14. A Future Large Enough for Small Places

The map has more than three capitals

A schoolteacher in a small island state wants help preparing lessons. A farmer wants earlier warning of a crop problem. A business owner wants to reach customers abroad. A public servant wants to find the relevant provision in a room full of records. None needs to win a global race to build the largest AI model. Each needs a useful capability on terms that leave room to act.

These are imagined examples, but they show what is lost when the debate centers only on major powers and companies able to finance enormous infrastructure. That framing assigns everyone else the role of market, data supplier, investment site, or population to be protected. Small states also need a place in deciding what the technology should do. The Caribbean roadmap discussed below is one attempt to claim that place.

Looking from Saint Lucia changes the question. What would count as success for a society that does not expect to manufacture the world’s most advanced chips or finance the largest training runs? Surely the answer cannot be permanent dependence followed by gratitude for whatever arrives. Neither can it be the expensive imitation of every project undertaken by a much larger economy.

The meaningful objective is greater practical agency: the ability to identify local problems, obtain useful tools, understand their limits, shape their application, and change course when the arrangement fails. This is a demanding objective, but it is more relevant than technological prestige. A modest system that helps a community solve a persistent problem may increase its freedom more than a nationally branded system that nobody can maintain.

The Caribbean does not lack a policy conversation of its own. UNESCO’s 2024 Caribbean Artificial Intelligence Policy Roadmap, prepared by Erica Simmons and Andrea M. Davis, brings culture and creativity, governance, education, and resilience into the same discussion. It advocates regional cooperation and attention to the region’s specific circumstances. It is a policy roadmap, not evidence that every recommendation has become law or practice in every Caribbean country.¹

The distinction matters. A document can make a future thinkable without making it funded, staffed, or enforceable. The distance between a regional aspiration and a functioning local institution is precisely where political responsibility begins. Someone must decide what to attempt first, what it will cost to sustain, and which existing task will receive fewer resources if the new one proceeds.

The same realism should shape international ambition. The Antigua and Barbuda Agenda for Small Island Developing States identifies science, technology, innovation, and digitalization as areas of action within a wider development agenda. Its significance here is that technological capability belongs inside development and resilience, alongside other needs, rather than above them as an independent contest for prestige.²

For a small state, opportunity cost is an ethical question. Money spent on a poorly chosen technology cannot simultaneously maintain a school, strengthen a public office, or repair essential infrastructure. Yet declining useful technology also has a cost. If a carefully selected tool allows a small team to perform work previously beyond its reach, refusing it can preserve dependence just as surely as buying badly can create dependence.

The choice is therefore practical and selective. Which use expands what people can accomplish? Which use makes an institution more competent? Which use creates a new obligation without providing the ability to meet it? A serious development strategy should be able to answer in the language of services, skills, and choices available to citizens.

Independence is the ability to change an arrangement

Digital sovereignty is often imagined as ownership: local servers, a national model, a domestic platform. Ownership can matter. But a country can own equipment while depending on outsiders to operate, repair, update, or explain it. It can also use foreign infrastructure while retaining substantial control through sound contracts, capable staff, accessible records, and credible alternatives.

The practical test is what happens when the relationship becomes difficult.

Suppose a ministry obtains an attractive cloud service. At first the price is low and support is attentive. Over time its records, procedures, and staff habits become tied to the platform. Then the price rises, an important feature changes, or the supplier withdraws support. Can the ministry export its information in a usable form? Can another provider understand the documentation? Can the public still obtain essential services during a transition? If the answer is no, the original discount bought dependence.

This is a hypothetical pattern, not an allegation about any particular Caribbean contract. Its logic applies wherever switching costs grow after a purchase. The government’s bargaining position can weaken as the service becomes indispensable and the cost of switching rises. Exit therefore belongs in procurement, not in the emergency meeting held after a relationship breaks down.

An exit clause alone is insufficient. A right to receive data is weak if the export is unreadable, incomplete, or prohibitively expensive to use. A promise of transition support is weak if no local team knows what questions to ask. A second vendor is not an alternative if it relies on the same underlying service and fails at the same moment. The arrangement should be tested through an actual export, a recovery exercise, and an honest account of shared dependencies.

These tests involve costs. Maintaining several fully equivalent systems may be wasteful. Local hosting may offer control while creating security and maintenance burdens beyond a small institution’s capacity. Foreign hosting may offer reliability while introducing legal, contractual, and geopolitical dependence. There is no universal location that settles these questions. The choice should follow the sensitivity of the information, continuity requirements, applicable law, costs, and the competence available to operate the system.

For especially important functions, the goal may be a simpler fallback rather than a duplicate of every advanced feature. An agency may need a protected copy of its records, an alternative communication channel, and a procedure for urgent cases during an outage. A school may need lessons that remain usable when a subscription expires. A public authority should not discover that it has lost the ability to explain its own decisions because its supplier’s interface has changed.

Open software and openly available models can improve options, but openness is not an automatic public service. Someone must evaluate the system, secure it, operate it, and correct failures. A tool available without a licensing fee may still have a substantial total cost. Equally, a proprietary service may be worth paying for when its performance and terms are strong. The important question is whether the choice preserves informed control rather than whether it displays the preferred label.

The same distinction applies to data. More local data does not automatically mean greater local power. Data collected carelessly can become a liability. Data held in an unusable archive can remain politically invisible. Data transferred under weak terms can improve someone else’s product while leaving the originating community without a usable service. The useful sequence begins with a purpose, identifies the minimum information needed, establishes rights and duties, and funds the people who will maintain it.

Independence should therefore be understood as an ongoing capability. It includes the power to enter a relationship, assess it, renegotiate it, and leave it without surrendering an essential public function. No small state can eliminate every dependency. It can avoid pretending that every dependency is beyond negotiation.

Buying together without becoming the same

A small buyer approaches a global supplier with a disadvantage. The supplier may treat local requirements as exceptions too costly to accommodate. A government may lack enough specialists to examine every proposal. A single public body may have too little work to employ a full team of security, procurement, accessibility, and evaluation experts.

This suggests cooperation. It does not require every island to adopt the same policy or send all its information into one database.

There is an existing regional foundation to build upon. In June 2020, the World Bank approved the Caribbean Digital Transformation Project involving Dominica, Grenada, Saint Lucia, Saint Vincent and the Grenadines, and the OECS Commission. Its stated aims included digital access, skills, public services, and a stronger legal and regulatory environment. Approval and stated aims should not be confused with proof that every intended outcome has been achieved.³

My proposal is more specific: interested states could pool selected buying and evaluation functions while retaining responsibility for local decisions. A shared technical team could investigate suppliers, examine security evidence, design common contract language, and test whether advertised capabilities work under agreed conditions. Each participating government would still decide whether a particular use is lawful and appropriate for its people.

The benefit is partly bargaining power and partly memory. A defect discovered in one jurisdiction should not need to be rediscovered at public expense in every neighboring jurisdiction. A well-negotiated portability clause should be available to the next buyer. A failed pilot should produce a usable lesson rather than disappear with the project team. Cooperation becomes valuable when it lowers the cost of learning.

The team would need constraints of its own. If it could select a single supplier for the whole region without meaningful challenge, it might create a more concentrated dependence than the one it replaced. Membership should not require surrendering every local option. Evaluation methods, conflicts of interest, and reasons for recommendations should be visible. Smaller suppliers should be able to qualify without hiring a firm of consultants simply to understand the process.

Joint procurement also creates a risk of common failure. If everyone buys the same vulnerable service, one problem can disrupt many institutions at once. Some diversity may be worth paying for. The aim should be shared standards for quality and accountability, with enough technical variety and recovery capacity to avoid turning efficiency into fragility.

Local businesses belong in this arrangement as more than ceremonial partners. They can maintain applications, adapt interfaces, train staff, assess local language performance, or build specialized services on top of broader technology. Contracts can require substantive knowledge transfer and documentation rather than a local logo attached to a foreign product. But local ownership should not exempt a supplier from evidence, fairness, or competition. Citizens deserve competent service regardless of where the owner’s passport was issued.

The region’s diaspora may also contribute expertise without becoming an alternative ruling class. A person working abroad can help review a technical proposal or mentor a team. That contribution should strengthen the institution’s ability to decide, not make it dependent on the continuing availability of one well-connected individual. Public systems need continuity beyond friendship and personal goodwill.

International partners should be judged by the same standard. Training that leaves an institution able to maintain a service is different from a demonstration followed by a photograph. A grant that covers a pilot but not the recurring cost may transfer a future liability. Assistance is most valuable when it leaves more competence and choice behind than it found.

Cooperation, properly designed, allows small places to become stronger counterparts. It need not erase their differences. The common work can concern the questions every buyer should ask, while the local work concerns which purposes are worth pursuing and what obligations the service owes to the people who use it.

Learning without surrendering a voice

Imagine that the teacher at the beginning of this chapter receives a powerful lesson-planning assistant. The system can explain a concept, generate exercises, suggest activities, and adapt reading difficulty. Its promise is real. It may reduce preparation time and make useful support available outside school hours. But the educational question cannot end with the number of worksheets produced.

Does the student become better able to reason, or merely better able to obtain a completed answer? Does the teacher gain time for individual attention, or inherit the work of correcting convincing errors? Does the curriculum become more connected to local life, or more standardized around whatever material the model knows best? These are questions for evaluation in a particular setting, not arguments that settle the value of every educational use in advance.

UNESCO’s 2023 guidance on generative AI in education and research advocates an approach centered on human agency, appropriate use, privacy, and the capacity of institutions to validate tools. It treats educational design and human capability as necessary parts of adoption. The guidance is not evidence that a particular classroom product improves learning.⁴

For a Caribbean classroom, a useful test might ask students to compare an AI account of a local issue with records, interviews, and direct observation. A confidently written paragraph about fishing, agriculture, a national celebration, or a neighborhood’s history becomes a claim to investigate. Students learn how to use the tool and how to resist being overawed by its tone.

Such an exercise values knowledge that may be absent from a large model. A fisher, a craftsperson, a farmer, or an elder may know something indispensable that has never appeared in a widely available dataset. The absence of that knowledge from the system does not make it less real. Education should help students connect different forms of evidence without treating either local familiarity or machine fluency as infallible.

Language is part of this responsibility. A system that handles a locally used language or variety poorly may misrepresent meaning, flatten humor, or mistake an ordinary expression for an error. Testing must involve speakers with relevant competence. A culturally familiar voice is not enough; a convincing accent can conceal a weak understanding of what was said.

Community language resources could support better tools, but participation should come with clear terms. What material is being collected? For what purpose? Who may reuse it? Can contributors correct misleading representation? Are commercial benefits shared in any form? These are proposed questions for a fair project, not a claim that one agreement can confer ownership over an entire living language.

Culture also changes. A community should not be represented as a museum exhibit whose only legitimate task is preservation. Young artists may want to experiment with new forms, use generative tools, or combine traditions in ways older authorities dislike. Cultural agency includes the freedom to innovate. Protection should prevent exploitation without freezing people inside an approved picture of authenticity.

The same principle applies to economic opportunity. AI could help a small enterprise prepare a proposal, compare suppliers, improve a website, or communicate with customers. These uses may matter more immediately than grand claims about a national AI industry. Public support can concentrate on practical competence: accurate records, basic security, clear communication, checking outputs, and understanding contractual obligations.

Training should also include the confidence to ask for help and the ability to notice when a tool exceeds its competence. A person who can create an impressive document but cannot verify its central claims has acquired a limited skill. A person who understands a business problem and can combine a tool’s speed with careful judgment has acquired a more durable capability.

A small place becomes less dependent when its people can question, adapt, and create. Access to a system is one part of that freedom. Access to education, meaningful work, trusted institutions, and a public voice determines what people can do with it.

The electricity behind the promise

Digital ambition has physical requirements. A screen does not show the power station, the cooling arrangement, the materials in the hardware, or the equipment eventually discarded. The apparent weightlessness of a service can hide the places where its costs accumulate.

The International Energy Agency’s 2025 Energy and AI report estimated that data centers used about 415 terawatt-hours of electricity in 2024, around 1.5 percent of the global total.⁵ Its April 2026 update estimated 485 terawatt-hours for 2025 and projected about 950 terawatt-hours in 2030, around 3 percent of global electricity then. Those totals cover all data centers, not AI alone. The 2030 figure is a central projection, not a measured future.⁶

Increased efficiency per task can coexist with rising total demand when more people use the systems and use them for more demanding work.⁶ Neither the global total nor its projected share tells a town what a new facility will mean for its own electricity supply and prices. Those questions require the project’s expected load, local supply, and the terms under which new infrastructure will be paid for.

Scroll sideways to see all columns. Keyboard: focus the table and use the left and right arrow keys.

Text equivalent: Electricity: an estimate and a projection

Part of figure	Value or meaning
Title	Electricity: an estimate and a projection
Date and scope	Global data centres • IEA, 16 April 2026 All data centres, not AI alone.
Unit and scale	Electricity per year • TWh TWh = terawatt-hours Shared scale: 0 to 1,000 TWh
Historical estimate	2025 • historical estimate 485 TWh Solid bar: estimated past use
Central projection	2030 • central projection About 950 TWh Dashed bar: projected future use Around 3% of global electricity demand in 2030.
Editorial questions	The local questions remain Who pays for the grid? Is water available? What happens to reliability and household bills? A global projection cannot answer those questions for a particular place.
Source and licence	Source: IEA (2026), executive summary. CC BY 4.0. Original visualization.
How to read the figure	Both bars begin at zero on the same scale. Their fill and labels distinguish the historical estimate from the central projection. No intervening annual path or statistical confidence band is implied. Global totals cannot determine a particular community’s costs.

Figure 3. Two rounded headline values from the IEA’s April 2026 executive summary. The future value is a central projection, not an observation or a confidence limit. Both cover all data centres, not AI alone.

UNCTAD’s Digital Economy Report 2024 examines the material side of digitalization, including resource extraction, energy and water use, and waste. It also draws attention to unequal distributions of benefits and environmental burdens. That wider view matters because a service can appear efficient at the point of use while shifting costs elsewhere.⁷

For a small state considering a data center or related investment, the relevant questions are concrete. What additional electricity is required, and when? Who pays for the infrastructure? What happens during shortages? What cooling system is proposed, under local conditions? What are the effects on water availability, noise, land use, and eventual equipment disposal? Which promises are contractual, which are modeled, and which are merely aspirations?

A general assurance that modern technology solves the water question cannot answer a site-specific question. Nor does a headline about enormous global energy demand establish that a particular local project is unwise. Different designs, locations, power sources, and operating arrangements can produce different results. The obligation is to examine the proposal that actually exists.

The community should also know what it receives. Temporary construction employment, continuing technical work, tax revenue, training, grid investment, and affordable access to services are different benefits. They should not be blended into one impressive number. A project may still be worthwhile with a modest number of permanent jobs, but the case should be made honestly.

The alternative use of scarce resources belongs in the calculation. A long-term commitment of electricity or land can limit choices that are not yet visible. This does not require preserving every resource unused. It requires recognizing that a contract distributes future options as well as present income. A government cannot responsibly value an investment only by what arrives at the signing ceremony.

At the same time, it would be self-defeating to treat environmental responsibility as a reason small states should remain consumers of technology developed elsewhere. There are promising questions for local and regional research: how to identify water leaks sooner, improve energy management in buildings, support agricultural decisions, assess damage after storms, or make scientific information more usable. These are proposed directions for investigation. They need appropriate data, competent partners, and evidence that they improve on available alternatives.

Some benefits will depend less on a new model than on completing the ordinary work around it. A predictive tool cannot repair a pipe by itself. A weather interpretation cannot reach a person without a trusted communication channel. A research result cannot improve a service if nobody has the authority or funds to act on it. AI can strengthen a chain of work, but the missing links remain public responsibilities.

This is where small places can contribute to the global debate. Their experience can make dependency, coordination, and opportunity cost unusually visible. A good solution must fit real limits. It must survive staff turnover, budget pressure, changing suppliers, and the next disruption. Those requirements are not obstacles to serious innovation. They define serious innovation.

The moral question also comes into clearer focus. A human family cannot mean a future designed by a few centers and offered to everyone else on nonnegotiable terms. It must include the ability of a small society to help choose what technology is for, to obtain a fair share of its possibilities, and to retain a voice when the arrangement needs to change. That ambition becomes credible when it is expressed in contracts, institutions, skills, and services people can actually use.

Sources and notes