The distribution of scientific power in the age of AI

Access to frontier artificial intelligence systems is becoming a decisive input into scientific discovery, yet that access is distributed unevenly across institutions and countries. We develop a proportional credit-sharing model, a Tullock contest, in which research laboratories compete for scientific credit, and a subset gain access to AI tools that raise their productivity. We characterise three results: AI access raises aggregate discovery; it concentrates credit and resources in AI-enabled laboratories through a self-reinforcing feedback loop; and the gap can be partially offset by equalising interventions that provide baseline AI tools to all laboratories at a bounded loss in aggregate output. A welfare analysis that incorporates the social cost of intervention characterises the constrained-optimal policy. These results sharpen under winner-takes-most credit allocation, where concentration is most severe and, under complete information, weaker laboratories may rationally withdraw from frontier priority races. Our framework reframes AI as scientific infrastructure: governance decisions about who can deploy frontier AI systems are decisions about the future distribution of scientific power.