U.S. firms increasingly adopt Chinese open-weight AI models, data shows
New spending data from Ramp indicates a growing share of U.S. businesses are using model-serving platforms that offer open-weight and Chinese-developed AI systems, rising from 4.5% in January to 6.1% in July 2026. These models, such as Moonshot's Kimi K3 and Z.AI's GLM-5.3-Flash, provide cheaper and more customizable alternatives to closed-source U.S. rivals. The trend suggests enterprises are increasingly looking beyond American AI providers for cost and control advantages.
Ramp's AI Index, which tracks spending across its customer base, recorded the rise in model-serving platform usage from 4.5% to 6.1% over six months. These platforms grant access to open-weight systems, including Moonshot's Kimi K3, noted for its large scale and coding performance, and Z.AI's GLM-5.3-Flash, priced at $0.15 per million input tokens and $0.50 per million output tokens. Hugging Face data shows Chinese labs produced the largest capable open model nearly every month in 2026.
Several organizations have publicly adopted this approach. Thomson Reuters built an in-house model, Thomson-1, adapted from Alibaba's open-source Qwen, to handle document review previously run on Claude. Harvey, the legal tech firm, post-trained its new model on an open foundation. Alex Brunicki of Backed VC observed companies fine-tuning open-source models on specialized datasets rather than relying solely on frontier systems.
This shift could reshape the competitive landscape of enterprise AI, potentially reducing U.S. dominance in the sector while offering smaller businesses access to capable, affordable models. However, reliance on Chinese-developed systems may raise data-security and supply-chain concerns for firms handling sensitive information. The trend could also accelerate price competition across the industry, though the article notes fears that larger GLM releases may introduce advanced cyber capabilities for which organizations are unprepared.