Why AI May Replace Traditional Geopolitical Competition
Or — at minimum — change the asset class that competition is for.
Traditional geopolitical competition has been about territory, energy reserves, populations, and trade routes. Each had a clear unit of account: square kilometers, barrels per day, GDP per capita, shipping tonnage. AI as a strategic resource fits none of those units cleanly. It is closer to a service than a stock, closer to electricity than to oil, closer to a research program than to a finished good. If it becomes the central strategic asset, the form of state competition changes, possibly fundamentally.
Consider what a sovereign-AI claim actually consists of in 2026. (a) Access to frontier-model weights, either via in-country fabrication of those weights or via export licenses from a friendly lab. (b) Sufficient compute infrastructure to train and run those models — typically tens to hundreds of thousands of GPUs in regional clusters. (c) Sufficient domestic technical talent to operate the stack without permanent foreign dependence. (d) A regulatory environment that permits the relevant deployments. None of these can be straightforwardly counted, embargoed, or moved across borders the way oil or grain can.
The consequence is that AI competition is run with different instruments. Export controls (on chips, not on models). Talent visas. Sovereign-investment partnerships with frontier labs (UAE's Mubadala into Anthropic and OpenAI; Saudi NEOM compute deals; Japanese MEXT funding for sovereign models; the 2025 Stargate compute project). Open-weights releases as soft-power instruments (DeepSeek, Qwen, Llama). University-by-university talent recruitment. The instruments are recognizable from past technology races but the asset they target — capability — is more abstract than past targets.
The most consequential possibility is that AI dissolves the boundary between economic competition and military competition. A frontier model in the hands of a productive economy is a continuous productivity multiplier; the same model in the hands of a military is an autonomy multiplier. The same artifact serves both purposes. This is unusual — past dual-use technologies (oil, electricity, computers, the internet) usually required substantial customization for military use. Frontier AI does not.
What does this predict for centers? If the AI-as-strategic-asset thesis holds, the centers that matter most are not necessarily the largest economies — they are the polities that combine compute, talent, regulatory tolerance, and capital availability. That is a different list than the GDP rankings produce. The United States and China dominate by combination. The UAE, Singapore, and Saudi Arabia outperform their GDP weight by virtue of capital and regulatory flexibility. Japan, Korea, and Germany underperform their GDP weight because none has a top-tier frontier lab. Israel and the UK punch above their GDP weight through talent density.
If the thesis fails — if AI turns out to be just another general-purpose technology, like electricity, that diffuses widely and stops conferring strategic advantage after a generation — the existing center pattern of London, New York, Tokyo, Beijing, plus a handful of regional capitals, persists with minor modification. Both scenarios are possible. The 2030s will resolve it.