Open Science, Closed Models: How Funding Shapes AI in Science

Funding source shapes which AI tools scientists use: Public funding is associated with researchers using open-weight models and off-the-shelf tools, while private funding correlates with scientists modifying and extending foundation models. Industry cloud credits specifically push researchers toward closed, proprietary models rather than open alternatives.

The real divide is control, not access: Across the world, scientists are increasingly adopting AI tools, but the Global South faces growing dependence on API-only access to proprietary models (rising from 0.4% to 10.0% of AI-engaged papers), while China maintains higher rates of model extension through greater public funding—suggesting unequal power dynamics in who controls AI infrastructure.

Corporate influence reaches science through collaboration: Industry involvement in research papers is strongly associated with model extension work, meaning corporate resources and priorities are shaping how scientists engage with AI, though this influence is concentrated in Computer Science and collaborations involving Global North institutions.

Public funding choices have structural consequences: The study of over 104,000 papers reveals that how governments and institutions fund AI research directly determines whether scientists build independent capabilities or become dependent on commercial platforms—a decision with long-term implications for scientific autonomy and equity.