This is the new battlefield for the technology giants. Africa, which will have 2.5 billion inhabitants in 2050, attracts everyone’s desire, but it is not only its demographics that attract digital players. According to the International Finance Corporation (IFC), the potential contribution of the African digital economy could reach $712 billion, or 8.5% of the continent’s GDP, by 2050.
With the world’s fastest growing population and a booming digital economy, Africa is questioning its place in the global technological revolution. Both governments and startups are multiplying initiatives to ensure that the continent’s future… does not happen without it.
But the challenge remains immense. On a continent where IT infrastructure and budgets remain limited, developing artificial intelligence ecosystems, training models on African data and languages, and then hosting these systems locally represents a challenge that is as strategic as it is complex.
It is in this context that “DeepSeek” enters the scene, a Chinese start-up founded in 2023, which is revolutionizing the global artificial intelligence market. The American economic media Bloomberg recalls that its model, “27 times less expensive to use” than its American competitors such as the Chat GPT-4, offers features comparable to most of its rivals in the sector thanks to an innovative architecture and modest hardware resources.
Chinese AI uses far fewer hardware and power resources than ChatGPT thanks to its so-called “Mix of Experts” (MoE) architecture. Unlike ChatGPT, which requests all of its parameters for each request, DeepSeek only activates a specific part of its modules, significantly reducing power and GPU consumption. And therefore its cost.
Significantly lower cost
But that’s not all. Unlike “proprietary models,” which charge for licenses and access to their infrastructure, DeepSeek only charges for the computing power actually used. The price is based on the use of tokens, that is, words or small fragments of text that the model processes to generate content. Huawei, a partner of the startup, also offers two million free tokens a day to its users.
This economic advantage is decisive and is added to a technical advantage. Bloomberg notes that several factors explain why Silicon Valley models are not suitable for African users. American models require more tokens to process foreign or rare words, often increasing computational costs for those not working in English.
Additionally, because the training data is predominantly Western, these models can miss cultural nuances and incorrectly assign a person’s gender based on their name, or generate images that reflect racial stereotypes.
For African startups, economy and efficiency are therefore tangible. Olubayo Adekanmbi, founder of EqualyzAI, notes that DeepSeek is much more affordable than its competitors. For example, DeepSeek Chat charges $0.27 for one million tokens processed and $1.10 for one million tokens generated, compared to $5 and $15 respectively for OpenAI’s GPT-4o. Result: EqualyzAI pays around $2,700 per month to train its language model, instead of $12,500 with the American solution.
Alexander Tsado, co-founder of Johannesburg-based Alliance4AI, told African Business magazine that AI-related tools could improve access to healthcare in rural areas, boost financial inclusion by helping banks reach underserved communities and help farmers detect crop and plant diseases, contributing to increased yields. “AI in Africa can transform the daily lives of hundreds of millions of people,” he added.
Chinese-American rivalry
DeepSeek’s success in Africa is also part of a broader geopolitical context. Over the past ten years, China has gradually developed its expertise in cutting-edge technologies, including artificial intelligence, under the ambitious “Made in China 2025” plan. Launched in 2015, this program aimed to transform the “made in China” brand, often associated with lower quality manufacturing, into a symbol of high technology and excellence.
By dethroning the famous ChatGPT on its own turf, DeepSeek illustrates the strategic success of this plan. But this should not be seen as Beijing’s desire to fill its coffers to the detriment of Americans. At least not for now. This strategy is reminiscent of another Chinese initiative “Belt and Road” in the field of physical infrastructure and is therefore not intended to generate immediate benefits.
With an African digital economy estimated today at $180 billion, well below OpenAI’s recent valuation of $500 billion, for example, the goal is above all strategic: winning over users, strengthening soft power and accumulating the immense volumes of data that will shape the future of artificial intelligence.
Forgotten sovereignty?
But harnessing this AI comes with risks. According to Ecofin, DeepSeek could become “a Trojan horse: millions of African users dependent on algorithms controlled from Beijing, trained from data absorbed locally without compensation.” A warning that reminds us of the need for the continent to maintain control of its digital infrastructure and its strategic data.
For Ecofin, data are “the new precious resources of the digital age.” And Africa already produces a wide variety of information (books, articles, tweets, financial transactions, medical or scientific data) that feeds the training of AI models. “We must sensibly negotiate the use of all this raw material,” the agency emphasizes, so as not to remain mere spectators in this global technological battle.
Kennedy Chengeta, a Pretoria-based AI entrepreneur and academic, reminds us that even if the benefits of AI are widely recognized, the question of sovereignty remains immense. “Cost has been one of the biggest barriers to AI adoption in Africa. Many African countries lack local data centers and cutting-edge IT infrastructure, forcing businesses to rely on expensive cloud services from international providers. This dependence not only increases operational costs but also limits the continent’s scalability,” he tells African Business.
Source: BFM TV
