By Qaiser Nawab

For developing countries, technological revolutions often arrive twice: first as a promise and later as a reminder of how far behind they remain. The industrial age was built around capital-intensive factories; the information age around computing power, networks and intellectual property. By the time many countries in the Global South acquired the infrastructure and skills needed to participate, the frontier had moved again.

Artificial intelligence may not follow the same script. That is the argument recently made by Justin Yifu Lin, former World Bank chief economist and a leading Chinese development economist. Lin believes open-weight AI — models whose trained parameters can be downloaded, adapted and run locally — could give developing economies an opportunity they did not have during earlier industrial transformations.

The claim deserves attention, but not celebration. Open-weight AI changes the economics of access. Whether it changes the economics of development is a harder question.

A lower barrier, not a level field

Countries do not necessarily have to build the most advanced AI model themselves to benefit from it. They can take an existing model, adapt it for a local language or industry, and deploy it for tasks ranging from agricultural advice and medical support to education, logistics and public administration.

This matters because the AI frontier is prohibitively expensive. Training the largest models requires enormous computing resources, advanced chips, specialist talent, and capital that most developing economies lack. Open-weight models reduce part of that barrier.

Chinese firms have become important to this shift. Models such as DeepSeek and Alibaba’s Qwen have attracted attention not only for their capabilities but because they can be downloaded and adapted under their respective licences. This widens the options available to developers who cannot afford to build frontier systems from scratch. Lin’s point is practical: if advanced tools become cheaper and more adaptable, latecomers gain room to experiment.

The World Bank has noted that open technologies can help developing countries adapt AI to local contexts without reinventing foundational technology. Yet its research also shows how unequal the landscape remains. Data-centre capacity, venture funding, advanced computing and model development are still heavily concentrated in richer economies.

So the door may be more open, but the road beyond it is hardly level.

It is also worth being precise about terminology. “Open-weight” is not always the same as fully open-source. A model may make its parameters available while withholding parts of the training data or development process. That does not negate its usefulness, but governments should resist treating openness as a binary category. The real question is what users are legally and technically able to inspect, modify, and deploy.

The development question comes after access

The temptation in policy circles is to assume that once a technology becomes affordable, development will follow. History offers little support for that assumption.

Electricity did not industrialise countries that lacked functioning institutions. Internet access did not automatically produce productive digital economies. Smartphones expanded connectivity, but did not by themselves create high-value industries. AI will be no different.

The World Bank’s recent work on AI identifies four foundations: connectivity, compute, context, and competency. The list shifts attention away from models and towards ecosystems. A country may have access to a powerful open-weight model and still lack reliable electricity, affordable cloud infrastructure, local-language data, skilled engineers or public institutions capable of supervising AI responsibly.

For Pakistan and many comparable economies, this is the central policy challenge. The useful question is not whether the country can produce its own equivalent of the world’s largest model. It is where AI can raise productivity in sectors that already matter — agriculture, textiles, logistics, health, education, financial services, small manufacturing and public administration — and what local capacity is needed to make those applications work.

That approach is less glamorous than announcing national “AI leadership”, but probably more consequential. A cotton farmer does not need a sovereign frontier model; she may need reliable crop advice in Urdu, Punjabi or Sindhi. A small exporter may gain more from AI-assisted compliance and market research than from a prestige data centre. A provincial hospital may benefit from decision-support tools adapted to local needs, provided privacy and accountability are built in.

This is where Lin’s idea of a latecomer advantage becomes plausible. Countries entering new technological niches can adopt newer tools directly. But the advantage exists only if local firms, universities and governments can translate general-purpose models into useful applications. Access creates possibility; institutions determine whether possibility becomes productivity.

Openness should lead to capability, not dependence

There is a geopolitical temptation to frame AI as a contest in which developing countries must choose one technological camp over another. That would be a mistake.

The Global South benefits most from competition that expands choice: more models, lower prices, interoperable systems, transparent licences and the ability to switch providers. Chinese open-weight models are valuable in part because they add pressure to a market that might otherwise become more closed and expensive. The same principle should apply to models developed elsewhere. No developing country should replace dependence on one foreign technology stack with dependence on another.

Policy should therefore focus on technological optionality. Governments should prefer standards that allow locally hosted systems, data portability, independent security testing and procurement from multiple providers. Universities should be supported to evaluate models rather than merely consume them. Public funds should build local datasets and skills that remain useful regardless of which company produces the next leading model.

UN Trade and Development has warned that AI’s economic gains could be highly concentrated and that many developing countries remain underrepresented in AI governance. That is the larger issue behind Lin’s argument. Access to weights may give the Global South more room to participate, but participation also requires a voice in standards, safety, trade rules, data governance and the distribution of value.

There is reason for cautious optimism. Useful AI can already be deployed without owning the entire frontier stack. Smaller models can run on modest infrastructure, open-weight systems can be adapted locally, and innovation can occur far from the world’s major technology hubs. This creates space for countries that have historically entered technological revolutions late.

But a second chance should not be confused with guaranteed catch-up.

The countries that benefit will be those that treat AI neither as a miracle nor as a threat imported from abroad, but as infrastructure for problem-solving. China’s open-weight ecosystem is helping widen access; that is significant. What developing economies do with that access is ultimately their own development test.

The fourth industrial revolution may offer the Global South a better starting position than the previous three. Whether it produces a different ending will depend less on who releases the models than on who builds the capacity to use them well.

Author: Qaiser Nawab is Chairman of the Belt and Road Initiative for Sustainable Development (BRISD), an international platform fostering cooperation and innovation across Asia, Africa, and Latin America. He can be reached at qaisernawab098@gmail.com