From Foreign Policy · Lizzi C. Lee, Zhuoning Li · · 3m
How China's AI is winning the global south
Beijing's open-weight AI models are gaining ground in developing markets by offering flexibility, affordability, and local control that U.S. competitors can't match.
In brief
Beijing's open-weight AI models are gaining ground in developing markets by offering flexibility, affordability, and local control that U.S. competitors can't match. Chinese AI adoption in emerging markets follows a proven playbook: provide modular components that local developers customize for their needs, creating…
Chinese AI models are surging
Alibaba's Qwen has 151,448 derivatives on Hugging Face, 4.7 times more than Meta's Llama. It draws 39.6 million monthly downloads, far ahead of Llama's 7.5 million.
U.S. AI firms use walled gardens
OpenAI and Anthropic meter access through APIs and subscriptions. This model struggles in cost-sensitive markets across the global south.
China's approach is organic
Unlike the Belt and Road Initiative, Chinese AI adoption isn't pushed by Beijing. Global developers are pulling it in, making influence harder to dislodge.
Linux offers a precedent
Red Hat built profitable businesses around Linux subscriptions and services without selling the software itself. Chinese AI could follow the same path.
The Lego-like model works
Chinese companies provide modular components. Local businesses assemble them into products tailored to their markets, bundling with microfinance and training.
Africa's EV story shows the pattern
Chinese suppliers provide batteries and motors to local companies like Zeno Uganda, which builds battery-swapping ecosystems around local infrastructure constraints.
Chinese AI fits local needs
Built under tight computing constraints, Chinese models run efficiently on less demanding hardware. They also perform well with non-Latin scripts like Burmese, Thai, and Vietnamese.
Cost and sovereignty matter
Chinese models are dramatically cheaper than U.S. proprietary options. Open weights also let governments control where data is stored, avoiding U.S. cloud providers.
Singapore builds on Qwen
SEA-LION, a regional AI system for Southeast Asia, is built on Alibaba's Qwen to give countries sovereign control and better support for local languages.
Malaysia pursues data control
Malaysia is evaluating a 494 million dollar sovereign AI initiative powered by Huawei's Ascend 910C chips to gain more control over national data.
Brazil goes further with Huawei
In August, Brazil announced a 251 million dollar supercomputing project with Huawei and iFlytek, plus a separate 193 million dollar supercomputer tender, both based on Qwen.
Adoption isn't ideological
Developers choose Chinese models for practical reasons: price, flexibility, and performance for specific needs. Adoption isn't binary; many use combinations of closed and open models.
Switching costs create dependence
Once applications, data pipelines, and services grow around Chinese models, switching becomes expensive. A practical choice today becomes structural dependence tomorrow.
“The stickiest technological dependence is the kind users choose for themselves.”
China gains lasting influence
By supplying modular components that others build with, China could gain the same lasting influence in AI that it already has in electric vehicles.
What matters most
Chinese AI adoption in emerging markets follows a proven playbook: provide modular components that local developers customize for their needs, creating structural dependence that's harder to reverse than top-down deals.
