The release of DeepSeek’s new artificial intelligence model has shaken assumptions about the dominance of US tech giants. Amid stock crashes, claims of IP breaches and talk of a ‘Sputnik moment’, a major implication has been overshadowed: With the resource barriers to generative AI advances lower than had been commonly assumed, the scope for developing countries to harness AI for their own needs is greater. Not just as customers of a few AI ‘hyperscalers’, but as creators and innovators. Making the most of this opportunity will require focus on building AI capacity in the Global South.
The claimed low cost of training the DeepSeek R1 large language model and its (more or less) open-source nature does not just mean cheaper AI that is more affordable to use. It also heralds a less centralised AI ecosystem in which advantages gained by massive investment are tougher to defend. DeepSeek is not a one-off but is rather an example of the shrinking gap between the performance of closed-source (think ChatGPT) and open-source AI, which some claim is ‘democratizing AI’.
With lower-cost, open-source AI available, investments in the AI capabilities of poorer nations can be expected to produce greater economic and developmental benefits.
This matters because in the shadow of the race to release the next big thing in AI has been another race, against a widening digital divide, which a recent UN report warns ‘could limit the benefits of AI to a handful of States, companies and individuals’.
While the digital divide is old news, AI’s rapid advances have triggered new concerns that developing countries could be left behind. Last summer, the UN General Assembly unanimously adopted a resolution on ‘enhancing international cooperation on capacity-building of artificial intelligence’. Capacity-building refers to work that strengthens the abilities of individuals, institutions and countries to meet challenges and achieve objectives. The General Assembly resolution called on states, international organisations, businesses and other stakeholders to contribute to cooperation through a long list of activities including knowledge sharing, technical assistance, research cooperation and personnel training.
The DeepSeek developments should give this work new urgency, because they suggest that the opportunities for Global South nations to develop and deploy AI are larger than previously appreciated and subject to less constraints. There are several reasons for this. First is the model’s apparent efficiency, which will be of great interest to resource-constrained developers. Second, the open source (more precisely, open weight) nature of the model makes it easier for others to learn from and iterate. Third is DeepSeek’s competitive performance, relative to much more expensively developed market leaders. Finally, DeepSeek’s emergence indicates that AI advances are difficult to keep within a single geopolitical bloc, regardless of trade restrictions. For all this, poorer nations will only be able to take advantage of these opportunities if they have the capacity to do so.
Work under the Paris Agreement on climate change shows what AI capacity-building in practice can achieve. There are multiple ways in which AI can help countries deal with the problems of climate change, from enhancing traffic flow to monitoring deforestation. At the same time, AI poses climate risks, like increasing energy and water consumption.
To help developing countries engage with these opportunities, challenges and risks, the UN Climate Technology Centre & Network launched a global capacity-building programme on AI for climate action. This has taken the form of regional workshops bringing together over seventy national climate officials from Africa, the Asia-Pacific and Latin America. The workshops not only delivered insights from AI leaders in research, business and international institutions but resulted in multiple new ideas for CTCN technical assistance projects, which help developing countries put technologies into action.
'The impact of more trade restrictions on China’s AI ecosystem is a question for the future, but the key point now is what a group has been able to produce with far more limited resources than were needed to produce similarly performing models.'
In a related initiative, a global competition attracted over a hundred project proposals for AI applications to combat climate change, with the winning entry – an AI-driven disaster resilience platform – announced at the Baku climate conference. Most participating project teams were from developing countries. The top finalists will receive support to turn their ideas into reality. An online hub providing free access to apps developed through the competition will further explore the open-source model’s potential expand AI access.
Climate change is just one of many issues that will be impacted by AI, alongside health, trade and many others. For every use case there is a corresponding need to help poorer nations get to grips with AI.
The DeepSeek breakthrough suggests that the future of AI – a general purpose technology – will not be dominated by a handful of trillion-dollar gatekeeper companies. Recognising this does not of course mean that it is somehow curtains for the companies that have dominated the AI story for the last two years. The likes of Meta and OpenAI will be learning from the DeepSeek developments just as surely as other Chinese firms are. More competition means an intensified scramble for innovation and efficiency.
The long-term significance of this moment is unlikely to be DeepSeek itself, but what DeepSeek showed is possible. The fact that this is a Chinese company has shaped much of the reaction, including from those who see AI primarily through the prism of Sino-US rivalry. In the United States, prominent figures in both politics and business have called for tougher export controls and more. This strand of reaction is captured by the title of a bill introduced by a United States senator, the ‘Decoupling America’s Artificial Intelligence Capabilities from China Act’. The impact of more trade restrictions on China’s AI ecosystem is a question for the future, but the key point now is what a group has been able to produce with far more limited resources than were needed to produce similarly performing models.
.png)
As researchers, entrepreneurs and policymakers explore the potential of open-source AI, capacity-building that empowers developing countries will be a critical means of putting AI at the service of all humanity.
Stephen Minas is Professor at Peking University School of Transnational Law and Vice-Chair of the UN Climate Technology Centre & Network Advisory Board.