AI agents and cloud models can send sensitive local or enterprise data to public networks, while traditional confidential-computing approaches have often been too slow for AI workloads. Jinghua says its operator-level redesign cuts AI confidential-computing time loss by three to four orders of magnitude and supports domestic heterogeneous GPUs, and its nearly RMB 100 million Pre-A round is earmarked for domestic-compute adaptation, core technology iteration and dedicated confidential-computing chips.