GPU四小龙，春节不打盹儿 | 海斌访谈

Core Viewpoint - Companies that do not take action in the AI wave, particularly in the release of large models and domestic chip enterprises, may fall behind in the industry Group 1: Domestic GPU and Model Development - Domestic GPUs are increasingly well-matched with domestic models, with a surge in large model releases around the Spring Festival [2] - ByteDance launched the video generation model Seedance 2.0, while Zhizhu and MiniMax released their respective models GLM-5 and MiniMax-M2.5 [5][6] - Domestic GPU companies are on standby during the Spring Festival to quickly adapt to new models, indicating the fast-paced nature of AI technology [2][4] Group 2: Model Adaptation and Performance - Companies like Moer Thread and Nuxi have successfully adapted new models to their GPUs, with Moer Thread's MTTS5000 and Nuxi's Xiyun C series being highlighted [4] - Adapting models to run efficiently on GPUs can significantly improve performance, with general GPU architectures allowing for quicker adjustments compared to specialized architectures like NPU [4] - The adaptation process is crucial for ensuring that large models perform optimally on domestic chips, which is essential for competitive positioning [4] Group 3: Market Dynamics and Future Expectations - The release of DeepSeek's models has raised expectations for domestic AI capabilities, with significant improvements noted in the past year [9][10] - Domestic chip companies are increasingly collaborating with large model enterprises, establishing a routine for Day 0 adaptations [14] - The anticipated release of new models from DeepSeek could further stimulate deployment efforts among state-owned enterprises, expanding market opportunities for domestic GPU companies [14] Group 4: Competitive Landscape and Technological Advancements - Domestic GPU companies are beginning to compete with international firms like NVIDIA, particularly in inference tasks where performance requirements are less stringent [11] - The rapid iteration of domestic chips is driven by the growing demand for AI computing power, with companies like Moer Thread and Nuxi accelerating their product development cycles [15][16] - The industry is optimistic about overcoming existing technological gaps, with a belief that domestic firms will eventually achieve parity with international competitors [16]