We are witnessing one of the most dangerous narratives to emerge in the ICT4D space in decades.
While Washington and Beijing battle for technological supremacy, development practitioners are being pushed to pick sides in a competition that misrepresents the global distribution of AI capabilities and undermines the potential for international research and regulatory cooperation.
The US-China “AI Cold War” Narrative by Yujia He & Richard Heeks shows how this framing between the US and China is fundamentally distorting how we approach artificial intelligence in development. It forces a binary choice that serves neither our constituents nor our mission to accelerate social and economic progress.
This zero-sum mentality is exactly what our field should reject.
Binary Traps Are Damaging Our Work
The evidence is clear: this “cold war” narrative is creating real problems for development organizations right now.
China largely decouples the capacity-building conversation from the governance conversation, arguing that governance is an obstacle to development in the Global South because it increases regulatory costs, imposes barriers to entry, and takes the focus away from capacity building.
Meanwhile, the US approach has become increasingly conditional and restrictive. The United States established new outbound investment screening measures for US investments into Chinese companies with activities relating to AI, semiconductor, and supercomputing technologies, effectively forcing development partners to navigate an increasingly complex web of geopolitical considerations.
For practitioners, this creates an impossible situation.
- Do we choose cheaper, more accessible Chinese AI solutions that come with governance concerns?
- Or do we opt for Western alternatives that may exclude the very communities we’re trying to serve?
Three Critical Blind Spots
1. Global South has AI innovation too
The binary framing also misrepresents the global distribution of AI capabilities. Data shows that in terms of both firm patents and market capitalisation, the US exceeds China, the EU and the rest of the world combined. Meanwhile, by the measure of firm patents, the EU is on par with China.
More importantly, an increasing number of middle- and low-income countries published their national AI development strategies in 2023 and 2024, with geographic diversity across Africa, Latin America, and Asia. These countries aren’t just choosing between US and Chinese solutions—they’re building their own.
2. Regional cooperation opportunities
There are initiatives that represent genuine alternatives to superpower competition.
- African Union Continental AI Strategy released in 2024
- ASEAN released its Responsible AI Roadmap for 2025-2030
- European countries created the Hamburg Declaration on AI.
3. Emerging technological colonialism
When we frame AI development as a choice between two superpowers, we implicitly accept that the Global South must remain dependent rather than develop sovereign digital capabilities.
President Luiz Inácio Lula da Silva asked why Brazil, a country with 200 million people, a nation 524 years old with a globally respected intellectual foundation, couldn’t create its own mechanisms instead of relying on AI from China, the United States, South Korea, or Japan?
The Real Competition We Should Be Focusing On
Here’s what the “AI cold war” narrative obscures: While Big Tech firms have sought to improve their models by incorporating more languages in their training datasets, most major LLMs still underperform for non-English languages and cultural contexts, especially for low-resource languages, leading to the “LLM digital divide“.
The real competition isn’t between Washington and Beijing—it’s between inclusive, locally-relevant AI development and continued digital exclusion.
Efforts by public and private sector players in Global South countries to develop local monolingual models and regional multilingual models are underway. These models are often built upon various existing open-source models developed by US, Chinese or European companies like Meta, Google, Alibaba, and Mistral, and fine-tuned using data from local and regional languages.
Four Strategic Recommendations
1. Embrace Multi-Alignment Over Binary Choices
Stop forcing false choices between US and Chinese AI ecosystems. ASEAN seeks to advance AI governance and development cooperation with both China and the US, as shown in the leadership statements for Comprehensive Strategic Partnerships with both countries in 2024. We should follow this model of strategic autonomy.
2. Prioritize Open Source and Interoperability
Open-source and lightweight models are especially attractive for adoption in the Global South. Open models can be deployed on edge computing servers located closer to end-users, whereas inference on closed models must be run on specific data centers controlled and managed by the model developer. Support initiatives that reduce vendor lock-in and increase local control.
3. Invest in South-South Collaboration
Some regional multilingual models, such as SEA-LION (Southeast Asian Languages In One Network) and AfriBERTa, benefit from technical collaboration with Global North firms and universities. These partnerships offer more balanced alternatives to great power competition.
4. Focus on Local Capacity Building
Rather than choosing sides in a geopolitical competition, we should focus on building local AI capabilities that serve development goals. China’s strategy involves four key components including calls for more representation of developing countries in global AI governance and emphasizing the role of the United Nations as a channel for communication—regardless of the source, this emphasis on local capacity deserves support.
Evidence-Based Decision Making
As the research suggests, given the urgent problems facing our planet as a whole, we need to take a data-led approach to honing regulation to benefit resilient, diverse markets and societies globally.
For development practitioners, this means evaluating AI solutions based on their effectiveness, accessibility, cultural appropriateness, and development impact—not their country of origin. It means supporting initiatives that empower local communities to shape their own technological futures rather than remain dependent on external powers.
The future of AI in development isn’t binary. Nor should we be binary in our approach to building it
