
由月之暗面Kimi系列领衔的中国人工智能模型,占据了全球企业在开源系统上预估支出的近十分之九,这是一项新研究的结果,尽管专有的美国竞争对手仍占据绝大部分总体收入。
Chinese artificial intelligence models led by Moonshot AI’s Kimi family accounted for nearly nine-tenths of estimated global enterprise spending on open-source systems, according to a new study, even though proprietary US rivals continue to capture the vast majority of overall revenue.
硅谷风险投资公司Menlo Ventures本周发布的报告称,中国AI系统占全球企业在开源模型上的支出的89%。Kimi以61%的份额领跑,其次是DeepSeek的15%和Z.ai的GLM系列的11%。
The report, released this week by Silicon Valley venture capital firm Menlo Ventures, said Chinese AI systems accounted for 89 per cent of global enterprise spending on open-source models. Kimi led with a 61 per cent share, followed by DeepSeek at 15 per cent and Z.ai’s GLM series at 11 per cent.
相比之下,Meta Platforms的Llama系列作为该类别中领先的西方产品,仅占开源企业支出的7%。
By contrast, Meta Platforms’ Llama series, the leading Western offering in the category, accounted for just 7 per cent of open-source enterprise spending.
根据Menlo的说法,这些全球数据是从9月对近400名美国企业决策者的调查中推算得出的,这些决策者管理的AI总预算超过40亿美元,同时参考了公开文件和第三方报告。
The global numbers were extrapolated from a September survey of nearly 400 US enterprise decision-makers managing more than US$4 billion in total AI budgets, alongside public filings and third-party reports, according to Menlo.
与封闭系统不同,开放权重模型允许开发者下载并调整底层参数,而非仅通过托管云服务访问。Menlo将此类开放权重架构广泛归类为“开源”。
Unlike closed systems, open-weight models allow developers to download and adapt the underlying parameters rather than access them exclusively via hosted cloud services. Menlo groups such open-weight architectures under the broad label of “open-source”.
尽管中国开源架构的采用不断增长,专有系统仍占据模型预算的绝大多数。
Despite the growing adoption of open Chinese architectures, proprietary systems still captured an overwhelming majority of model budgets.
报告显示,美国公司Anthropic和OpenAI合计占据了企业通过应用程序编程接口访问的AI模型支出的86%,分别占55%和31%,而Google占9%。
US firms Anthropic and OpenAI commanded a combined 86 per cent of enterprise spending on AI models that were accessed via application programming interfaces, capturing 55 per cent and 31 per cent respectively, while Google held 9 per cent, the report showed.
相比之下,开源模型仅占支出的4%,尽管它们占企业Token(AI数据处理单元)使用量的19%。
In contrast, open-source models represented just 4 per cent of expenditure, even though they accounted for 19 per cent of enterprise usage of tokens – the processing units of AI data.
在全球AI模型聚合平台OpenRouter上,中国模型的Token使用量已连续第27周超过美国模型,时间跨度为10月26日至11月1日。
On global AI model aggregation platform OpenRouter, Chinese models’ token usage exceeded that of US models for the 27th consecutive week from October 26 to November 1.
这种动态凸显了全球 AI 供应链中正在形成的分工:美国巨头在将闭源 API 商业化方面处于领先地位,而中国的开源模型则充当基础,供海外开发者在此基础上改造、构建自己的服务。
The dynamic underscores an emerging division across the global AI supply chain: US giants lead in monetising closed-source APIs, while Chinese open models serve as foundations upon which overseas developers adapt to build their own services.
一个例子是美国 AI 编程平台 Cursor。在 3 月发布的一份技术报告中,Cursor 披露其 Composer 2 模型是基于 Moonshot(月之暗面)的 Kimi K2.5 构建的。这家美国公司在代码密集型数据集上对该中国模型进行了精调,随后应用强化学习以提升其在软件工程方面的表现。
One example is US-based AI coding platform Cursor. In a technical report published in March, Cursor disclosed that its Composer 2 model was built on Moonshot’s Kimi K2.5. The US company refined the Chinese model on code-heavy data sets before applying reinforcement learning to improve its performance in software engineering.
报告中,Menlo 指出,编程工具和企业聊天机器人目前占全球企业级 AI 应用市场 773 亿美元规模的 76%,自主智能体正在推动企业需求的下一阶段增长。
In the report, Menlo said that coding tools and enterprise chatbots now made up 76 per cent of the US$77.3 billion global enterprise AI application market, with autonomous agents driving the next phase of business demand.
Menlo 是 Anthropic 的活跃投资方,同时也投资了报告中考察的多家 AI 基础设施提供商。
Menlo is an active investor in Anthropic, as well as several AI infrastructure providers examined in the report.