China’s AI companies are increasingly selling more than models overseas. They are offering local partners a way to build systems around their own languages, data, and industries. In a report from Kenya, Cairo, Kazakhstan, and Southeast Asia, Xinhua describes Chinese companies helping emerging markets create large language models that are designed for local use rather than simply importing an English-first AI service.
The approach is important because most frontier models were initially trained and optimized for languages with the largest digital footprints. Countries whose languages have less web data can find themselves dependent on systems that understand their users imperfectly and route sensitive information through foreign infrastructure. Chinese providers are using multilingual speech technology, open-weight models, and on-site deployment to address that gap. Their pitch is not only lower cost. It is a path to local AI capability.
iFLYTEK Builds an ASEAN Base for Ten Languages
iFLYTEK’s Spark speech large language model is at the center of the company’s Southeast Asia strategy. Xinhua reports that Spark supports speech recognition and simultaneous interpretation in more than 130 languages. The company has worked with Guangxi, the Chinese region bordering several ASEAN markets, to create the Spark ASEAN Multilingual Large Model Base.
That base covers 10 languages, including Malay, Indonesian, Vietnamese, and Thai. iFLYTEK says the system can deliver performance comparable to leading international models with fewer parameters, a design intended to make industry-specific deployments more efficient and more secure. The company has signed cooperation agreements with partners in Laos, Malaysia, and Thailand.
In Malaysia, iFLYTEK and Enjoy TV & Film Broadcasting Corporation have established an AI multilingual intelligent dubbing and translation center. The center supports more than 130 languages and is intended to build an Asian-oriented dubbing ecosystem. That is a more concrete use case than a general chatbot. It combines translation, speech, media localization, and the distribution needs of regional content companies.
Dong Bin, vice president of iFLYTEK’s brand marketing center, told Xinhua that mainstream English large models have “clear shortcomings” in many overseas markets. He said there is real demand for systems customized for local languages. iFLYTEK has also open-sourced its large model, and the company says nearly 600,000 overseas developer teams are using its algorithms and speech tools to build customized products.
Huawei, DeepSeek, and 01.AI Localize the AI Stack
Other Chinese companies are pursuing the same idea through different technical and commercial models. Xinhua reports that Huawei launched its first 100-billion-parameter Arabic large language model in Cairo. The model was trained on local Arabic datasets from finance, power, and oil-and-gas sectors. Huawei says it achieves 96% speech-recognition accuracy and serves more than 20 Arabic-speaking countries and regions.
The Arabic model is notable because it combines language localization with industry data. An energy company, financial institution, or public-service provider needs more than general-language fluency. It needs a system that recognizes local terminology, regulations, workflows, and speech patterns. Huawei’s model is therefore positioned as infrastructure for local sectors rather than merely a Chinese-language export.
DeepSeek represents a different route. Its open-source model is presented as a lower-cost starting point for African developers. Xinhua cites input pricing of about $0.27 per million tokens and output pricing of $1.10 per million tokens. Open weights also let developers adapt a model without beginning from scratch, lowering the financial and technical barrier to a locally useful system.
Beijing-based 01.AI has chosen a partnership model. The company jointly established Q.AI with Kazakhstan to develop and deploy local large language models, AI agents, and enterprise AI platforms. Kai-Fu Lee, 01.AI’s chief executive, told Xinhua that the company prioritizes local data sovereignty, on-site deployment, and local operations. His argument was direct: emerging economies should not simply buy a model or outsource their AI capability. They should build durable national capacity around their own data, languages, industries, and governance rules.
That model of deployment is closely related to EastFrontier’s earlier reporting on how Chinese AI models are gaining ground in Africa. Adoption alone can increase a model’s international footprint. Localization goes further by giving local institutions a role in data, deployment, and product design.
Open Weights Turn AI Exports Into Local Capability
The commercial logic behind this strategy is clear. A provider that offers only a closed API keeps most of the value chain at home. A provider that supplies open weights, language tools, developer support, and local deployment can become part of a country’s broader digital infrastructure. It may earn less from one centrally hosted model, but it can build durable technical relationships across education, healthcare, office software, media, and public services.
This is also why China’s open-model strategy matters beyond benchmark scores. EastFrontier previously described China’s use of free and open-source AI to reach the Global South. The new Xinhua report adds a more specific picture: iFLYTEK is working on ASEAN languages, Huawei is building an Arabic model with sector data, DeepSeek offers low-cost open-source access, and 01.AI is creating a joint venture in Kazakhstan.
China’s policy language reinforces that commercial push. At the World AI Conference in July, China argued for open and mutually beneficial AI development and pledged 5,000 AI training and seminar opportunities for developing countries over five years. Xue Lan, dean of Tsinghua University’s Schwarzman College, told Xinhua that China’s approach includes helping countries develop applications suited to their own conditions and nurturing local expertise.
The strategy will face practical limits. Local data quality, computing resources, regulatory capacity, and developer training vary widely across countries. A multilingual model is not automatically useful if a customer lacks the infrastructure or governance systems needed to operate it. Yet the emerging-market opportunity is increasingly defined by those requirements, not just by the size of a model.
Chinese companies are positioning themselves accordingly. They are trying to make local-language AI an industrial partnership rather than a one-way software export. For countries that have been underrepresented in the first wave of foundation models, this may offer a more direct route into the AI economy. China’s AI industry creates a global market built not only on usage, but on local ownership of capability.