The annual NeurIPS conference is happening this week in San Diego, with satellites taking place in Mexico City, Copenhagen and beyond. The AI World team analysed 100% of the accepted papers from the OpenReview platform to show the affiliated institutes of authors featured at NeurIPS. While many prestigious institutions and companies are represented, cutting-edge AI research is now increasingly being shaped in three hubs: Beijing, Shanghai and San Francisco.
Typically, much of the spotlight falls on the extraordinary AI investments flowing into Silicon Valley, even though China is widely recognised as a major AI research player. The NeurIPS paper analysis largely confirms this pattern. Chinese and US institutions account for a substantial portion/share of authorships, with the rest being split between other Asian, European and Canadian institutions and companies. For instance, the three best papers were led by researchers from Qwen, Princeton and the University of Washington respectively.
Beyond China and the United States however, several other institutions stand out, such as the Mohamen bin Zayed University of Artificial Intelligence (established in 2019 in Abu Dhabi, UAE), Singapore´s Nanyang Technological University (NTU) and National University of Singapore (NUS), and the Korea Advanced Institute of Science & Technology (KAIST) from South Korea.
New this year is the position paper track, which welcomes submissions that reflect the field's broader impact on society. This highlights a small but important shift, showing that alongside technical breakthroughs, NeurIPS is providing more space for researchers to critically assess how AI systems reshape economies, institutions and everyday life. Many of the leading authors/ researchers now combine academic roles with research positions in major AI labs, blurring the line between AI research and applied development. The visualization shows the unique active affiliations per paper, which can include academic and industry affiliation for a unique author.
Our analysis builds on previous efforts to map the geography and structures of NeurIPS, most notably the work done by Macro Polo, as well as last year's assessment, which focused mainly on spotlight and oral papers.
Combined, these developments suggest that NeurIPS is becoming not just a leaderboard of models, but a mirror of the global AI ecosystem itself. With new tracks addressing societal impact and a growing number of double-affiliated authors, the conference increasingly reflects the directions in which AI research is going.