StepFun 擬香港 IPO 集資 5 億美元

🔑 增強重點摘要

• •StepFun, founded in April 2023 by former Microsoft executive Jiang Daxin, prioritizes multimodal AI models capable of processing video, images, and text, releasing China's first 1-trillion-parameter model in 2024.[2][6]
• •The company appointed Yin Qi, co-founder of Megvii and chairman of Chongqing Afari Technology, as its new chairman in early 2026 to drive commercialization in end-user terminals like smartphones and smart cars.[1][3]
• •StepFun partners with major Chinese firms including Honor, Oppo, ZTE, and Geely, achieving deployments on over 42 million smartphone devices with nearly 20 million daily active users by end-2025.[1][5]
• •Its leadership includes CEO Jiang Daxin, CTO Zhu Yibo (ex-ByteDance/Microsoft), and Chief Scientist Zhang Xiangyu, forming a top-tier team focused on AI+terminal ecosystems.[3][5]

🛠️ 技術深入

• •Flagship Step 3 model is a native multimodal reasoning system with up to 3x inference speed of DeepSeek-R1 on domestic chips and 70% higher throughput on NVIDIA Hopper GPUs.[5]
• •Released China's first 1-trillion-parameter AI model in 2024, prioritizing progression from unimodal to multimodal models toward embodied intelligence and AGI.[2]
• •Development roadmap: unimodal → multimodal → unified multimodal understanding/generation → world models → AGI, with investments in infrastructure like Shanghai Intelligent Computing Technology.[2][5]

🔮 前景展望AI analysis grounded in cited sources

⏳ 時間線