Chinese Models Take 89% of Enterprise Open-Source AI Spending, Menlo Ventures Says

In Menlo's breakdown, Kimi takes 61 per cent of open-source spending, followed by DeepSeek at 15 per cent, Z AI's GLM models at 11 per cent and Qwen at 3 per cent. Meta's Llama is the most popular Western open model in the report, with 7 per cent.

The result gives Chinese labs a commanding lead in one corner of the enterprise market, but a small one. Menlo estimates that open-source models handle 19 per cent of the tokens enterprises process, up from 11 per cent a year earlier, but account for only 4 per cent of spending, down from 6 per cent. The report puts the gap down to price: average open-source token prices fell 3.3 times over the year to about $0.13 per million tokens. Usage of Chinese open models is growing quickly, but they bring in comparatively little revenue.

Anthropic Widens Its Lead

At the other end of the market, spending is concentrating among the three frontier labs. Menlo estimates that Anthropic, OpenAI and Google together hold 69 per cent of enterprise AI deployment dollars, or $122.7 billion, up from 50 per cent.

Anthropic's share of enterprise spending on large language model APIs rose to 55 per cent from 40 per cent, according to the report. OpenAI's rose to 31 per cent from 27 per cent, while Google's fell to 9 per cent from 21 per cent. That leaves Anthropic and OpenAI with 86 per cent between them. Near-frontier and open models, a group in which Menlo includes xAI's Grok 4.7, Kimi K3, DeepSeek V4 and GLM-5.3, hold just under 5 per cent of API spending, even though the report describes several of them as close to frontier performance at a fraction of the price. Menlo Ventures is an investor in Anthropic.

A $177 Billion Deployment Market

Menlo puts enterprise spending on deploying AI at $176.8 billion by September 2026, 5.4 times the $33 billion it estimated for 2025. Separately, it estimates $102.5 billion was spent on developing models, mostly on training compute. Infrastructure takes $99.5 billion of the deployment total and applications $77.3 billion.

Coding is the biggest application category at $34.0 billion, up 7.7 times, followed by enterprise chatbots at $25.0 billion. Together the two account for 76 per cent of application spending. The labs' own coding products, Claude Code and OpenAI's Codex, account for $24.0 billion, or 70 per cent of coding spending, according to the report. Among business departments, customer success is growing fastest, at 4.3 times.

The report also finds that software agents now generate 80 per cent of tokens on OpenRouter, the model routing service, about seven times the volume of human-driven use. Menlo says agent traffic overtook human traffic in February and has grown 37 times since then. It notes that OpenRouter is only one channel.

The findings draw on a survey of nearly 400 US enterprise decision-makers responsible for more than $4 billion in AI spending, which Morning Consult conducted between September 8 and 26. The market sizes are Menlo's own bottom-up estimates, expressed as annualised run rates and built from the survey, public filings and third-party data. The authors are Tim Tully, Joff Redfern, Deedy Das and Derek Xiao.

Menlo expects the frontier labs' grip on enterprise spending to loosen over time, but the report presents this as a forecast, not something its data yet shows.