Major AI Startups Rarely Publish Research Results

According to an analysis by Professor John Ioannidis's team at Stanford University, more than 50% of AI unicorn companies (valued at $1 billion or more) have not published a single research paper or preprint.

Among AI-related papers published in 2025, papers in which these unicorn companies played a leading role accounted for only about 1 in 1,000 of the total. This phenomenon runs counter to academic norms that emphasize scientific validation and reproducibility.

Experts offer the following perspectives on this phenomenon:

• Difficulty of scientific validation: Because research results are not disclosed, it is difficult to objectively assess the effectiveness, energy usage, and safety of the technology.
• Commercial incentives: The analysis suggests that companies' goals lie in generating profit rather than advancing science, and that keeping technology proprietary rather than disclosing it is more advantageous for their business models.

This study covered 317 AI unicorn companies that existed from 1998 to 2025, and was based on data on papers in which company researchers played a leading role.