斯坦福大学研究团队对1998年至2025年间的317家AI独角兽公司进行了调查，发现超过半数的此类公司没有公开发表过学术论文[1]。这项研究表明，尽管AI领域在全球范围内获得巨大关注，但科学成果却高度集中。
研究数据揭示了这一现象的规模。2025年全球AI论文总量超过90万篇，但与这些初创公司相关的论文仅有950篇，占比仅为0.1%[1]。进一步分析表明，排名前5%的公司贡献了超过90%的引用量，其中OpenAI一家公司就贡献了近40%[1]。在132篇高被引论文中，仅有27名高产作者就贡献了近四成署名[1]。
区域差异明显。近三分之二的中国AI独角兽公司曾发表过论文，而美国公司则采取了截然相反的策略，超过半数的美国公司未发表符合条件的论文[1]。通讯作者John Ioannidis对这一现象表达了疑虑，称"对于一个声称正在重塑科学、并被认为具有巨大科学潜力的领域来说，几乎没有科学文献是一个非常奇怪的悖论"[1]。
A research team from Stanford University has published findings examining the scientific output of 317 AI unicorn companies, uncovering stark disparities in how these high-valued startups contribute to peer-reviewed literature.[1] The study, covering the period from 1998 to 2025, reveals that over half of AI unicorns have produced no publicly available research papers, with scientific contributions heavily concentrated among a small number of firms.[1] Across 2025, the global AI research landscape generated over 900,000 papers, yet only 950 papers involved these unicorn startups—representing just 0.1 percent of total output.[1]
The research demonstrates significant regional differences in publication practices.[1] Chinese AI unicorns prove more research-oriented, with nearly two-thirds having published papers, while the majority of their American counterparts have adopted closed-source business models, with over half producing no qualifying publications.[1] Citation influence remains extraordinarily concentrated: the top 5 percent of companies account for over 90 percent of all citations, with OpenAI alone contributing nearly 40 percent.[1] Among 132 highly-cited papers, just 27 prolific authors generated close to 40 percent of all authorships.[1]
John Ioannidis, the corresponding author, highlighted the paradox underlying these findings, stating: "For a field claiming to reshape science and widely regarded as possessing enormous scientific potential, having almost no scientific literature is a very strange contradiction."[1]
