原文 | Nowfal 编译 | 王启隆 出品丨AI 科技大本营（ID：rgznai100） 在科技行业的舆论场中，常能听到一句带着几分轻蔑的评价："这不就是个套壳 AI 吗？" 对于那些正在绞尽脑汁试图构建新事物的开发者而言，这句话听起来格外刺耳。它宛如一盆冷水，径 直泼在刚刚燃起的创新火苗之上。这句话背后的潜台词极具杀伤力： 这意味着缺乏核心技术，意味 着只是在巨人的地基上搭建积木，意味着随时可能被推倒的脆弱性。 然而，反击的声音同样响亮，且逻辑坚硬得令人无法反驳。 Perplexity 的首席执行官 Aravind Srinivas 曾直言不讳：" 世上万物皆是套壳 （ Everything is a wrapper ）。 OpenAI 套的是英伟达的算力和 Azure 的云服务；Netflix 套的是 AWS 的基础设 施；就连市值高达 3200 亿美元的 Salesforce，归根结底也不过是 Oracle 数据库的一个高级外 壳。"你 此言确实切中肯綮。但在深入这场关于"定义"的口水战之前，有必要先厘清公众口中的这个"AI 套 壳"（AI Wrapper）究竟是何种物种。 简而言之，这往往是一个被 ...

Core Insights - A recent analysis reveals that 73% of 200 AI startups are essentially "shell companies," relying heavily on third-party services like ChatGPT and Claude for their core functionalities [1][5][40] - The findings challenge the perception of innovation in the AI startup ecosystem, highlighting a significant gap between marketing claims and actual technological capabilities [1][5][40] Group 1: Investigation Background - The investigation was initiated after a software engineer discovered discrepancies in a startup's claims about its proprietary technology, which was found to be dependent on OpenAI's API [2][5] - The engineer aimed to provide a data-driven analysis rather than subjective opinions, leading to a comprehensive examination of 200 AI startups [3][5] Group 2: Methodology - The analysis involved reverse engineering and monitoring API calls, as well as examining the marketing claims against actual technology stacks [4][5] - The focus was on startups that had received external funding and claimed to possess "exclusive technology," excluding those less than six months old [4][5] Group 3: Findings - The investigation categorized the startups into three models based on their actual technology use: - **Model 1**: Companies claiming to have proprietary models but actually using GPT-4 with minimal modifications [8][10] - **Model 2**: Startups utilizing Retrieval-Augmented Generation (RAG) techniques but misrepresenting their technology as proprietary [15][19] - **Model 3**: Companies claiming to fine-tune their models, but often just using OpenAI's services without significant innovation [22][25] Group 4: Economic Implications - Many of the startups operate with high profit margins, often charging significantly more than their actual costs for API usage, leading to questions about their business models [19][21] - The analysis suggests that while the profit margins are substantial, the lack of genuine innovation raises concerns about the sustainability of these business practices [19][21] Group 5: Recommendations for Stakeholders - For founders: Transparency about technology stacks is crucial, and companies should avoid misleading claims about their capabilities [45][46] - For investors: Understanding the true nature of AI startups is essential, as many are service-oriented companies relying on API costs rather than employee costs [46][51] - The market is expected to mature, with a potential shift towards rewarding transparency and genuine innovation [47][48]