Kimi Family Leads Chinese AI Models in Open-Source Enterprise Spending

Chinese artificial intelligence models, particularly those from Moonshot AI’s Kimi family, represented nearly 90% of the estimated global enterprise spending on open-source systems, as reported by Menlo Ventures, a Silicon Valley venture capital firm.

The report indicates that Chinese AI systems accounted for 89% of global enterprise spending on open-source models, with Kimi leading at a 61% market share. Other notable contributors included DeepSeek at 15% and Z.ai’s GLM series at 11%.

In contrast, Meta Platforms’ Llama series, the leading Western offering, captured only 7% of the open-source enterprise spending.

The global figures were derived from a September survey involving nearly 400 US enterprise decision-makers managing over $4 billion in total AI budgets, along with public filings and third-party reports.

Open-weight models, unlike closed systems, allow developers to download and modify the underlying parameters rather than relying solely on hosted cloud services. Menlo Ventures categorizes these open-weight architectures under the broader term “open-source.”

Despite the increasing adoption of open Chinese architectures, proprietary systems still dominate the overall model budgets. US firms Anthropic and OpenAI together accounted for 86% of enterprise spending on AI models accessed via application programming interfaces (APIs), with 55% and 31% respectively, while Google held 9%.

Open-source models, however, represented only 4% of total expenditure, despite making up 19% of enterprise usage of tokens, which are the processing units of AI data.

On the global AI model aggregation platform OpenRouter, the token usage of Chinese models surpassed that of US models for the 27th consecutive week from October 26 to November 1.

This trend highlights a growing division in the global AI supply chain, where US companies excel in monetizing closed-source APIs, while Chinese open models provide a foundation for overseas developers to create their own services.

For instance, US-based AI coding platform Cursor revealed in a technical report that its Composer 2 model was built on Moonshot’s Kimi K2.5. The company enhanced the Chinese model using code-heavy data sets and applied reinforcement learning to improve its software engineering performance.

Menlo Ventures noted that coding tools and enterprise chatbots now constitute 76% of the $77.3 billion global enterprise AI application market, with autonomous agents driving the next phase of business demand.

Menlo is an active investor in Anthropic and several AI infrastructure providers mentioned in the report.

Source: SCMP Tech (Asia)