Chinese models get 89% of US open-source model spending
Kimi gets 61% of it, but 3 US labs with closed models get 69% of all enterprise AI spending.
Menlo Ventures, a US venture firm, published its 2026 report on enterprise AI on 3 November. It asked almost 400 US enterprise AI leaders. It found that 89% of US enterprise spending on open-source models goes to Chinese models. Open-source models are models that anyone can use and operate. Kimi from Moonshot AI gets 61%, DeepSeek gets 15%, GLM from Zhipu gets 11% and Qwen from Alibaba gets 3%.
How it works
The report says that Chinese labs set the open-source frontier. Low price is not the only cause. The average price for open-source tokens fell from $0.45 to $0.13 for each million in one year. A token is a small piece of text that a model reads or writes.
The numbers
Anthropic, OpenAI and Google have 69% of enterprise AI deployment spending. On OpenRouter, a marketplace for models, DeepSeek is first for text use, before OpenAI. Deedy Das of Menlo Ventures says that no top position is permanent.
What is not known
Artificial Analysis, a test firm, made some tests non-public. The distance between the top open model and the top model then grew from 3 points to 12. Sridhar Ramaswamy of Snowflake questions if open models improved or if test content leaked.
Insight: Open models do the work and closed models get the money
Menlo says open-source models are 19% of platform tokens but 4% of enterprise spending. Vercel reports that open-weight models processed more than 7 in 10 tokens in September, but almost 1 in 4 dollars. This reads as a divide: open models get the use, and closed labs get the revenue. For leaders: a low token price does not make a low budget. Count the value of each task.
Sources
Posted
