Artificial intelligence is increasingly being viewed as critical national infrastructure. As governments across Asia weigh their AI strategies, the debate is shifting from whether to adopt AI to how much control countries should retain over the models powering it.
Discussing the shifting tech landscape and the growing importance of sovereign AI with iTNews Asia at the ATxSummit, Professor Mohan Kankanhalli, Director of the NUS AI Institute and Deputy Executive Chairman of AI Singapore, says AI innovation is now core to national policy. He also explores the trade-offs nations must make when navigating between building and buying their AI models.
While many nations and corporations have historically been content to adopt ready-made tech infrastructure from foreign hyperscalers, the increasing use of advanced AI is forcing an urgent re-evaluation of how and where nations should invest in their AI technology stack.
Kankanhalli explained that true sovereignty is no longer just about where data is geographically stored, but about who owns and builds the underlying models that process it. He defines this as the ability to create and govern the full AI value chain – from generating and safeguarding data to building, deploying and continuously improving models, while retaining the economic and societal benefits that flow from them.
“I think it’s a combination of producing data, building AI models, exploiting AI models, improving the AI models, and getting the upside benefit of AI,” he said.
The cost and risks of dependence
Kankanhalli argues that reliance on foreign AI models creates a self-reinforcing cycle of dependence that permanently damages a nation’s economic and technological future. The concern, he explained, is that countries that rely exclusively on foreign models risk becoming permanently dependent.
As citizens and enterprises use these systems, their interactions and data help make overseas models more capable, while domestic ecosystems struggle to catch up.
This creates a state of "digital colonisation" where domestic companies are locked out of the foundational AI game because they lack the pre-training data to build competitive models, rendering their own technologies permanently inferior,” he explained.
Furthermore, he warned that this dependency leaves an economy incredibly fragile.
If a foreign corporation or government suddenly decides to restrict access to their model, as occurred when Meta altered its Llama strategy, local enterprises that built their applications on top of that tech are left completely stranded with nowhere to turn.
- Professor Mohan Kankanhalli, Director of the NUS AI Institute and Deputy Executive Chairman of AI Singapore
For this reason, he believes that sovereignty must be considered at the model layer, not just at the data layer. Data residency laws alone cannot guarantee sovereignty if the intelligence processing that data resides abroad.
Should we build, buy or blend?
Kankanhalli stressed, however, that sovereign AI does not necessarily mean every country must build the world’s largest frontier model. Nations and corporations navigating sovereign AI can choose between three distinct pathways identified by the ACM Global Technology Policy Council – buying a model, building a model, or adopting a hybrid "blend" approach.
Each strategy presents its own professional trade-offs. Buying an existing model allows an organisation to completely avoid upfront development costs, though it introduces long-term dependency and data exposure risks.
In contrast, developing a homegrown model demands sustained investment in compute, research and talent, but it enables organisations and nations to retain control of the AI stack while safeguarding the value and sovereignty of their data.
Alternatively, a viable hybrid mode allows an organisation to blend these strategies by initially buying a model to deploy applications quickly and then systematically moving toward building its own capabilities over time.
Kankanhalli said nations should pursue strategies aligned with their capabilities and priorities. Smaller and mid-sized models tailored to sectors such as healthcare, finance or logistics can deliver significant value while fostering local expertise.
For instance, he pointed out that Singapore's approach reflects this philosophy. Through AI Singapore, the country is developing the Sea-Lion family of Southeast Asian language models while also investing in specialised medical AI built on local healthcare data.
Why AI is a national security imperative
The professor also sees a growing national security dimension to the debate. Frontier AI models are becoming increasingly adept at coding and could eventually be used to discover software vulnerabilities or mount cyberattacks. In such a world, countries without comparable defensive capabilities may find themselves at a strategic disadvantage.
“Countries which possess these frontier models can attack the cyber infrastructure of other countries not having such a model,” Kankanhalli said. “In order to defend from such models, you need to have an equally powerful model.”
He added that nations must either develop these capabilities themselves or establish deep trust with partners willing to provide access to advanced defensive models.
Kankanhalli also cautioned that countries lacking advanced AI capabilities risk falling behind in scientific discovery and intellectual property creation, as agentic systems increasingly accelerate research in fields such as materials science and healthcare.
He said sovereign AI is now a prerequisite not only for protecting critical infrastructure, but also for safeguarding a nation’s future capacity for innovation and strategic autonomy.
Moving beyond a "one size fits all" strategy
How should a nation plan its AI innovation and ensure its independence and control?
Kankanhalli advised that achieving AI sovereignty does not require every country to replicate massive, multi-billion-dollar frontier stacks from scratch. Instead, the strategic imperative dictates that nations look inward to identify their specific geopolitical and structural advantages, optimising their resources accordingly.
Rather than viewing the proliferation of national AI strategies as a fragmentation of technology, Kankanhalli frames it as a healthy development that introduces crucial redundancy into global supply chains. He does not agree with the notion that the rise of sovereign AI will splinter the global AI ecosystem into incompatible silos.
Instead, he sees the emergence of multiple national and regional models as a source of strength, one that will better reflect the world’s linguistic and cultural diversity while reducing concentration risk. "We need multiple providers of models, multiple languages and cultures covered by the models... If there is only one provider or two providers of the model, I think there is a supply chain risk out there."
Ultimately, a broader array of AI providers and specialised models, Kankanhalli argues, will make the global AI landscape more resilient, foster innovation, and ensure that no single country or company becomes an indispensable gatekeeper of this foundational technology.
