Governments worldwide are funneling billions into their own homegrown AI technologies— but is this colossal investment really justified, or just a costly misstep? The race to develop so-called "sovereign AI" is heating up, with nations vying for autonomy in a field dominated by a few tech titans. But here's where it gets controversial: smaller countries pouring money into these projects may be facing an uphill battle against industry behemoths, raising questions about the real value of such efforts.
Take Singapore, for instance. Its government-backed AI model impressively communicates in 11 languages, from Bahasa Indonesia to Lao, addressing the rich linguistic diversity of the region. Malaysia's ILMUchat, crafted by a local construction giant, prides itself on nuanced understanding—for example, knowing the exact Georgetown being discussed: the Malaysian city in Penang, not the American university. Across Europe, Switzerland’s Apertus, launched in September, showcases deep cultural language awareness by correctly choosing between the Swiss German "ss" and the German "ß." These examples illustrate a growing trend: tailoring AI to capture local languages and cultural subtleties that global models often overlook.
This movement is part of a vast, globally escalating AI arms race valued at hundreds of billions of dollars, largely driven by dominant players based in the US and China. Tech giants such as OpenAI, Meta, and Alibaba are investing enormous sums to develop increasingly powerful AI models. Amidst this landscape, middle powers and developing nations watch closely, sometimes placing their own, financially hefty bets to carve out a space in the AI ecosystem.
These ventures fall under "sovereign AI," a strategy where countries like the UK, India, and Canada strive to build their own cutting-edge AI technologies or position themselves strategically in the sector. But here's the million-dollar question: Can relatively small-scale investments truly make a meaningful impact when competing with the massive budgets and resources of global leaders?
"US-based companies and the government, as well as China, are able to surge aggressively toward AI dominance," explains Trisha Ray, a fellow at the Atlantic Council, a strategic think tank. "But for smaller or middle powers, building a large language model (LLM) from the ground up is an enormous challenge and financial burden."
And yet, for many countries, relying on foreign AI solutions isn’t an option, especially for sensitive applications like national defense. India, which ranks as the second-largest market for OpenAI users globally with over 100 million ChatGPT downloads in recent years, has encountered limitations with imported AI systems. Abhishek Upperwal, founder of India’s AI developer Soket AI, points out real-world issues: an AI tutor deployed in a remote Telangana village spoke English with a heavily American accent, hindering understanding, while an Indian legal tech company’s attempt to retrofit Meta’s LLaMa model resulted in confusing, mixed US-Indian legal outputs.
Security concerns loom large too. For India’s defense ministry, China’s AI model DeepSeek is a non-starter. Upperwal highlights the risks: "That model could be trained on datasets questioning India’s territorial claims, such as regarding Ladakh, making its use in defense unthinkable." Even US-sourced AI tools face distrust; defense personnel worry about data sovereignty and potential leaks if relying on overseas providers.
In response, India is backing domestic development. Soket AI is among select companies building a national LLM under the government-funded IndiaAI Mission, which has committed about $1.25 billion to AI innovation. This model will be smaller than the colossal versions produced by US or Chinese firms, perhaps akin to those from French startup Mistral. AI experts agree that immense investments in computing resources and specialized chips are fundamental to pushing AI frontiers toward artificial general intelligence (AGI), the ultimate yet elusive goal. Hence, India plans to leverage its pool of talented engineers and scientists to compensate for funding limitations.
"We can’t match billions being poured in by the US," Upperwal acknowledges. "So the edge has to come from brainpower and expertise."
Singapore offers another illustrative case with AI Singapore supporting SEA-LION, a family of language models focused on Southeast Asian languages often sidelined by major Western and Chinese AI. Leslie Teo, senior director of AI Singapore, explains these models aren’t intended to replace giants like ChatGPT or Gemini but to complement them by embedding cultural nuances. For example, a generic AI might awkwardly use overly formal Khmer or suggest pork recipes to Malaysians, missing local contexts entirely. By developing regional-language models, governments can become smarter consumers and more representative players in global AI.
Teo cautions against overhyping sovereignty: "Our goal is better representation and understanding, not claiming absolute control."
Beyond individual national efforts, there’s growing interest in multinational collaboration. Researchers at Cambridge’s Bennett School for Public Policy recently proposed forming a public AI company run by a consortium of middle-income countries. Dubbed "Airbus for AI," this concept draws inspiration from Europe's success in challenging Boeing in the 1960s by pooling resources across multiple nations—including the UK, Spain, Canada, Germany, Japan, Singapore, South Korea, France, Switzerland, and Sweden—to create an AI competitor to US and Chinese corporations.
Joshua Tan, lead author of the proposal, notes it’s already attracting attention from AI ministers in several countries and has sparked interest from sovereign AI initiatives itself. Even developing nations such as Mongolia and Rwanda have shown enthusiasm. "Trust in current US administration’s tech promises is declining," Tan observes. "Countries ask, ‘Can we really depend on this technology or might access be cut off?’"
Although optimistic about collaborative potential, some remain skeptical. Tzu Kit Chan, a Malaysian AI strategist, warns that despite the appeal of a multi-country approach, poorly conceived sovereign AI programs can squander public funds and ultimately flounder.
"Governments need to grasp just how fast and far AI technology is advancing," Chan stresses. "The real risk is that flawed strategies lead to wasteful spending with little to show for it."
Chan advocates a different approach: focusing on strengthening AI safety regulations rather than trying to outdo established international products that dominate the market. He cites a telling anecdote from Kuala Lumpur: "If you ask the average finance professional what AI model they use, eight out of ten say ChatGPT or Gemini—not any sovereign solution."
This raises a provocative question: Are sovereign AI projects genuine efforts to foster independence and innovation, or are they mainly symbolic, costly endeavors that miss the bigger picture? Should governments double down on developing proprietary AI, or would their resources be better spent on regulation, education, and cooperation? The debate is wide open, and the stakes have never been higher.
What’s your take? Do you believe sovereign AI is a strategic necessity for national security and cultural preservation, or is it a misguided race doomed by scale and speed? Share your thoughts and join the conversation below.
