Chinese AI models gain real world adoption on global cloud platforms

Amazon Web Services has announced that it has integrated GLM-5.3, the latest flagship model from China’s Z.ai, into the Amazon Bedrock platform. Sources familiar with the matter confirmed to China Daily that AWS will share revenue with Z.ai based on customer usage of the model.

Industry experts noted that this development highlights a broader evolution in the global expansion of Chinese AI models. Rather than merely attracting overseas developers, these systems are increasingly achieving real-world adoption and being actively distributed through major international commercial platforms.

[See more: China expands digital yuan network with eight new bank operators]

According to China Daily, Z.ai is not an isolated case. AWS has incorporated six open-weight Chinese models into Amazon Bedrock, including DeepSeek V3.2, MiniMax M2.1, and Alibaba’s Qwen3 Coder Next. Additionally, Moonshot’s Kimi K3 became available on the platform in September.

Wang Peng, a researcher with the Beijing Academy of Social Sciences, explained that under this cloud revenue-sharing model, open models help cultivate developer ecosystems while cloud infrastructure providers supply the necessary computing power and access to corporate clients. Developers then receive a portion of the income generated when customers utilise the technology.

Expanding the reach of Chinese AI models

Recent data from the AI aggregation platform OpenRouter demonstrated that Chinese models processed about 57.46 trillion tokens during the week ending Sunday, compared with approximately 16.2 trillion tokens processed by United States models. This marks the twenty-third consecutive week that Chinese models achieved higher weekly token usage on the platform. DeepSeek alone accounted for 21.8 percent of text requests during OpenRouter’s full week, with Z.ai and Qwen also ranking among the leading model providers.

Further indicators come from Hugging Face, where Chinese-developed models accounted for about 41 percent of platform downloads over the past year, surpassing United States counterparts. By summer, Alibaba’s Qwen family recorded roughly 2.045 billion downloads, while derivative models built on Qwen reached 151,448 – about 2.6 times the number of derivatives based on Meta’s models.

Zhang Yaqin, founding dean of Tsinghua University’s Institute for AI Industry Research, stated that openness could become a structural advantage for the industry. 

Meanwhile, Shen Yang, a professor at Tsinghua University, told China Daily that Chinese artificial intelligence companies are encountering diverse market conditions internationally, with Southeast Asia offering commercial certainty, Gulf markets providing higher value per customer, and Latin America presenting significant unmet demand.