the massive cost savings, what are the hidden security and geopolitical risks of relying on Chinese AI?
2026 AI Realignment: U.S. Enterprises Embrace Chinese Models for 30–46% of Workloads—Balancing Cost, Risk, and Regulation
Overview
In mid-2026, U.S. companies are rapidly adopting Chinese AI models, moving them from niche alternatives to mainstream solutions. This shift is driven by the strong technical competitiveness and overwhelming cost advantages of Chinese models, which now account for 30-46% of token usage on platforms like OpenRouter. Chinese AI providers achieve dramatic cost efficiency through lower operational expenses and by offering open models with subsidized plans. Models such as GLM-5.2 deliver performance close to the cutting edge but at much lower prices, making them increasingly attractive and difficult for U.S. organizations to ignore.