Only One Poor Country Has a National AI Strategy

As of mid-2026, more than 80 countries have published a national AI strategy, a government plan for how artificial intelligence will be regulated, adopted and turned to national advantage. The World Bank’s World Development Report 2026 adds the sentence that turns this from a statistic into a problem: of the world’s 25 low-income countries, exactly one has done so, Rwanda. Among high-income countries, more than half have. The report’s own summary is hard to improve on: the countries with the most to gain appear today to be the least prepared.

The gap deserves attention precisely because AI is the rare technology whose benefits do not require building anything. No poor country needed to invent the mobile phone to be transformed by it, and none needs to train a frontier model to use one. What a strategy represents is not laboratories or chips but decisions: what the schools teach, what the state digitizes, what data rules apply, and which public services get rebuilt around tools that already exist and mostly cost little. Twenty-four of the twenty-five poorest countries have not yet made those decisions on paper.

The Preparation Gap in One Picture

A fair objection arrives immediately: a strategy is a document, and documents do not run schools or power grids. The report’s answer is that the document is a leading indicator of everything that does. Rwanda’s strategy is already visible as AI-awareness modules in secondary schools, folded into a broader digital transformation program. A published plan assigns ministries, budgets and timelines; its absence usually means the questions have not been asked. The pattern echoes every previous technology wave, in which the gap that mattered was never invention but adoption, the same human-capital machinery we set out in our explainer on skills, education and health.

Why the Stakes Are Higher for the Poorest

The reason preparation matters more at the bottom of the income distribution is that AI’s plausible gains there are larger, not smaller. Low-income economies run on scarce expertise: too few doctors, agronomists, teachers and engineers per person. AI’s economic character is precisely the mass reproduction of expertise, which is why the report argues that developing countries should ignore the rich-world debate, dominated by job-loss fears on one side and the race to build frontier models on the other, and treat AI as what it is for them: the cheapest productivity tool ever offered, the mechanism our piece on output per hour and living standards shows compounding into everything else.

The report’s prescription is deliberately unglamorous: adopt, adapt, and only then advance. Do not build models; deploy them into agriculture extension, public health triage, teacher support and government paperwork, and concentrate public spending on the constraints that gate adoption, electricity, connectivity, data systems and skills. The same logic ran through the labor-market evidence we examined in AI job displacement is no longer a forecast, but with the sign reversed: in economies where the binding constraint is absent expertise rather than expensive labor, the technology substitutes for what does not exist, which is the definition of a windfall.

Table 1. The Divide the Report Describes

Group	National AI strategies	What AI mainly threatens or offers
High-income countries	More than half	Automation of white-collar work; the frontier race
All countries	More than 80	Mixed, dominated by the large middle-income adopters
The 25 low-income countries	One: Rwanda	Mass reproduction of scarce expertise, if adoption is prepared

The Familiar Shape of a New Divide

What makes the 1-of-25 statistic uncomfortable is its familiarity. Every general-purpose technology has produced the same geometry: the economies best placed to absorb it move first, compound their advantage, and the distance grows before it shrinks. The countries in Figure 1’s empty circles are, in most cases, the same fragile and low-income economies whose incomes have stopped converging at all, the stall we documented in nearly a decade of lost convergence. A technology that raises productivity wherever it is adopted, adopted everywhere except where productivity is lowest, widens exactly the gap that already stopped closing.

None of this is destiny, and the report is explicit that the window is open rather than closed. The inputs a strategy coordinates are buildable: connectivity is spreading, the models themselves are largely free at the margin, and Rwanda demonstrates that a strategy does not require wealth, only decision. The report’s warning is about time. Preparation, in its phrase, has begun in a substantial number of developing countries, but not in the countries that need it most, and each year of that sentence staying true converts a potential windfall into another layer of divergence. For the ordinary citizen of a low-income country the stake is concrete: whether the clinic, the school and the land office they use this decade work better, or whether those improvements arrive everywhere else first.

MASEconomics Explains

Conclusion

One national AI strategy among 25 low-income countries, against more than 80 worldwide and more than half of the rich world, is the cleanest single measure yet published of who is preparing for the decade’s defining technology and who is not. The World Bank’s framing removes the usual excuse: for poor countries the relevant AI is not the frontier model but the free or cheap tool that multiplies scarce expertise, and the binding constraints are electricity, connectivity, data and skills, all of which respond to exactly the coordination a strategy provides.

The risk the number encodes is not that AI harms the countries in the empty circles, but that it bypasses them while compounding everyone else, arriving as a second divide on top of incomes that have already stopped converging. Rwanda’s existence in the filled circle is the counter-argument to fatalism: the entry ticket is a decision, not a budget. The report’s sentence stands as the decade’s quiet warning, that the countries with the most to gain are currently the least prepared, and every year it remains true it becomes more expensive.

Frequently Asked Questions

Thanks for reading! The cheapest development tool ever offered is waiting on twenty-four unwritten documents. Happy learning with MASEconomics