怎么看算力的天花板和成长性？ | 投研报告

Core Insights - The AI industry has a higher ceiling for growth, driven by a multiplier effect of "intelligence" from the "computing power - model - application - data" flywheel, with platforms like OpenRouter showing exponential growth in token usage [2][3] - AI's effectiveness is realized more quickly compared to traditional technologies like 4G/5G, which have longer payback periods due to their engineering properties and limited ARPU growth [2] - Hardware iterations in AI are faster, optimizing price and demand cycles, with significant reductions in unit computing costs occurring every 12-18 months, allowing for more complex models and systems to be developed [3] Industry Analysis - The AI industry's ceiling is determined by the space of tasks that can be automated rather than traditional metrics like population or device count, indicating a much larger scalability potential [2] - AI models and computing power can be deployed immediately and iteratively to meet new demands and scenarios, leading to rolling realization of value [2] - The rapid iteration of AI computing power breaks traditional hardware price-volume cycles, enabling developers to set new benchmarks with more complex models as computing becomes cheaper and more accessible [3] Investment Recommendations - Companies with higher AI computing power ceilings and favorable competitive landscapes are expected to have stronger valuations compared to 4G/5G, with specific investment targets identified in optical interconnects and switches [3]