Chinese AI models lag in monetisation

By the Editorial Team · Thursday, October 8, 2026

Chinese AI developers like DeepSeek and Moonshot AI face challenges in monetizing their open-weight models, as third-party providers capture a large share of the revenue generated from their technologies. To improve earnings, these firms are implementing new licensing arrangements that require revenue sharing from major cloud platforms and service providers distributing their models.

China’s leading developers of open-weight artificial intelligence (AI) models are gaining traction worldwide, but a significant portion of the revenue generated by their technologies is being captured by third-party providers rather than the original creators, according to a recent study.

AI developers such as DeepSeek and Moonshot AI have adopted a strategy of making their model weights—the core components powering their AI systems—freely available for download. This open approach enables other companies to operate the models on their own infrastructure and customize them independently. While this has accelerated adoption, it has also exposed the original developers to steep competition from rival platforms that host these models and offer access at competitive prices.

For instance, DeepSeek, a start-up based in Hangzhou, reportedly secured just about 8 percent of estimated customer spending on its models accessed through OpenRouter, a widely used third-party aggregator. This data is part of a report released Tuesday by the U.S.-based research organization Epoch AI.

A similar dynamic is seen with Beijing-based Z.ai, also known as Zhiji AI. After releasing the weights for its GLM-5.3 Flash model in August, Z.ai initially dominated access, holding an 88 percent share of the tokens served through OpenRouter. However, within 20 days, the number of third-party providers offering the model increased from 12 to 26, with several charging lower prices than Z.ai itself. Consequently, Z.ai’s share dropped to 22 percent, even as overall usage of the model increased by 17 percent. Their daily token volume fell by 70 percent during the same period.

These figures illustrate the challenges Chinese AI developers face in monetizing direct access to their models. Z.ai reported a gross margin of 24.6 percent for its application programming interface (API) platform during the first half of this year, a rise from 18.9 percent in 2025. This platform allows clients to integrate Z.ai’s models into their own applications and tools.

In contrast, U.S. companies employing closed-weight models retain tighter control over distribution and pricing, yielding higher profitability. Epoch AI estimates that Anthropic, for example, achieved an 85 percent gross margin by offering its Opus 4.8 model exclusively through its own API.

DeepSeek stands out among open-weight developers, estimated to have gross margins between 70 and 80 percent due to comparatively low infrastructure costs. Yet even this company sees a substantial share of end-user demand captured by third-party hosts.

Overall, revenue generation remains uneven. China’s top six AI firms—including ByteDance, Alibaba Group Holding, Z.ai, Moonshot AI, DeepSeek, and MiniMax—collectively commanded roughly one-tenth of the AI revenue earned by OpenAI and Anthropic combined as of September.

To address this gap, Chinese developers are exploring new licensing frameworks aimed at securing greater portions of revenue from third parties using their models. In July, Moonshot AI launched the Kimi K3 model under a license requiring major providers offering the model as a service to enter into revenue-sharing agreements. Additionally, Moonshot has sought to claim up to 30 percent of service revenues associated with Kimi K3 on prominent cloud platforms such as Amazon Web Services, Microsoft Azure, and Google Cloud. The model was made available on Amazon’s cloud platform last month.

These developments highlight the evolving strategies among Chinese AI firms as they seek to balance open innovation with sustainable monetization in a highly competitive global market.

This article was reported by synthesizing coverage from multiple independent news outlets to present a balanced, unbiased account. Read about our editorial standards. Spotted an error? Let us know.