The Compute Divide as an Access-Countercapacity Gap

Advanced artificial intelligence is often discussed through model capability, productivity, safety, or regulation. This exploratory report examines a more infrastructural problem: the compute divide. It argues that the compute divide is best understood as the gap between AI access and AI countercapacity: the difference between being able to use AI systems and being able to independently scrutinize, reproduce, regulate, or contest them. Drawing on selected academic, policy, technical, and institutional sources, the report maps infrastructure concentration across compute, cloud platforms, advanced chips, data centers, energy, model access, transparency, and evaluation environments. The strongest current evidence concerns concentrated AI infrastructure, cloud and chip dependencies, rapid data-center electricity growth, transparency deficits in foundation models, and uneven global AI readiness. The strongest present-day governance risk is a preparedness gap: AI adoption is advancing faster than many public-interest actors' capacity to independently inspect or challenge high-impact systems. Open-weight models, smaller models, algorithmic efficiency, public compute, structured access, competition policy, and regulatory expertise are real counterforces, but they mitigate different layers of the divide unevenly. The report concludes that public compute, audit rights, transparency duties, regulatory capacity, competition remedies, and democratic contestation mechanisms should be understood as attempts to build public-interest countercapacity against an emerging inequality in the means of AI-enabled knowledge production.