The Eight Worlds of AI Geopolitics: A Strategic Framework

If you could map the future of AI geopolitics across eight possible worlds, which one would you bet on?

Former U.S. National Security Advisor Jake Sullivan has introduced a provocative new framework that challenges how we think about AI strategy. Writing from Harvard Kennedy School, Sullivan argues that Washington doesn't need another prediction about the AI age—it needs a way to make smart choices under radical uncertainty.

His solution? A 2×2×2 matrix that considers three fundamental questions: Will AI progress toward superintelligence or plateau? Will breakthroughs be easy to copy or hard to replicate? And is China truly racing for the frontier, or playing a different game entirely?

The Hidden Assumptions Behind AI Policy

Sullivan's central insight cuts through the noise of AI debates: "Every confident policy argument rests on hidden assumptions." Those pushing massive investments in frontier research assume breakthroughs will compound and be difficult to replicate. Those focused on spreading American AI systems abroad often assume the opposite.

The stakes couldn't be higher. If these assumptions are wrong, strategies built on them will waste resources and could cost America its technological edge. OpenAI's multi-billion-dollar training runs, Google's quantum computing investments, and the Biden administration's export controls all rest on specific beliefs about how the AI future will unfold.

Eight Worlds, Three Dimensions

Sullivan's framework turns on three axes. The first is the nature of AI progress itself. At one extreme lies superintelligence—AI that far outpaces humans and can recursively improve itself. At the other lies "bounded and jagged intelligence"—impressive but limited, reaching incredible performance in math or coding while struggling with judgment and creativity.

The second axis examines how easily rivals can catch up. In one world, breakthroughs spread quickly through espionage, leaked model weights, or clever training techniques. In another, frontier capability depends on the full technological stack—proprietary hardware, institutional expertise, vast datasets, and talent ecosystems that can't be easily replicated.

The third dimension focuses on China's strategy. Is Beijing racing aggressively to the frontier with massive training runs and competing labs? Or is it prioritizing deployment of existing technology while waiting to reverse-engineer American innovations later?

The Fast-Follow Problem

The "fast-follow" question deserves special attention. Silicon Valley's biggest fear isn't that China will innovate—it's that Beijing will let American companies burn through billions in R&D costs, then quickly copy the results at a fraction of the price.

This isn't just theoretical. China has already demonstrated sophisticated model distillation techniques, where smaller systems learn to mimic more advanced ones. If catching up proves easy, the race becomes less about breakthrough research and more about embedding American systems globally before rivals can spread their own.

Policy Tools in an Uncertain World

Sullivan acknowledges a paradox: The U.S. government doesn't own leading AI labs or set production targets like Beijing can. Yet Washington's influence runs deeper than direct control.

Export controls and investment restrictions have effectively subsidized the domestic AI industry by constraining competitors and channeling private capital toward American firms. When senior officials describe AI leadership as a national priority, companies and investors anticipate favorable regulations and closer government coordination—expectations that influence risk-taking and investment flows perhaps more than congressional appropriations.

Beyond Binary Thinking

The framework's power lies not in predicting which world will emerge, but in forcing policymakers to confront their assumptions. Reality will likely blend elements from multiple scenarios. China may pursue a middle path in frontier R&D. Catching up may be only somewhat difficult. AI may be transformative but still bounded by certain limitations.

Sullivan suggests thinking probabilistically along each axis. A partial Chinese investment strategy, for instance, increases the odds that Beijing could unexpectedly close the gap with the United States.