SiliconFlow Hong Kong listing application accepted by the CSRC with over 10 million users and 2.9 billion yuan raised
SiliconFlow's Hong Kong listing application has been accepted for filing by the China Securities Regulatory Commission. According to Frost & Sullivan data, it is China's largest independent ecosystem token supplier with over 10 million registered users and more than 13000 enterprise clients.
SiliconFlow's Hong Kong listing application has been accepted for filing by the China Securities Regulatory Commission. According to Frost & Sullivan data, it is China's largest independent ecosystem token supplier with over 10 million registered users and more than 13000 enterprise clients. Founded in 2023, SiliconFlow positions itself as an independent ecosystem token supply platform that does not build its own large models and does not sell compute directly. Its model is to schedule heterogeneous chips from Nvidia, Huawei Ascend, MetaX and Moore Threads through its own inference engine, run mainstream models such as DeepSeek, Qwen and Kimi on pooled compute, and sell service per token. As of June 2026 the platform supported more than 170 AI models across over ten chip types. Its binding to domestic chips is central to its story: it entered public view during the 2025 Spring Festival by launching DeepSeek inference on Huawei Ascend first, and Moore Threads and MetaX later joined its compatibility list, with a certain scale of deployments among central and state-owned enterprise clients. By 2025 annual token throughput it ranked fourth in China's overall token supply market with about 1.5 percent share, and first among independent ecosystem platforms. On funding, SiliconFlow raised close to 2.9 billion yuan in equity financing during 2026 from investors including the China Internet Investment Fund, Guoxin Fund, China Mobile's Changleng Fund, China Orient Asset International, Jingneng Fund and Shanghai Yidian Intelligent Computing Fund, with existing investors Yaotu Capital, Shengyi Capital and GTJA Venture Capital adding on. The company said the two rounds will fund inference engines, heterogeneous compute scheduling and model-chip adaptation. On the developer side its model downloads lead the first tier on Dify Marketplace and its token call volume ranks among the leaders on OpenRouter.