Chinese AI models are doing the work but not getting paid for it, Dimension Capital tells investors

Chinese AI labs have quietly become the engine behind a huge share of Western AI workloads. The money, though, stays in America.

That’s the core finding from Dimension Capital, which shared its analysis with limited partners after a research trip to Beijing and Shanghai. The venture firm, which closed an $800 million fund in July 2026, found that open-weight models built by Chinese teams now account for more than 60% of token usage on platforms like OpenRouter, up from essentially nothing not long ago.

The work flows east, the dollars stay west

The pattern Dimension describes is a kind of lopsided symbiosis. US frontier labs like OpenAI and Anthropic develop leading proprietary models. Chinese teams, including DeepSeek, Alibaba’s Qwen, and ByteDance’s Doubao, produce open-weight alternatives that American application companies then refine and deploy. Eight of the top 10 models on OpenRouter now originate from Chinese labs, capturing more than 25% of total token volume on the platform.

But token volume is not revenue. The inference providers actually collecting payments from developers and enterprises are overwhelmingly American: companies like Fireworks and Baseten. Chinese model builders, despite powering a massive share of usage, see only a sliver of the financial return.

The revenue gap is staggering. OpenAI is generating roughly $40 billion in annual recurring revenue. Anthropic has surpassed $6.5 billion. Chinese model companies, by contrast, sit somewhere below the hundreds of millions.

Open-weight models: a gift to competitors

Open-weight models are, by design, free to use. Labs release the model weights publicly, allowing anyone to download, fine-tune, and deploy them without licensing fees. For US companies building applications on top of these models, it’s a remarkable deal: cutting-edge AI capabilities at a fraction of the cost of licensing proprietary alternatives.

For the Chinese labs producing them, the calculus is different. Open-weight releases build reputation, attract talent, and create ecosystem influence. They do not, however, generate direct revenue at scale. The economic value gets captured downstream, at the inference layer, where American providers charge customers for compute and API access.

Dimension frames this as a “two-way trans-Pacific flow” of AI technology. Despite the revenue imbalance, Chinese frontier labs are commanding valuations 5-10x higher than their US counterparts on a revenue-multiple basis.

Decoupling is harder than it looks

The Dimension report also complicates the popular narrative of US-China tech decoupling. Washington has imposed sweeping export controls on advanced semiconductors and chipmaking equipment, designed to slow China’s AI progress by restricting access to cutting-edge hardware. The integration Dimension observed runs deep, encompassing not just models but also data pipelines and inference frameworks. American companies are actively building on Chinese-developed models, while Chinese labs continue to iterate rapidly despite hardware constraints.

Chinese labs have shown a remarkable ability to do more with less. Operating under tighter resource constraints than their American counterparts, they’ve produced models competitive enough to dominate usage rankings. That efficiency has been a consistent theme since DeepSeek first drew attention for achieving strong benchmark results with fewer high-end chips than analysts expected.

US inference providers are well-positioned as the toll collectors on a highway built with Chinese technology. That arrangement is profitable today, but it also means American revenue depends, in part, on the continued willingness of Chinese labs to release their work openly.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.