A $2 billion equity placement and a $3 billion convertible bond issuance announced September 13 provided the capital for Zhipu's shift toward autonomous AI development. The funding supports a "Fully Self Training" system where the next generation of GLM foundation models will generate their own training data and construct the environments used to train subsequent versions. This approach creates a closed loop of recursive self-improvement for model architecture and infrastructure.

Zhipu will allocate approximately 60% of the net proceeds to compute resources, production inference, and the GLM technical stack. The remaining funds are split between 15% for strategic investments and potential acquisitions and 25% for general working capital. The company's roadmap follows reports of similar recursive capabilities at Google DeepMind and discussions of AI-driven research by OpenAI.