Chinese models take 15% market share
- Menlo Ventures said Chinese models led by DeepSeek, Alibaba’s Qwen and Moonshot’s Kimi reached about 15% of global usage by November 2025. - The same report said enterprise generative-AI spending hit $37 billion in 2025, up from $11.5 billion in 2024, with $19 billion in applications. - Open-weight releases and lower-cost deployment widened adoption beyond U.S. leaders. (menlovc.com)
Chinese models led by DeepSeek, Alibaba’s Qwen and Moonshot’s Kimi reached about 15% of global model usage by November 2025, according to Menlo Ventures’ year-end enterprise AI report. (menlovc.com) (trendforce.com) That share was about 1% a year earlier, the report said, marking a sharp jump in 12 months as Chinese labs pushed open-weight models that developers could download and adapt. (trendforce.com) Menlo said companies spent $37 billion on generative AI in 2025, up from $11.5 billion in 2024 and $1.7 billion in 2023. More than half of that 2025 total, or $19 billion, went to applications rather than model makers or infrastructure. (menlovc.com 1) (menlovc.com 2) The shift helps explain why model market share moved so fast. When a model is open-weight, developers can run it inside their own products, trim costs, and avoid paying a closed provider for every query. (trendforce.com) DeepSeek’s R1 release in January 2025 accelerated that trend. TrendForce, citing Nikkei, said the model’s low-cost, high-performance launch helped push Chinese systems into coding, design and other advanced tasks. (trendforce.com) Alibaba’s Qwen family then became a distribution engine. Xinhua reported on January 13, 2026 that Qwen had reached 700 million downloads on Hugging Face and had passed Meta’s Llama in cumulative downloads by October 2025. (english.news.cn) The enterprise data points the same way. Menlo surveyed about 500 U.S. enterprise decision-makers and said generative AI now accounts for more than 6% of the software market, with at least 10 products above $1 billion in annual recurring revenue. (menlovc.com) Menlo also said 76% of enterprise AI use cases in 2025 were bought rather than built internally. That favors models and tools that are cheap to test, easy to integrate and already embedded in applications. (menlovc.com) The report argues the next fight is less about raw model access and more about proprietary data and constant evaluation. In practice, that means companies are trying to build feedback loops that keep improving answers after deployment, not just at launch. (menlovc.com) So the 15% figure is not just a leaderboard update. It shows Chinese open-weight models moved from fringe adoption in late 2024 to a meaningful share of global usage by late 2025, while enterprise spending kept rising fast enough to support multiple winners. (trendforce.com) (menlovc.com)
