AI's Top Startups Have Gone Quiet: Half Have Never Published a Paper

A new analysis of the AI startup landscape finds that peer-reviewed research has become the exception, not the rule. More than half of the companies studied have never produced a single qualifying paper, and scientific output is extraordinarily concentrated: the top 5% of firms account for over 90% of all citations. OpenAI alone is responsible for nearly 40% of citations in the dataset, trailed by Chinese computer-vision firm Megvii and the model-hosting platform Hugging Face.

The drop-off holds even at the largest and best-funded labs. Despite a headcount of roughly 4,500, OpenAI has only eight researchers who have authored five or more qualifying papers — a striking ratio for a company built on frontier research. The pattern reflects a deliberate shift away from the open-publication norms that defined academic and early-industry AI work.

The economics explain the silence. Unlike pharmaceutical firms, which can publish discoveries while protecting them with patents, AI companies capture little competitive value from disclosing technical advances and much of the risk — handing rivals a roadmap. As commercial stakes have risen, secrecy has become the rational default, raising concerns about reproducibility, independent safety scrutiny, and the field’s ability to build on shared knowledge.