Over the last couple of months, I had several phone calls and WhatsApp conversations with Adam Satariano, one of the journalists behind a New York Times article published on 5 August 2026 about the growing use of Chinese artificial intelligence models across Africa.

Adam and I went back and forth quite a few times as he sought to understand what Chinese open-weight models could mean for AI solutions in Kenya and the rest of Africa, particularly given their affordability compared with the significantly more expensive and less flexible frontier models from American companies such as OpenAI, Anthropic and Google.

The article ultimately quoted me making a comparison that captures the economics rather well: “Why use an expensive Ferrari to do the school run when a Toyota hatchback can do the same?”

My point was not that the Ferrari has no value. It was that most African businesses do not need the most powerful model available for every task. If a less expensive model can reliably handle customer service, document processing, content localization, business analysis, legal workflows or routine software development, the commercially logical decision is to use it. Depending on the model, workload and computing setup, Chinese open-weight models can be substantially cheaper — and in some situations up to 90% less expensive — than proprietary American alternatives.

The larger implication of the New York Times article, however, goes well beyond pricing. It shows that Kenya and the rest of Africa are becoming an important proving ground for a new global AI order in which model capability, cost, customization, data control and geopolitical reliability matter more than the nationality of the company that built the model.

The Kenyan Examples Make A Compelling Case For Chinese AI Models

What makes the article especially relevant is the range of Kenyan technologists, entrepreneurs, companies and policymakers whose experiences bring the issue to life.

Michael Michie of EverseTech, for instance, is building AI systems for Kenyan banks and government agencies. His customers are price-sensitive, so his company builds most of its solutions on Chinese open-weight models. His perspective is pragmatic: the decisive question is not who built a model, but whether it provides the capability the customer needs at a price the customer can sustain.

This is clearly an important distinction. Much of the global AI conversation is driven by benchmark leadership and which AI lab has produced the most advanced model(s). Kenyan businesses and organizations are more likely to ask whether a solution works on local infrastructure, can be adapted to local requirements, protects sensitive data and delivers a measurable return on investment (ROI).

The experience of Ahmet Acar, a former principal adviser at Amazon Web Services who lives in Nairobi, adds another dimension. He found Anthropic’s developer programme slow to access, while joining Alibaba’s programme took only minutes. That difference may sound simple but in emerging markets like Kenya and the rest of Africa, responsiveness is part of the product. Therefore, an AI company that makes developers feel like a priority can build an ecosystem much faster than one that treats Africa like a rounding error, which is more often than not the lived reality. 

Kamal Budhabhatti of Craft Silicon demonstates why the answer will not always be a Chinese model. Using Anthropic’s Claude Code, a team of 10 completed in half the time a project that previously would have required 100 engineers. The investment was high, but the speed and capability justified it. At the same time, Huawei offered Craft Silicon significant incentives to consider Chinese systems, including free computing capacity. Craft Silicon’s international banking clients, however, may impose restrictions on Chinese technology because of security, compliance and geopolitical concerns.

This is the real enterprise decision. It is not simply a contest between cheap and expensive AI. There are critical considerations such as overall performance, total cost, data residency, cybersecurity, regulatory compliance, model transparency, client requirements and vendor risk.

Shikoh Gitau of Qhala highlighted another advantage of open-weight models: operational control. Her business was affected when access to DeepSeek was disrupted around China’s national examinations. A hosted service can be withdrawn, restricted or repriced regardless of whether it comes from China or the United States. If the weights can be downloaded and deployed on infrastructure that a Kenyan company controls, the organization has a stronger measure of continuity and independence.

This does not remove every risk. A model can still contain bias, security vulnerabilities, political sensitivities or problematic training data. The article’s Ugandan example is a great example: Sunbird AI initially used an Alibaba model for its Sunflower service because it performed well across local languages, but later moved a product to Google’s Gemma model after concerns emerged about how the Chinese model answered politically sensitive questions. Open-weight access gives developers more control, but it does not eliminate the need to test models rigorously.

The article also features Bernard Momanyi Nyagaka of Sanifu, whose company uses OpenAI models while being courted by a Chinese provider, and John Tanui, Principal Secretary in Kenya’s Ministry of Information, Communications and the Digital Economy. Tanui’s position — that Kenya should not lean exclusively West or East — is probably the most strategically logical response. Kenya does not need to choose one geopolitical bloc. It needs the capacity to choose the right model for each use case and to switch when cost, policy, performance or risk changes.

China’s Emerging AI Belt & Road Initiative

For roughly two decades, Chinese companies have helped build Africa’s telecommunications, cloud, data-centre and digital-service infrastructure. Huawei’s role in African networks, the Huawei-built data centre at Konza Technopolis and the technology underpinning platforms such as M-PESA have already created deep commercial and technical relationships.

AI is becoming the next layer of that relationship. The New York Times reports that a DeepSeek-based project promoted at Konza is intended to combat telecommunications fraud, while Chinese companies are offering African developers free computing resources, engineering assistance and easier access to their ecosystems. Chinese-made smartphones sold across Africa are also increasingly capable of distributing Chinese AI services directly to consumers.

This begins to look like an AI version of China’s Belt and Road Initiative: not necessarily a single centrally branded programme, but a connected combination of models, cloud infrastructure, devices, developer incentives, standards and diplomatic partnerships. In July 2026, Kenya joined other countries in signing the agreement establishing the World Artificial Intelligence Cooperation Organization in Shanghai. Reuters reported that 29 countries, including 10 African states, became founding signatories.

The strategy is gaining traction. According to the New York Times analysis, Chinese models now account for roughly half of AI use on OpenRouter, up from less than a quarter a year earlier, while 19 of the 25 most downloaded open models on Hugging Face are Chinese. The question is therefore no longer whether Chinese AI will be used in Africa. It is how deeply it will become embedded in the continent’s digital economy and on whose terms.

Kimi K3 Shows How Quickly The Gap Is Narrowing

Moonshot AI’s Kimi K3, released on 16 July 2026, makes this debate even more significant. Moonshot describes Kimi K3 as a 2.8-trillion-parameter open-weight model with native vision, a one-million-token context window and capabilities designed for long-horizon coding, reasoning and knowledge work. Independent evaluations place it close to leading American frontier systems on several agentic and coding tasks, although performance varies by benchmark and use case.

The important point is not that Kimi K3 beats every model from Anthropic, OpenAI or Google. It does not. The point is that a Chinese open-weight model can now operate close enough to the frontier for many high-value tasks while giving developers greater deployment flexibility and, for some workloads, substantially lower costs.

China has achieved this while operating under American export controls on advanced AI chips. Even here, nuance is important. China’s access to cutting-edge computing is constrained, but it is not zero. Reuters reported that Moonshot has access to around 20,000 Nvidia Hopper-generation chips through an Alibaba computing agreement. Chinese laboratories have nevertheless been forced to focus intensely on architectural and computing efficiency, and those efficiencies are producing models that are increasingly attractive to cost-sensitive markets.

Open-weight should also not be confused with unrestricted or cost-free. Companies still have to pay for servers, electricity, engineering, cybersecurity, model evaluation and maintenance. Licenses may also impose commercial conditions. The advantage is not “free AI”; it is greater control over where and how a model runs, alongside the possibility of lowering the total cost at scale.

Kenya Is Not A Peripheral AI Market

This global contest matters because Kenya is already one of the world’s most enthusiastic AI markets. Omdia estimated that, as of January 2026, ChatGPT was used monthly by 22.9% of Kenya’s active smartphone base — almost 7.5 million people. The same research found that ChatGPT was Kenya’s seventh most-searched term in 2025 and that searches had doubled year on year.

Those numbers show that the demand for AI is not hypothetical. Kenyans are already using it to learn, work, build businesses, write software, create content and solve everyday problems. OpenAI’s decision this past week to introduce unlimited text chats for free ChatGPT users will lower one consumer-access barrier even further. However, free consumer chat and the economics of building AI into an enterprise product are different questions. For Kenyan businesses deploying AI across thousands or millions of interactions, inference, integration and infrastructure costs still matter enormously.

The opportunity is therefore much bigger than choosing between ChatGPT, Claude, Gemini, DeepSeek, Qwen or Kimi. Kenya can become a market where global models are tested, localized and combined with local data to build solutions for agriculture, financial services, health, education, government services and the creative economy. It can also export those solutions across a continent facing many of the same affordability and infrastructure constraints.

Kenya’s Policy Framework Is Catching Up

Kenya launched its National AI Strategy 2025–2030 in March 2025 around three pillars: AI digital infrastructure; data and AI governance; and AI research, innovation and commercialization. When I analysed the strategy last year, I argued that its ambition to build local infrastructure, local datasets, local talent and homegrown solutions could position Kenya as a leading African AI hub — but that delivery would matter more than the policy language.

That implementation challenge is now colliding with fast-moving regulation. The Artificial Intelligence Bill, 2026 was introduced in the Senate in February and received its first reading in April. It proposes a risk-based framework, an independent AI Commissioner, stronger obligations for high-risk systems, transparency requirements, human oversight and safeguards linked to Kenya’s Data Protection Act.

Separately, the Ministry published the Draft Kenya Artificial Intelligence and Other Emerging Technologies Policy for public consultation in July 2026, with submissions closed on 4th August. The draft policy proposes a National AI and Other Emerging Technologies Council and makes sovereignty and strategic autonomy one of its policy pillars.

The existence of both a Bill and a separate draft policy is itself a governance issue. Their proposed oversight structures need to be aligned so that Kenya does not create duplicate institutions, conflicting mandates or compliance uncertainty. More importantly, regulation must address model provenance, data residency, cybersecurity, auditability, procurement standards and cross-border dependence without making it impossible for startups and smaller businesses to innovate.

Kenya Needs Model Optionality, Not Model Loyalty

The key lesson from the New York Times article is not that Kenya should replace American AI with Chinese AI. That would merely exchange one form of dependence for another.

Kenya should instead pursue model optionality: the technical, commercial and regulatory capacity to use proprietary frontier systems where their superior performance and enterprise safeguards justify the cost; open-weight Chinese, American or other models where affordability, localization and control matter more; and locally developed models where Kenyan languages, datasets and cultural contexts require something neither bloc provides.

That means investing in shared computing infrastructure, local cloud and data centre capacity, interoperable systems, rigorous model evaluation, cybersecurity, AI literacy and procurement rules that prevent unnecessary vendor lock-in. It also means ensuring that the companies and people building Kenya’s AI future — from EverseTech, Qhala, Craft Silicon and Sanifu to researchers, universities, startups and public agencies — have a strong input in how policy is designed and implemented.

Kenya’s advantage is that we are not starting from zero. We have a digitally curious population, a strong mobile and digital financial services technology ecosystem, experienced software developers, regional influence and a history of adapting global technologies to local realities. But high adoption does not automatically create sovereignty, and access does not automatically create value.

If Kenya can combine American frontier capability, Chinese open-weight economics and Kenyan talent, data and context, it could build an AI ecosystem that is both globally connected and locally grounded. The countries that benefit most from AI will not necessarily be those that build the single most powerful model. They may be those that learn how to combine the best available models, deploy them affordably and govern them in their own interests.

That is why the developments captured by the New York Times matter. They are not simply evidence that Chinese AI is gaining ground in Africa. They are early signs that Kenya and Africa could have more leverage in the global AI economy than many people realize — provided we use it deliberately.

If you missed it earlier in this blog post, here is the link to the New York Times article.