Tencent-backed Cloud Leopard DPU maker targets $4.2B AI chip IPO

Cloud Leopard’s planned ChiNext raise is a reminder that the AI race is as much about networking silicon as it is about GPUs. DPUs sit in the data-path between compute and storage, offloading networking, security and virtualisation work so that GPUs and CPUs can stay focused on tensor math. A domestically controlled, Tencent-backed DPU supplier that already dominates China’s all‑feature DPU market is strategically important: it helps Chinese hyperscalers build AI clusters with less dependence on foreign NICs and accelerators, and lets them optimize for local workload patterns.

For AGI timelines, more capable and cheaper DPU infrastructure essentially lowers the effective cost of running large models and agentic workloads at scale. It also makes it easier to stitch together heterogeneous compute – domestic GPUs, specialized accelerators, and perhaps future AI-specific ASICs – into a coherent cluster. The IPO, if successful, would give Cloud Leopard billions of yuan to invest in next-generation AI‑focused DPUs, further raising the floor of what “commodity” AI infrastructure looks like in China. That puts additional pressure on Western vendors, but also increases the odds that frontier‑class training and inference remain economically viable even as model sizes, context windows and agent complexity grow.