Skip to content
MeetMrStewart.com
Perspectives

PERSPECTIVE | AI, ECONOMIC DEVELOPMENT AND WORKFORCE READINESS

Perspective No. 3 · Economic Development

Sovereign AI Buys the Machines. Who Builds the People?

Building the People Layer of National AI Policy

Download this Perspective (PDF)

The money is moving first, and it is moving to very few places.

Stanford's AI Index Report 2025 puts corporate AI investment at $252.3 billion in 2024. Private investment in the United States reached $109.1 billion, nearly 12 times China's $9.3 billion. Governments have read those numbers as a call to build. Sovereign AI programs now promise national compute, national models, and national data centers, and the announcements compete on scale.

Preparedness is not keeping the same pace. The International Monetary Fund's AI Preparedness Index tracks 174 economies across four areas: digital infrastructure; human capital and labor market policies; innovation and economic integration; and regulation. Its central finding is uncomfortable. Wealthier economies tend to be better equipped for AI adoption than low-income countries, which means the places with the least capital also start with the weakest foundations.

Using AI turns out to be a different contest from building it. Visual Capitalist, charting Microsoft's estimates of the share of the working-age population engaging with AI for at least 90 minutes a month, puts the United Arab Emirates first at 70.1% in the first quarter of 2026 and Singapore second at 63.4%. The United States, home to most of the investment, ranks 21st at 31.3%.

Three findings, one pattern. The machines concentrate. The foundations diverge. The use spreads where institutions carry it. That pattern is the real subject of sovereign AI, and it is mostly a workforce subject.

Sovereignty Is Usually Defined Too Narrowly

Sovereignty, as most national strategies use the word, means ownership. Own the data centers. Own the compute. Own the models, or at least host them under national law. Those are real assets, and the concern behind them is legitimate. A country that depends entirely on foreign infrastructure accepts real risk.

But ownership is the part of the strategy money can buy. The part money can't buy outright is the people layer: the technicians who keep a data center running, the engineers who adapt models to local language and law, the teachers who make the tools usable in classrooms, the public employees who decide how the systems serve residents, the founders who build companies on top of the capacity. The pattern runs through the work itself. The construction jobs a buildout announces are real, and they end. The operating jobs, held by the technicians, maintainers, and teachers, are the ones that stay.

Every country controls that layer in a way it doesn't control global capital. Investment can be courted, refused, or lost to a better offer. Models built elsewhere can be licensed, restricted, or repriced. A population's capability is built at home, by domestic institutions, over years. It is slower than a procurement cycle and harder to announce. It is also the only part of a sovereign AI strategy that can't be taken away by someone else's decision.

So the sharper definition is this: a country is sovereign in AI to the extent that its own people can build with it, run it, question it, and benefit from it. By that definition, many strategies that call themselves sovereign are only well equipped.

Starting Points Differ, So Provision Must

The IMF index is useful precisely because it separates the foundations. A country can be strong on infrastructure and exposed on human capital, strong on innovation and thin on regulation. The same unevenness runs through its workforce and its institutions. An expert in hospital administration may be new to AI. A ministry may be expert at procurement and new at evaluating model risk. Starting points differ by subject, for people and for governments alike.

That is why buildouts that buy capacity first and design provision later produce a familiar result: impressive infrastructure, a thin layer of specialists able to use it, and a wide population watching from outside. The readiness levels and the learning experiences proposed earlier in this series are design tools against exactly that outcome, and a sovereign buildout is where they earn their keep.

Capacity Without Pathways Is Just Inventory

A sovereign buildout creates a provision question that most announcements leave unanswered. Who gets to use the new capacity, and how do they learn to?

Compute granted to a national laboratory doesn't reach a community college instructor by itself. A model trained on national data doesn't teach a small business owner anything by itself. Between the asset and the citizen there has to be a designed pathway, and pathways are workforce policy.

The learning experiences proposed in this series, from guided exploration to self-paced study, exist for exactly this junction: they design the pathway at the same time as the asset, in the formats different people actually use. It is a proposal, not an existing program, and it would have to earn its standing through design, testing, and evidence. Their value for sovereign AI policy is practical. They force the provision question early, while the data centers are still on paper, instead of after the ribbon is cut and the capacity sits idle.

The politics around these projects has changed as well, on several continents at once, and it is worth reading clearly, because it is not a politics of rejection. Ireland, the Netherlands, and Singapore have each, in different ways, slowed or conditioned new data center approvals over power, land, and sustainability. The United States' elections in November 2026 told the same story in a different system: support for AI held across parties, while unconditional support for building it disappeared, and the winning positions converged on practical conditions. The builder pays its full share of grid costs. Household ratepayers are protected. The host community keeps a real say. Agreements are negotiated in public rather than under secrecy clauses. Water is stewarded. And the workforce benefit is named and local. Different countries, different politics, one pattern. A sovereign strategy that treats those conditions as obstacles will keep colliding with them. A strategy that treats them as the terms of a durable welcome builds the one asset no announcement can buy: a public that believes the project is also theirs.

Each Institution Holds One Part of the People Layer

No single institution can build the people layer, and none should be asked to. The work divides naturally.

National governments set the terms. They decide whether publicly supported AI investment carries any obligation to develop domestic capability, and they fund the foundations the IMF measures: infrastructure, education, labor market policy, and trustworthy regulation.

Local and regional governments know where the people are. They can connect national capacity to actual employers, schools, and neighborhoods, and they hear first when a program is missing the residents it was meant to serve. They also hold many of the permits, and increasingly they negotiate the conditions, on grid cost, water, and local workforce benefit, under which a project proceeds.

Economic development organizations can do for workforce what they already do for capital: identify demand, convene the parties, and ask that an investment announcement arrive with its people chapter alongside its construction chapter.

Universities and community colleges convert capacity into capability. They are where national infrastructure becomes courses, credentials, applied research, and graduates.

Industry partners hold the current knowledge: which skills the work requires this year, which tools are standard, which roles are changing. Their contribution is currency, not just capital.

Community organizations reach the people who don't see themselves in technology announcements. Where they are absent, adoption programs reliably serve the already connected and call the result diffusion.

The workforce chapter has to be written by the institutions that will have to live it, at the same time as the infrastructure chapter, not after it.

What Sovereignty Should Be Measured By

Data centers will keep getting announced, and the investment figures will keep concentrating. Neither trend sits within the control of most governments, and neither is the right scoreboard.

The scoreboard a government does control is capability. How many of its technicians can run the new facilities. How many of its teachers can teach with the new tools. How many of its small businesses use the national capacity for something that matters. How many of its residents moved from new to capable, and from capable to expert, while the machines were being installed.

A sovereign AI strategy that builds those numbers has done something no procurement can undo. A strategy that doesn't has purchased equipment and called it independence.

TRANSLATION

Policy priorityPractical first actionEvidence of progress
Workforce chapter in AI investmentA publicly supported sovereign AI project arrives with a published workforce and access annex, covering the roles it will need, the training partners who will prepare people for them, and the routes by which the public can use the capacity, with workforce starting points baselined by sector and occupationAnnex template issued; share of approved projects filing one
Utilization of public capacityRequire publicly funded AI facilities and compute programs to report actual use, by type of user, within their first year of operationFirst utilization reports published
Educator and public-sector capabilityFund a first training cohort for teachers and for public employees in the agencies adopting AI systemsCohort completion; share of adopting agencies with trained staff
Access pathwaysOpen a published application window through which community colleges, small firms, and independent researchers can use national AI capacityApplications received and awards made, by institution type

Buying the machines is the purchasable part of sovereignty. The durable part is a population that can build with them, run them, and benefit from them, and that part is built by policy or not at all.

Sources

Download this Perspective (PDF)