The AI Power Gap Is Moving From Models to Adoption

The first phase of the AI power race was easy to photograph: chips, data centres, sovereign funds and enormous capital commitments. The next phase is less visible. It concerns whether firms adopt the technology, whether workers use it productively and whether infrastructure produces economic returns outside a small group of technology companies.

independent specialist publication's research network increasingly treats infrastructure, company adoption and project delivery as separate layers rather than assuming that capital expenditure automatically equals economic transformation. That is a useful framework for political power because governments ultimately need diffusion, not just flagship assets.

Adoption changes the distribution of gains

The OECD repeated its employer and worker AI surveys in 2026 with wider country and sector coverage, extending earlier work on why firms implement AI and how workers experience it. IMF research using observed AI-usage data similarly finds that gains are unevenly distributed across occupations and countries, with high-wage occupations capturing a disproportionate share of usage-based value.

Those findings complicate the idea that national AI strength can be measured by model rankings or installed compute alone. An economy can possess infrastructure while leaving much of its business base unable to exploit it.

Infrastructure still matters, but it is not sufficient

The World Economic Forum's review of national AI investment strategies describes different approaches to building domestic capability, from broad US and Chinese investment across the value chain to more selective strategies in European economies and government-led ecosystem building in places such as the UAE.

The political question is what each strategy makes possible. Compute can reduce dependence and attract firms. Skills can spread productivity. Procurement can create demand. Regulation can either clarify deployment or increase uncertainty. Power emerges from the interaction of these systems rather than from any one investment announcement.

The next scoreboard should measure diffusion

Governments increasingly have incentives to advertise capacity, funding and national champions. A more revealing scoreboard would track business adoption, measurable returns, workforce use, access to compute and the concentration of benefits. The AI race is becoming a contest over institutional ability to turn technology into broadly usable capability.