AI Without a Budget?

Artificial intelligence is increasingly appearing in public-sector modernization plans.

Chatbots, AI-powered surveillance, document automation, predictive analytics, fraud detection and platforms capable of assisting citizens 24 hours a day.

It sounds promising. But Peru and many other Latin American countries face a much less futuristic question:

What happens when a public institution wants artificial intelligence but still struggles to finance cameras, computers, connectivity or basic information systems?

The uncomfortable answer is: Yes, AI can be implemented with limited resources.

But not every type of AI, not in every institution and certainly not by skipping the foundations.

AI does not necessarily mean cameras

When people think about artificial intelligence in government, they often imagine large monitoring centers, facial recognition systems, hundreds of cameras and giant screens.

That is only one category of AI. Artificial intelligence can also classify documents, answer citizen questions, summarize files, identify duplicate information, analyze databases and assist with administrative work.

Some of these applications can operate through cloud infrastructure and use information an institution already possesses.

Peru already has examples. In 2026, Ceplan deployed a generative-AI chatbot to provide information about public-sector strategic planning, while the Peruvian government maintains an official catalogue of AI applications being used by public institutions.

A municipality without hundreds of cameras is therefore not automatically excluded from using artificial intelligence.

The real question is: What does it want AI to do?

The mistake is starting with the technology

Imagine a municipality whose information is scattered across spreadsheets, paper documents and disconnected systems.

Now add artificial intelligence. The likely result is a modern tool attempting to operate on disorganized information.

OECD and CAF research has emphasized that governments with greater digital maturity are better positioned to benefit from public-sector AI. Governance, data, interoperability, infrastructure, skills and digital security form part of that foundation.

So before asking: “How much does AI cost?”

perhaps governments should ask: “Is the organization ready to use it?”

Some AI projects really do cost millions

When artificial intelligence depends on physical infrastructure, the economics change dramatically.

In Sullana, Peru, a public-safety project worth more than PEN 25 million includes fiber-optic infrastructure, an operations center, communications equipment and 194 cameras, including AI-supported recognition and advanced video analytics. The municipality pursued the project through Peru's Works for Taxes financing mechanism.

Another initiative announced by Peru's Ministry of the Interior aims to connect at least 3,200 municipal cameras to a national intelligent video-surveillance platform, with an announced investment of PEN 37.7 million.

These examples illustrate an important point: AI may represent only one part of the cost.

Before AI comes cameras, connectivity, fiber, servers, control centers, integration and maintenance.

In that environment, promising “artificial intelligence” without explaining how the surrounding infrastructure will be financed leaves major questions unanswered.

Does basic infrastructure always have to come first?

Not necessarily.

That would be the opposite mistake.

If the problem is answering citizen questions, processing documents or analyzing administrative information, purchasing cameras would be irrelevant.

Investment should begin with the problem.

A resource-constrained institution could start with smaller applications: classifying documents; building internal assistants for regulations; analyzing citizen complaints; detecting duplicate records; summarizing large volumes of information; or helping employees locate documents and procedures faster.

These projects are very different from building an AI-powered surveillance network. They also involve very different budgets, infrastructure requirements and risks.

Latin America does not start from the same position

This challenge goes well beyond Peru.

Digital-government development across Latin America and the Caribbean remains uneven. An OECD and IDB assessment covering 23 governments found that the region is progressing, but important differences remain in the foundations required for coherent digital transformation.

OECD and CAF research has also emphasized institutional capacity, digital talent, data governance, interoperability, digital infrastructure and procurement as important foundations for scaling AI in the public sector.

Copying a project designed for a large capital city and transferring it directly to a small municipality may therefore make little sense.

The problems may look similar. The available capabilities may not be.

Peru already recognizes this difference

In April 2026, Peru approved its National Artificial Intelligence Strategy 2026–2030, establishing a national framework for the development, use and governance of AI.

There is also an important principle in Peru's AI regulatory framework.

Implementation is designed to be progressive and phased, taking into account the different capacities and resources of organizations participating in the country's digital transformation ecosystem.

That distinction matters. Not every government institution needs the same AI. And not every institution needs it at the same time.

Cheap AI does not mean free AI

Cloud-based services and existing AI models can reduce the need to purchase large amounts of computing infrastructure.

But costs remain. Licensing. Integration. Cybersecurity. Specialized staff. Training. Data preparation. Human oversight. Maintenance. And potentially continuous cloud-computing expenses.

A seemingly small project can therefore become a significant recurring expense if it is poorly designed.

Governments should not only ask how much a system costs to launch.

They should also ask: What will it cost to operate for the next five years?

Is AI viable for the Peruvian public sector?

Yes.

But adopting AI should not mean pursuing the most spectacular technology available.

A resource-constrained institution may generate more value through a small, clearly defined project that uses existing infrastructure and solves a specific problem.

Another institution with more advanced capabilities may be able to integrate cameras, sensors, predictive systems and sophisticated control centers.

Both can use artificial intelligence. They simply should not use it in the same way.

Problem first. AI second.

Perhaps Latin America's greatest risk is not being left behind by the AI revolution.

The greater risk may be spending scarce resources trying to look technologically advanced.

Governments do not need AI because AI is fashionable.

They need it when it can solve a concrete problem faster, better or at lower cost.

If a municipality still lacks cameras, connectivity or basic information systems, building those foundations may be the right priority.

But if an affordable, controlled AI application can eliminate months of administrative work, waiting for perfect infrastructure would make little sense either.

Digital transformation is not about reaching the newest technology first.

It is about choosing the right technology.

And sometimes the smartest decision about artificial intelligence will be to deploy it. Other times, it will be to wait.