Chinese Models Lead US Open-Source AI Spending
TMTPost cites Menlo Ventures’ 2026 State of Enterprise AI report saying Chinese models received 89% of US enterprise spending on open-source models, led by Kimi at 61%, DeepSeek at 15%, and GLM at 11%. The report also placed open-source models at 19% of token usage but only 4% of total enterprise model spending. TMTPost additionally argues that long-running agent sessions can make memory and storage costs diverge from revenue billed by token count.
The reported figures show strong demand for Chinese open-source models, while the article’s analysis suggests that memory, storage, and session-state management may become central to the economics of overseas AI inference services.
(本文作者为 硅碳变量,钛媒体经授权发布)
文|硅碳变量,作者|温宁
硅谷知名风险投资机构门罗风投(Menlo Ventures)发布的《2026年企业AI现状报告》显示,美国企业用于开源模型的支出中,89%付给了中国模型。前三名中,Kimi占61%,DeepSeek占15%,GLM占11%。
市场很愿意接受一套简洁叙事:依靠性能与定价,国产大模型抢走海外开发者预算。
但同一份报告里还有另一组数字:虽然开源模型在Token使用量上已经拿下了接近五分之一的份额(19%),但开源模型实际上仅占了企业模型总体支出的 4%,绝大部分的企业AI预算依然流向了闭源大模型。
换句话说,89%是一份浓度很高但体量有限的份额。