Chinese models take 89% of US enterprise open-source AI spend, led by Kimi at 61%
What's the deal? Chinese AI models now account for 89% of US enterprise spending on open-source models, with Moonshot's Kimi alone taking 61%, according to Menlo Ventures' 2026 State of Enterprise AI report, published November 3. Meta's Llama, the leading Western open-source choice, took just 7%.
The breakdown: Behind Kimi came DeepSeek at 15%, GLM at 11%, and Qwen at 3%. The figures reflect a survey of nearly 400 US enterprise AI buyers, extrapolated alongside other data — so the precise model-level shares are estimates, not audited purchase records.
Keep it in perspective: Open-source is a narrow slice of the market. Menlo estimates open-weight models handle 19% of enterprise token usage but only 4% of model spending, after token prices fell 3.3x.
Why now? Recent releases from Chinese labs — Kimi, DeepSeek, Qwen, and GLM — have strengthened their open-weight position through competitive performance and low-cost deployment. That makes them attractive for cost-sensitive workloads and firms wanting more control over hosting.
What's the bigger picture? Proprietary APIs still dominate enterprise AI. Menlo says Anthropic and OpenAI together take 86% of that spend, with Anthropic's share rising from 40% to 55% and OpenAI's from 27% to 31%; Google's fell from 21% to 9%. Frontier labs now command 69% of overall deployment dollars, up from 50% a year ago.
The market itself has exploded. Enterprise AI deployment spending rose from $33 billion to $177 billion in a single year — growth accelerating to 5.4x from 3.9x the year before.
The signal: Kimi's 61% share is not a measure of enterprise AI leadership — it is dominance within a small, price-driven corner. But it marks a notable shift: the main open alternative to US closed labs is now largely Chinese, a trend worth watching as open-weight models compete on cost.
Read more: menlovc.com
Image credit: U.S. Army Combat Capabilities Development Command
