Infinigence, Zhipu, and Moore Threads Launch Domestic Heterogeneous AI Token Solution

People's Daily reported that Infinigence partnered with Zhipu and Moore Threads to launch a domestic heterogeneous inference AI Token factory solution at the 2026 GAI+C conference. The architecture uses full-system heterogeneous collaboration to accelerate inference, connect models, inference, and chips, and lower comprehensive costs under equal hardware investments. The system relies on heterogeneous collaborative inference technology to allocate distinct hardware strengths across different operational phases. Resources with strong computing power handle the high-density prefill stage, while resources with substantial memory bandwidth execute the decode phase across integrated CPUs, GPUs, memory, and high-speed networking. On the business side, related domestic heterogeneous solutions from Infinigence have fully entered production environments. The deployment stably supports ultra-large models with trillions of parameters while continuously managing high-concurrency real business traffic and enabling regular computing resources to function in core enterprise operations. Across its broader ecosystem, Infinigence is accelerating the large-scale replication of technical and delivery standards. The firm is participating in the construction of multi-regional and multi-source computing power supply chain networks.