From Southeast Asia to Europe, governments are pouring billions into building their own artificial intelligence systems in a bid to assert “digital sovereignty.”
Yet analysts warn that these expensive projects may struggle to compete with tech giants like OpenAI, Meta, and Alibaba - and could end up wasting taxpayer money.
In Singapore, a state-backed AI program has built a language model that speaks in 11 regional tongues, while Malaysia’s ILMUchat can tell the difference between Penang’s Georgetown and its American namesake. Switzerland’s newly launched Apertus model, designed with cultural precision, can distinguish between the Swiss “ss” and the German “ß.”
These initiatives form part of a growing global movement known as “sovereign AI,” where countries from India to the UK aim to reduce dependence on foreign technology. But developing large language models (LLMs) from scratch is an enormous challenge, both financially and technically.
Analysts say the US and China dominate AI research because of their ability to invest hundreds of billions of dollars into the field. For middle or smaller powers, the barriers are steep. Developing competitive models requires vast amounts of data, infrastructure, and capital.
In India, concerns over data security and cultural accuracy have pushed the government to back its own national LLM under the $1.25 billion IndiaAI Mission.
Local developer Soket AI is among the companies leading the effort, after finding that American models struggled with Indian accents and produced confusing legal advice when adapted to Indian law. Developers say the goal is not to match US or Chinese capabilities but to create smaller, localised systems that reflect Indian languages and values.
Singapore’s SEA-LION model, developed under the AI Singapore initiative, takes a similar approach. It focuses on underrepresented regional languages like Malay, Khmer, and Lao. Program head Leslie Teo says the aim is not to replace global models but to ensure Southeast Asia’s cultures are represented.
Beyond solo efforts, some experts are pushing for collective solutions.
Researchers at Cambridge University’s Bennett School for Public Policy have proposed “Airbus for AI” — a shared public company pooling the resources of middle-income countries such as Germany, Japan, and South Korea to build a competitive alternative to US and Chinese systems. The proposal has reportedly drawn interest from multiple governments and AI firms.
But critics argue that even multinational cooperation may not be enough to close the gap. Malaysian AI strategist Tzu Kit Chan cautions that many sovereign AI projects risk “burning through huge sums without tangible results.” He argues that governments would be better off investing in AI safety frameworks and governance instead of trying to outbuild global tech leaders.