Pak-Sun Ting, founder of Hong Kong-based Votee AI, told us that the buzzword "sovereign AI" is less about political control and more about giving each market the ability to run artificial intelligence in its own language and on its own terms.
What is sovereign AI?
Across the globe, governments and firms are wary of becoming dependent on AI services that are hosted and governed by a single foreign power. In Europe the focus is on keeping personal data within the bloc, while in parts of Asia and the Middle East the priority is to nurture a domestic AI industry that can generate economic returns. For smaller economies the goal is simply to avoid a scenario where a foreign supplier could switch off a critical service.
Why language matters
Votee AI's flagship model operates in Cantonese, the dialect spoken in Hong Kong and neighbouring Guangdong province.
"The whole AI revolution is in English and Mandarin," Ting said. "Cantonese is used in education, health care and police communications. If those don't get covered, then AI is essentially useless."Other regions are following suit. Indonesia's telecom giant Indosat is developing Sahabat AI for Bahasa Indonesia, a language spoken by more than 200 million people. South Korean firms are competing in a government-backed "AI Squid Game" to produce a home-grown model for Korean speakers, another 80 million-strong market. Even Saudi Arabia's Public Investment Fund has backed an Arabic-language model built by Chinese developer MiniMax.
These languages are often described as "low-resource" because they lack the massive text corpora that English or Mandarin enjoy, making model training more challenging and costly.
Cost and practicality
Building a sovereign model does not require the same scale of investment as the multi-billion-dollar projects of firms like OpenAI or Anthropic. Ting estimated that Votee AI spent roughly $250 000 to train its Cantonese model, a figure that is modest compared to the tens of billions poured into the most powerful large-language models.
Governments, which are likely to be the biggest customers for such models, do not necessarily need the most powerful versions. A more modest model can meet public-sector needs while keeping expenses in check. Moreover, the rise of open-source Chinese models that can be downloaded and run locally offers an additional source of technology that can be adapted to local languages.
What happens next?
As the European Union continues to tighten data-protection rules, see the European Commission's digital single market framework, the demand for AI that can be kept within national borders is likely to grow. Companies that can deliver language-specific models at a reasonable cost may find a ready market across the continent and beyond.
For Ting and his peers, the next step is to expand the capabilities of their models beyond basic text generation, adding domain-specific knowledge for sectors such as health care and public safety. The broader trend suggests that "sovereign AI" will evolve from a political slogan into a practical choice for organisations that value autonomy and linguistic relevance.