When Do AI Gains Become Broadly Shareable? A Policy Threshold for AI-Driven Automation

Public capture of AI gains matters as much as technical capability: The model shows that increasing public capture from low levels (33%) can substitute for substantial AI capability improvements, meaning policy choices about who benefits from AI are as important as the technology itself.

Market structure and deployment costs are critical barriers: The paper finds that capability alone is not decisive—factors like market concentration, deployment expenses, and safety costs jointly determine whether AI benefits can be widely shared, suggesting that technical breakthroughs won't automatically translate to broad social benefit.

Moderate public capture is more efficient than full nationalization: Moving from 33% to moderate public capture yields significant gains in shareable benefits, but pushing toward full government nationalization shows diminishing returns, particularly when deployment or safety costs are high.

Competition policy has trade-offs worth considering: Opening concentrated AI markets can improve fairness and resilience, but may actually reduce the total pool of gains available for public benefit unless governments build alternative institutions to capture value through other means.

Different tax and ownership models offer alternative paths: Countries with tax-heavy systems and those using public-asset models (like Singapore and Abu Dhabi) can both effectively capture AI gains—through revenue collection or through direct ownership and investment returns—suggesting multiple policy approaches can work depending on national context.