Chinese open-source AI is gaining traction among US enterprises as they look for cheaper, more adaptable alternatives to proprietary systems from American labs. Recent spending data shows a modest but clear rise in the share of businesses that pay for model-serving platforms offering access to these models.
Rising Share of Open-Source Model Spending
Ramp's AI Index, which monitors token and subscription spend across its customers, recorded that the proportion of firms using model-serving platforms rose to 6.1% of total AI-spending businesses in July, up from 4.5% in January 2026. The index tracks purchases of both open-weight and Chinese-developed models, indicating a growing interest beyond the dominant US providers.
Key Chinese Players and Pricing
Among the most talked-about models is Moonshot's Kimi K3, an unusually large open-weight model that has demonstrated coding and agentic performance close to leading proprietary systems. Another lab, Z.AI, announced the launch of GLM-5.3-Flash, pricing its service at $0.15 per million input tokens and $0.50 per million output tokens. The aggressive pricing follows similar moves by DeepSeek, which is seeking a $7.4 billion raise at a $74 billion valuation.
Enterprise Adoption Cases
Two established organisations have publicly shifted parts of their operations to open-source models. Thomson Reuters built an in-house model called Thomson-1, based on Snowdon, an adaptation of Alibaba's open-source Qwen model. The model now handles document-review tasks previously run on Claude. CTO Joel Hron explained that "companies do not need ever-larger, more expensive models to get useful results, and that starting from a strong open foundation and specialising it deeply can produce capable AI at lower cost."
Legal-tech firm Harvey, backed by OpenAI and other investors, released Harvey Tenet, a model post-trained on Moonshot's Kimi K3. The company claims the new model outperforms both its base and US frontier systems on complex legal tasks.
Implications for US AI Leaders
Despite the shift, the dominant US labs still command the majority of spend. In July, Anthropic held 43.5% of the market share among businesses, while OpenAI accounted for 39.7%. However, Anthropic's most advanced model, Fable 5, represented only 6% of tokens purchased from the company, suggesting that customers are reluctant to pay a premium for top-tier performance when cheaper alternatives meet their needs.
"We've seen a lot of companies developing industry-specific foundation models using open-source models that are then fine-tuned on very particular data sets," said Alex Brunicki, a partner at Backed VC. "They're not necessarily using the frontier models for all of the work that they're doing. They're actually using these open-source models which are free to use."
The data implies that US frontier labs may have reached a ceiling on what customers are willing to pay for the newest, most expensive models. Open-source offerings appear ready to fill the gap, providing comparable capabilities at a fraction of the cost.
Future Outlook
Analysts expect the trend to continue as more Chinese labs release larger, higher-performing models and as US firms seek to control data usage and reduce expenses. Discussions are already underway between Moonshot and major US cloud providers about revenue-sharing agreements that could broaden the availability of Kimi K3. If these talks succeed, Chinese open-source models could become a mainstream option for a wider range of enterprises, further challenging the dominance of US-based AI providers.
For now, the modest increase in open-source model spend signals a shift in enterprise strategy: rather than chasing the most advanced proprietary system, many companies are opting for flexible, cost-effective solutions that can be tailored to their specific needs.