启明创投周志峰：大模型公司没有价值吗？今年AI应用将全面落地

Core Insights - Many investors believe that large model companies lack value, but the company disagrees, emphasizing that enterprises aim to create significant value through applications rather than just model development [1] - Qiming Venture Partners has invested over 10 billion RMB in more than 80 AI projects since 2013, with over 20 of these becoming publicly listed companies or unicorns [1] - The company has invested in 14 leading enterprises in large language models, multimodal models, embodied intelligence models, or end-to-end autonomous driving models [1] - The company predicts that by 2025, AI applications will fully materialize, marking a significant turning point in the application of AI technology [1][2] Investment Strategy - Qiming Venture Partners categorizes AI investment into three dimensions: infrastructure layer, model layer, and application layer [2] - The company believes that today's so-called "model companies" will ultimately become application enterprises, which is the significance of investing in model innovation [2] - It is estimated that 99% of enterprises in the AI market will be application companies, capturing 70%-80% of the value generated by the AI technology wave [2] Technological Evolution - The company highlights that any technological wave begins with foundational technology, focusing on two core indicators: performance and cost [2] - AI usage costs have shown a significant decline trend of over 100 times annually, and both performance and cost are expected to reach critical thresholds by 2025, leading to an explosion in applications [2] - The release of OpenAI's models has demonstrated significant advancements, with the o1 model scoring 120 on the Mensa test, surpassing 75% of the human population, and the o4 model scoring 140, nearing genius levels [3] Future Outlook - The company plans to continue its layout in the infrastructure, model, and application layers of AI, believing that AI products and services are gradually transforming people's lives [3] - The transition from single tools to platform products in AI applications is anticipated to be a crucial milestone in 2025 [2]