China's Open-Weight AI Models Face Severe Profitability Challenges

Business Insider reports that China's open-weight AI models, while technically impressive, are proving to be a terrible business compared to traditional open-source software. Unlike software, AI inference requires expensive chips, electricity, and data center capacity, meaning each new customer increases infrastructure costs rather than improving margins. Publicly traded Chinese AI labs like Zhipu and MiniMax have posted massive losses—$500 million and $250 million respectively—and their stocks have plunged over 40% and 50% in the past month. Moonshot AI had to halt new sign-ups due to insufficient computing power. Analysts from William Blair and Barclays note that open-weight models primarily generate revenue through hosting and inference compute, but those workloads flow to cloud giants like Amazon, Microsoft, and Alibaba, not the model creators. The strategy of giving away models may pressure Western leaders like OpenAI but creates severe profitability uncertainty for Chinese AI labs.