瀚博半导体：争做全球AI推理芯片的领导者

Core Viewpoint - Nvidia plans to acquire core technology assets from AI chip startup Groq for approximately $20 billion, signaling a shift in focus from "training" to "inference" in AI computing, as real-time, low-cost, and deployable inference capabilities become the new competitive focus in the industry [1][9]. Group 1: Nvidia's Acquisition and Market Shift - Nvidia's acquisition of Groq is seen as a clear signal that the focus of AI computing is shifting towards inference capabilities [1][9]. - The move is expected to help Nvidia cover a broader range of AI inference and real-time workloads [1][9]. Group 2: Insights from Hanbo Semiconductor - Hanbo Semiconductor was founded in 2018 with the insight that cloud AI inference would have greater explosive potential than AI training chips [4][10]. - The CEO of Hanbo, Qian Jun, emphasizes the vast and untapped market potential for cloud AI inference and rendering in the era of AGI [4][12]. Group 3: Hanbo's Product Development and Market Position - Hanbo has launched two series of AI inference and rendering chips, SV and SG, achieving commercial success with major internet companies and operators in China [4][12]. - The SV series chips can achieve data throughput speeds twice that of general-purpose GPUs while requiring lower bandwidth, and they are among the few products in China that natively support FP8 large model inference [6][14]. Group 4: Strategic Focus and Future Outlook - Hanbo focuses on specific applications, such as video processing and AI acceleration, to create value through hardware and software collaboration [5][13]. - The company aims to become a global leader in AI inference chips, with a prediction that 2027 will be a key year for the large-scale replacement of domestic AI inference chips [8][16].