Developing countries should avoid dependence on a single Artificial Intelligence (AI) provider and retain control over the local data and expertise that determine how AI systems are used, according to technology entrepreneur Tiffany Xingyu Wang.
Speaking at the 2026 World Bank Group and IMF Annual Meetings in Bangkok, Wang argued that AI adoption presents a question of economic and institutional sovereignty as governments seek to deploy the technology in healthcare, agriculture, education and public services.
Her warning comes as international development institutions encourage countries to adopt affordable, task-specific AI applications to improve productivity and expand access to essential services.
Wang said the differences in performance between leading AI models may matter less for many of the specialised tasks associated with small AI than the ability of governments and businesses to retain control over their technology choices.
“Do not depend on one model,” she said, arguing that users should be able to change providers if commercial incentives shift or financial support for a particular AI system diminishes.
She also called on countries to retain control over their local context, including languages, laws, agricultural conditions and healthcare needs, rather than relying entirely on models developed using data that may not adequately represent their populations.
“Rent the intelligence, but own the context,” Wang said. She argued that organisations should establish clear rules governing the use of their data, including limits on training models with local information and retaining data only for as long as necessary.
The issue has economic implications because AI tools developed for one market may not perform reliably in another. Local data, knowledge and institutional capacity are therefore important to ensuring that applications work for the communities they are intended to serve.
Wang also urged governments to diversify their AI suppliers and build the human expertise needed to assess the reliability of automated outputs.
“Being capable does not mean being beneficial,” she said, warning that AI systems can produce confident but incorrect answers and that users may surrender their own judgement without recognising the risks.
The discussion reflects a wider concern that countries could adopt technology without acquiring the skills or institutional capacity to determine how it should serve their development objectives.
Vilas Dhar, president of the Patrick J. McGovern Foundation, said development finance institutions should help governments build the capacity to define their own technology requirements rather than simply accept products offered by technology companies.
He said decisions involving data centres, surveillance, data protection and employment should be guided by public policy rather than driven exclusively by commercial technology providers.
The issue is particularly relevant as countries seek investment in computing infrastructure and AI services. Governments must balance the potential productivity gains against the risks of weak oversight, data misuse and long-term dependence on external suppliers.
Thailand’s Deputy Prime Minister and Finance Minister Ekniti Nitithanprapas said the government was connecting digital public infrastructure with fiscal programmes to improve the targeting of support, while also stressing the need to protect people from online scams and deepfakes.
Wang’s argument extends beyond regulation to the capabilities required for meaningful local ownership. She said AI deployment should involve people who know how to assess its output, verify its recommendations and reject them when necessary.
The goal for developing economies, is therefore not simply to gain access to AI technology. It is to develop the technical skills, data governance and public-sector capacity required to use the technology on their own terms.
The distinction could determine whether AI becomes a tool for strengthening domestic productivity and institutions or another area in which countries depend heavily on decisions made by external providers.
