Chinese Models Capture 89% of US Corporate Open-Source AI Spending, Menlo Ventures Report Finds

Moonshot AI’s Kimi alone holds 61% of the money US companies spend on open-source AI models, and Chinese models together take 89%, Menlo Ventures reported in its annual survey of enterprise AI.

Open-source tokens sell for 13 cents per million, against a blended 70 cents for closed models, so the price gap is wide. A company that moves workloads to open-source models to cut that cost would, on these numbers, mostly be paying Chinese-built ones.

Who holds the share?

Kimi’s 61% leads, followed by DeepSeek at 15% and Z AI’s GLM at 11%. Meta’s Llama is the most popular Western open-source model, at 7%.

Chinese models’ share of open-source spending has grown to 89%, Menlo said, as their capabilities have come to define the open-source frontier.

How big is the pool?

Open-source models now handle about 19% of tokens, up from 11% a year earlier, but take only 4% of model spending, down from 6%. Between September 2025 and September 2026, open-source volume rose 22.5 times, to an estimated 2.9 quadrillion tokens a month.

The price gap explains the small dollar share. The average open-source token now sells for 13 cents per million, down 3.3-fold, while closed-model prices fell 24%. By The Denominator’s arithmetic, 89% of a 4% slice is about 3.6% of all model spending: large inside open-source, small beside the frontier labs.

How solid is the figure?

The 89% comes from a Menlo survey of nearly 400 US enterprise decision-makers who oversee more than $4 billion in AI spending, which Menlo turned into estimated dollars. The share is a model of US corporate buying, not a tally of invoices.

Menlo also invests in companies on both sides of the split, including Anthropic, the frontier lab with the largest share of spending, and Fireworks, which runs open-source models for customers.

If cost pushes more of a company’s workloads to open-source, how much of that spend would end up with Chinese-built models?