appConfigEntityId

bodyType

customContents

entityId

entityUuid

isVideo

articlePageQuery

article

articleSchemaContent

content

hooksArticlePageAdvertisementConfigArticle

hooksArticlePageFloatingButtonConfigArticle

appConfig

filter

json

liveArticle

Jevans Nyabiage

The United States may want the world to choose sides in the AI race with China, but in Africa, many businesses are pursuing a more pragmatic strategy: using Chinese models where they offer better value or flexibility, while keeping their options open.
For example, when African insurtech platform Curacel expanded the range of artificial intelligence (AI) models powering its internal infrastructure, it did not look only to Silicon Valley.
Instead, the Nigeria-based firm – which provides AI-powered claims and fraud-detection infrastructure to African insurers and fintech companies – added GLM-5.3, a Chinese model developed by Beijing-based Zhipu AI, alongside models from Western providers.
Henry Mascot, chief executive and co-founder of Curacel, said Chinese models delivered comparable results at lower cost on high-volume tasks such as coding, data extraction, classification and customer support, although Western systems remained stronger on the most demanding reasoning and reliability-sensitive work.
“In our experience, the leading Chinese models have closed much of the performance gap,” Mascot said.
He said Curacel’s multi-model set-up allowed it to use GLM-5.3 selectively.
“We can route suitable workloads to it based on quality and cost without moving the entire stack or locking ourselves into one vendor,” Mascot said. “We have also used lower-cost [Chinese] models within client deployments.”

Curacel’s approach reflects a broader shift, with models such as DeepSeek, Qwen from Alibaba, the owner of the South China Morning Post, and Moonshot AI’s Kimi challenging established systems from OpenAI, Google and Anthropic.
Many Chinese models make their weights – the files that allow companies to adapt them and, with sufficient infrastructure, run them locally – available for download.
Their growing use in Africa coincides with Beijing’s efforts to expand its role in the continent’s digital infrastructure, technical training and AI governance.
Huawei Technologies is one example of how Chinese companies are combining AI models with the infrastructure needed to run them.
In May, its South African cloud business introduced a “model as a service” or MaaS platform, providing pay-per-token access to DeepSeek, Qwen and GLM, together with development and deployment tools.
China last month launched the World Artificial Intelligence Cooperation Organisation (Waico), with Beijing presenting its open-weight technology as an alternative to the largely proprietary systems offered by leading American companies.
The contest extends beyond models to semiconductors, critical minerals, cloud infrastructure and the rules governing AI. African businesses may choose technology according to price and performance, but governments could face growing pressure to align with Washington or Beijing.

That pressure is becoming more explicit. A draft State Department communication reviewed by Reuters on August 14 warned that countries joining Waico could be excluded from the US-led Pax Silica coalition.
Michael Michie, co-founder of Nairobi-based EverseTech – an AI consultancy that also builds AI systems for banks, government agencies and other African organisations – said Chinese models had begun to advance more quickly.
“A year or so ago, you might have expected to wait quite a long time for a Chinese model release that represented a major step forward. Now we are seeing meaningful improvements within six months or less,” Michie said.
EverseTech has moved some of its clients to Chinese models. A legal-sector client switched because of the lower cost per million tokens. Another, which relied heavily on retrieval-augmented generation – a process in which an AI system searches company documents before producing an answer – moved after EverseTech found that a Chinese model could convert the documents into searchable numerical representations faster and more cheaply, Michie said.
“For most customers, the question is no longer simply, ‘Which model is the most powerful?’ It is, ‘Which model gives me the capability I need at a price that makes commercial sense?’” he said.

Price is far from the only consideration. In Uganda, the developers of Sunflower – an AI model providing farmers with weather updates and agricultural advice in over 30 local languages – built the system on Qwen3 after finding it handled Ugandan languages better than models from Meta Platforms or Google.
Djamo, an Ivory Coast fintech company serving francophone Africa, assigns models according to the job at hand. Co-founder and CEO Hassan Bourgi said a lightweight Chinese model could handle a simple customer service task such as closing spam requests, while a more advanced model might be needed for complex work such as writing or reviewing code.
“It’s getting commoditised at some point for the simple stuff,” Bourgi said. “But when it’s very complex, I think the Americans are still a bit ahead today.”
Bernard Momanyi Nyagaka, co-founder of Sanifu AI, said open weights could allow banks and fintech companies to run models within their own infrastructure instead of sending sensitive information to an outside service. Sanifu helps internal teams build AI agents and data applications that automate manual work.
“Open weights mean you can self-host, fine-tune and keep sensitive data – bank records, KYC [know-your-customer] documents – inside your own infrastructure, rather than sending it to a third-party API [application programming interface],” Nyagaka said.


However, self-hosting requires computing capacity and technical expertise. Open-weight models may reduce reliance on a model developer, but dependence can shift to the cloud platforms, chips and technical support needed to run them.
Nyagaka advised that sensitive data should be kept local, regardless of geopolitics. “For banking, data sovereignty should be a design principle, not a vendor feature. Whatever model you use, the architecture should ensure regulated data doesn’t leave your controlled environment,” he said.
And while lower prices can make a model attractive, companies must also consider the cost of leaving the provider later.
Michie said: “The strategic mistake would be designing the institution so that five years from now, it has no realistic ability to move away from that model or provider.”
He said organisations should examine data-portability rules, termination rights and ownership of organisational data before committing to a provider.
Curacel limits that risk through a model gateway – a single interface that can route tasks to Chinese, American, European or self-hosted models. It also tests models against its own workloads rather than relying only on published benchmarks. “That makes switching a routing decision instead of a product rewrite,” Mascot said.

Nyagaka argued that open-weight Chinese models could, in some cases, create less dependence than proprietary Western services because companies could retain and host the model weights themselves. “The lock-in risk is highest with closed, hosted APIs of any origin,” he said.
But some tech executives have rejected the idea that African businesses must choose between Chinese and American AI.
“I would push back on the idea that Africa is merely a testing ground for US-China competition,” Mascot said. “African companies are active buyers. They will choose the technology that solves the problem at the right price, regardless of where it was built.”
That pragmatism, according to Michie, should ultimately give African firms more room to manoeuvre.
“Africa’s strongest position in the US-China technology competition is not choosing one side,” he said. “It is building enough capability and leverage to work with both and eventually contribute more of our own technology to that ecosystem.”


Somali piracy rise tests China’s long-running Gulf of Aden mission

Is Africa becoming a ‘testing ground’ for Chinese power projection?

Digital Silk Road: Hong Kong acts as financial bridge between China and MENA

China bets on early port and power deals in Sudan to secure post-war economic edge

China-Africa relations

The African firms adopting Chinese AI as US-China rivalry intensifies

Chinese AI models including DeepSeek, Alibaba’s Qwen and Moonshot AI’s Kimi are gaining ground with African businesses. Photo: Shutterstock

Nigeria-based insurtech Curacel is using both Chinese and Western AI models, selecting them by task, cost and performance. Photo: Curacel/Facebook

Chinese President Xi Jinping poses alongside delegates to last month’s World AI Conference in Shanghai. Photo: Reuters

Moonshot AI promotes its Kimi K3 model at the World AI Conference in Shanghai in July. Photo: Reuters

Visitors gather at Alibaba’s Qwen booth at the Apsara Conference in Hangzhou, China, in September 2025. Photo: CFOTO/Future Publishing via Getty Images

As Xi’s US visit approaches, basic details of planned AI talks remain uncertain

The bilateral discussions come amid growing US-China rivalry over competing visions for global AI development. Photo: Shutterstock

Making better robots depends on better data capture, Chinese firm 51World says

Li Yi, founder and CEO of 51World, speaks during a briefing in Shenzhen on Tuesday. Photo: YouTube/ @51world

Chinese robotaxi firms see eased regulatory bottleneck in Europe as deployment picks up

A Pony AI robotaxi on an exhibition stand during the 2026 Auto China show in Beijing. Photo: dpa

US urged to follow China’s lead in treating data as an economic asset

An employee programmes a robot to sort items at Maniformer, a physical AI data service platform, in Shanghai on August 5. Photo: Xinhua

Hong Kong firm bets on Chinese open-weight models to rival CoreWeave

A technician works at an Amazon Web Services AI data centre in the United States. Photo: Getty Images

News

China

Diplomacy

Lower costs, downloadable models and greater language flexibility are helping Chinese AI models gain ground across the continent


Lower costs, downloadable models and greater language flexibility are helping Chinese AI models gain ground across the continent.


The African firms adopting Chinese AI as US-China rivalry intensifies

.css-1c6uqr6{color:inherit;font-weight:inherit;font-size:inherit;font-family:inherit;line-height:inherit;overflow-wrap:break-word;}Lower costs, downloadable models and greater language flexibility are helping Chinese AI models gain ground across the continent

Lower costs, downloadable models and greater language flexibility are helping Chinese AI models gain ground across the continent