This week, the World Bank published its World Development Report 2026, with an unusually blunt message to developing countries. "AI has thrown developing economies a lifeline, and they should seize it," writes chief economist Indermit Gill. Developing economies in the midst of their weakest average growth performance in three decades, and the Bank argues that AI, applied to healthcare, education, justice, and agriculture, could compress decades of service delivery into years. By one estimate cited in coverage of the report, AI adoption could help sub-Saharan Africa achieve in ten years what would otherwise take a century.

We generally agree with the diagnosis. Our concern, however, is with the prescription.

Developing countries, the report notes, "do not need large models or big data centres to reap its benefits." That's true, and it's a useful corrective to the idea that every nation must burn billions racing toward frontier models. But it can also harden into a policy of permanent dependency: adopt what others build, forever. A lifeline you don't hold both ends of is a leash.

The gap is structural

The report identifies AI's productivity potential at more than 16% of jobs in developing economies against more than 18% in advanced ones. Two points is not much on its own, but compounding annually against economies already growing at their slowest rate in three decades it certainly is.. That gap sounds small until you remember it compounds every year. Meanwhile, the frontier models most broadly adopted are overwhelmingly built in the United States and China, trained mostly on the languages, data, and assumptions of the countries that built them. In sub-Saharan Africa, roughly a third of schools lack electricity and two-thirds lack internet access. The IMF estimates AI will affect 60% of jobs in advanced economies and 40% globally.

Report co-author Gaurav Nayyar frames the risk directly: "This is a golden opportunity, and if developing countries don't put the policies in place… there is a risk that they will fall behind." 

So the question isn't whether to adopt AI. The Bank's three-step ladder (adopt available tools, adapt them to local conditions, advance toward homegrown capability) is the right shape. The question is what keeps a country climbing that ladder instead of getting stuck on the first rung, renting intelligence priced,, governed, and trained somewhere else.

What Sovereign AI truly means

Sovereign AI is often caricatured as techno-nationalism, with every mid-sized economy building its own frontier lab. That's not what it means, and the report itself points to the most useful version: the UAE's Falcon models, trained on local languages and dialects, show what "adapt and advance" looks like when a country decides it wants agency in the stack rather than just access to it.

Advising governments on this, we've landed on a working definition. Sovereignty is not owning every layer. Sovereignty is enforceable agency at the layers that matter

Data and identity. Who your citizens are, and what is known about them, is the most sensitive layer of the stack. A national AI strategy that routes citizen verification and citizen data through foreign platforms hasn't adopted AI as much as it has been adopted by it.

Compute access. If intelligence becomes a factor of production like electricity, then who can afford and access it becomes a distributional question, not just an infrastructure one.

Rules and values. Models make judgment calls. Whose language, law, and social context those judgments encode is a sovereignty question, whether or not anyone frames it that way.

What we've been building toward

At Self Labs, we've spent the past years building privacy-preserving identity infrastructure. Self leverages zero-knowledge verification against government-issued credentials like biometric passports, national IDs, and Aadhaar, so a person can prove they are real, unique, and eligible without surrendering their data to anyone, Self Labs included. Credentials live on the citizen's device, not stored on a server that is primed for data breaches. 

That architecture is exactly what sovereign AI needs at its foundation. Governments get verifiable proof of personhood for public programs without building centralized biometric databases, while citizens get privacy by design rather than privacy by promise.

That foundation is already in production, used by millions and trusted by industry leaders. It powers proof-of-human verification for programs run by Google, Opera, Aave, and others, and last month we launched a dedicated Sovereign Workspace for these kinds of government-scale use cases.

It's also why we've been advising governments on sovereign AI strategy directly, helping policymakers translate "adopt, adapt, advance" into concrete institutional choices about identity, data, and compute.

Why we're particularly focused on Universal Basic Compute (UBC) programs for the same reasons. As AI becomes a general-purpose input to economic life, a baseline entitlement of compute for citizens, students, small businesses, and public institutions should be part of a nation's social and economic infrastructure, the way universal electrification and universal schooling once were. UBC is where sovereignty becomes tangible for ordinary people. It's the difference between AI as something that happens to a population and AI as something a population is equipped to use. It also depends entirely on the infrastructure described above, becauseyou cannot distribute a per-person entitlement fairly without a privacy-preserving, sybil-resistant way to prove personhood. The system must offer one allocation per person, without bots or surveillance.

The narrow window

The World Bank is right that this is a golden opportunity, and right that the window is narrow. But windows like this one don't close evenly. They close differently for countries that planned for agency and countries that didn't. Countries that treat this moment as a procurement exercise will get tools. The nations that treat it as a sovereignty exercise, building the identity, data, and compute foundations to adopt today, adapt tomorrow, and advance on their own terms, will get capability.

Self Labs advises governments and organizations on sovereign AI strategy and Universal Basic Compute programs, and builds the privacy-preserving identity infrastructure that makes them work. If you're a policymaker thinking through your country's AI strategy, we'd love to talk.