Chinese models led by Kimi captured 89% of enterprise spending on open-source AI

If the last time you looked at the open-model market was a year ago, your view may be outdated. This week, venture capital firm Menlo Ventures published research estimating that Chinese AI systems account for 89% of global enterprise spending on open-source models. Moonshot AI's Kimi family leads with a 61% share.

What exactly was measured

An important caveat: this is not the entire AI market, only its open segment. Proprietary US models accessed through APIs still capture the overwhelming majority of total revenue. In effect, these are two different markets. In the proprietary segment, US labs make the money. In the open segment, where a company downloads model weights and runs them on its own infrastructure or through an infrastructure provider, Chinese developers lead.

The figures are estimates. They are calculations by a venture capital firm, not audited financial statements. The methodology should be reviewed in the report itself; it is not included in the excerpt available to us.

Why businesses choose Chinese weights

The summary of the report does not explain the reasons, so what follows is our interpretation, not the authors' conclusions. Open models are chosen when control matters: data does not leave through an external API, the model can be fine-tuned for a specific task, and inference costs are predictable. Judging by the figures, Chinese models currently offer the most competitive balance of quality and price among open options. Other alternatives trail them significantly in spending share.

Consider a simple example. A team is building an internal assistant for contracts. The data is sensitive, request volumes are high, and paying for every token from a proprietary model is expensive. The team chooses an open model, deploys it on its own GPUs or with a cloud provider, and fine-tunes it on internal documents. According to Menlo's data, spending in scenarios like this is increasingly going to Kimi and its Chinese peers.

What to do about it

First, do not treat open source as synonymous with Western models. If you are selecting an open model for a product, Kimi and other Chinese model families should appear on your default comparison list, not be treated as exotic options.

Second, check the license and legal risks. For some companies, especially those in regulated industries or the public sector, a model's country of origin may become a compliance issue. Determine your customers' requirements before signing a contract, not afterward.

Third, do not lock yourself into a single model. With the open-weights market this concentrated, an architecture that lets you quickly swap models behind a common interface reduces both technical and business risk. Run your own tests on real data instead of relying on public benchmarks.

Limitations

This is one report from one venture capital firm, and the 89% and 61% shares are based on estimated spending, not the number of requests or deployments. The rising share of Chinese models in the open segment does not mean they outperform US models in overall quality. Proprietary US models continue to capture most revenue and remain the default choice for many tasks.

The practical conclusion is brief: the open segment is already largely Chinese. If you are building a product on open source, assess these models soberly, with testing and legal review, rather than judging them by habit or country of origin.