By Devmini Bandara and Prihesh Ratnayake
Introduction
It’s 2026, and Artificial Intelligence (AI) is everywhere. From AI powered Chatbots to autonomous weaponry – AI is embedded in the systems that shape how people learn, work, communicate, access services, and even fight wars. US and China are the leaders in frontier AI models but the latter is rapidly expanding its influence by releasing capable open-weight AI models that are cheaper, more accessible, and increasingly adopted worldwide. This Sino-American tech race has resulted in reciprocal economic, political, and national sovereignty influence. As states increasingly rely on foreign AI systems, they are faced with one tough question – what is the impact of AI on sovereignty?
This is where the discussion on “Sovereign AI” comes in. The idea that nations should develop and maintain independent AI capabilities rather than rely on systems controlled by foreign entities or multinational corporations to serve national interests, local priorities, and societal values is a popular debate, but is this achievable? Against this backdrop, two critical questions emerge: Can states, particularly those in the Global South truly retain control over the full AI stack to reduce dependence on foreign AI models that compromise their sovereignty? If not, what is the solution?
Defining Sovereign AI
The term Sovereign AI is most closely associated with Jensen Huang, founder and Chief Executive Officer of NVIDIA. Huang played a key role in bringing the concept into mainstream technological and geopolitical discussions by arguing that countries should not rely entirely on foreign companies to develop and control the AI technologies that will shape their economies, security, and public services.
The term gained widespread attention during NVIDIA’s GPU Technology Conference (GTC) in March 2024, where Huang made Sovereign AI a central theme of his keynote address. Speaking to policymakers, researchers, and industry leaders, he argued that every nation should have the capability to build, deploy, and govern its own AI systems. While the phrase itself was relatively new, the underlying idea had already begun to take shape through NVIDIA’s AI Nations initiative, launched in 2019. But what is it? And why is there such a growing interest in the concept?
Sovereign AI refers to the ability of a state, and in some contexts a corporation, to independently develop, deploy, and govern artificial intelligence systems using infrastructure, data, and regulatory frameworks that remain under its own jurisdiction and strategic control. These include the full AI stack: data sovereignty, which involves controlling where data is stored, processed, accessed, and protected; compute infrastructure, through ownership or strategic control of data centers, cloud systems, semiconductors, and high-performance computing resources; AI models and systems, by developing locally governed and contextually relevant AI applications and foundation models; and regulatory frameworks, through the establishment of domestic rules governing AI deployment, risk management, accountability, and ethical oversight.
As AI becomes more deeply embedded in military operations, intelligence gathering, cybersecurity, critical infrastructure, and the economy, it is increasingly becoming a key element of national power. Governments are already using AI to strengthen cyber defences, detect threats more quickly, and improve intelligence analysis. So, having control over AI infrastructure, technology, and expertise is increasingly seen as essential to a country’s security and resilience.
The case of International Criminal Court judge Nicolas Guillou highlights the risks of relying on foreign technology providers. After being sanctioned by the United States over the ICC’s arrest warrant for Israeli Prime Minister Benjamin Netanyahu, Guillou reportedly lost access to major US based services including Visa, Mastercard, and other US based platforms, with no other practical alternatives. For many governments, the message is clear. If access to essential digital platforms can be switched off so easily, so too could access AI infrastructure and services. Sovereign AI is, in part, an attempt to reduce that vulnerability.
This dependency becomes a major vulnerability, nay a national security threat, as AI is now being integrated into defense systems, intelligence gathering, cybersecurity, financial services, and critical infrastructure. It is expected to play a major role in driving future economic growth and industrial competitiveness. For many governments, relying entirely on external providers for such strategically important technology is becoming increasingly difficult to justify.
Another important motivation is the preservation of language and culture and less biases. Many of today’s leading AI models are trained primarily on datasets reflecting Western languages and cultural norms, raising concerns that smaller languages and local knowledge systems may be overlooked. The datasets that are used to train these AI systems carry biases of both the developer and the data sets that were used. As a result, many Sovereign AI initiatives seek to develop AI models that better reflect national languages, local contexts, cultural values, and indigenous knowledge, helping ensure that outputs are more inclusive, accurate, precise and relevant to the societies they serve.
It is important to also acknowledge that the promotion of Sovereign AI also aligns closely with NVIDIA’s commercial interests. NVIDIA occupies a dominant position within the global market for AI-related computational hardware, particularly graphics processing units (GPUs) that underpin modern AI training and deployment. By framing AI infrastructure and computational capacity as strategic national assets analogous to public utilities or critical infrastructure, NVIDIA effectively positions its technologies as indispensable components of national AI strategies. From this perspective, Sovereign AI functions simultaneously as both a legitimate geopolitical framework and a commercially advantageous narrative that reinforces demand for NVIDIA’s products and services. Understanding Sovereign AI therefore requires an appreciation of its complex intersection with geopolitics, technological governance, national security, and global political economy.
The Myth of Sovereign AI: Who holds power?
The question of whether AI sovereignty is a myth or a reality lies at the heart of contemporary tech diplomacy. This question is also deeply nuanced and complex. Governments globally are on track to invest over $1 trillion by 2030 to achieve full-stack technological independence, this includes massive initiatives like the U.S. spending $12 billion on Arizona chip production, the EU committing $50 billion, and China investing $150 billion to replicate advanced lithography. However, the hyper-specialized and deeply interconnected nature of the modern AI supply chain suggests that absolute digital autarky is an illusion. Pursuing it forces nations to sacrifice innovation speed and competitive advantages.
Firstly, owning every component is neither necessary nor practical. production of frontier artificial intelligence relies on complex, global choke points that no single nation can completely replicate or isolate, ranging from the hyperpure polysilicon distributed across the U.S., Germany, and Japan, to the monopoly on extreme ultraviolet (EUV) lithography held by ASML in the Netherlands, to the cutting-edge semiconductor fabrication dominated by Taiwan’s TSMC. In this globally interconnected supply chain, the idea of owning every component from silicon chips to the software libraries seems completely unrealistic.
Secondly, because these distinct nodes continuously co-evolve through cross-border collaboration, any state-led attempt to build an entirely enclosed domestic stack forces a country to chase a continuously moving frontier, risking a sacrifice in both innovation speed and competitive advantage.
Thirdly, the cost of cutting-edge AI foundation models is prohibitively expensive and complex for all but a handful of global tech giants. It would be nothing but a distant dream for underdeveloped and developing nations to adopt full AI sovereignty.
True AI sovereignty remains a myth in terms of total and absolute self-reliance; what is more practice is to redefine AI sovereignty not as total independence, but as “strategic autonomy“( selective autonomy) where a nation secures geopolitical leverage and a seat at the global table by mastering a narrow, indispensable niche within the broader interdependent ecosystem. A country could strategically decide which workloads must run on sovereign AI infrastructure, and which are acceptable for public services.
The Path Forward
For many countries on the global south, Sovereign AI is less a realistic policy goal than an aspirational vision. The enormous financial and technological barriers are simply too high. Building sovereign AI requires billions of dollars in data centres, advanced computing infrastructure, semiconductor technologies, and a highly skilled workforce, resources that remain beyond the reach of many developing nations. At the same time, the world’s AI leaders and technology giants have little incentive to share their most advanced models, proprietary technologies, or strategic expertise. Their latest AI systems are increasingly treated as strategic assets that confer economic, technological, and geopolitical advantage. From a realist perspective, this is entirely predictable. States and corporations act to protect their competitive edge, not to empower potential rivals.
For developing nations, India’s “India AI Mission” offers an exemplary blueprint to follow. India demonstrates how a state can build contextually relevant AI capabilities without owning every layer of the global hardware stack. It achieves this by establishing state subsidized, public-private compute infrastructure with empaneled GPUs. Furthermore, the country actively supports open multilingual datasets through initiatives like Bhashini while backing domestic startup innovation.
While replicating India’s national expenditure is unfeasible for smaller economies, they must pursue a pragmatic strategy grounded in South-South cooperation and targeted South-North alliances. Through South-South partnerships, regional blocs can pool financial resources to build shared compute facilities, create collective data trusts, and fund the localized fine-tuning of open-weight models. Simultaneously, strategic South-North alliances allow developing states to negotiate hybrid deals with global technology giants. By exchanging access to local markets and demographic data, states secure subsidized cloud infrastructure, technology transfer, and workforce development, all without bearing the multi-billion-dollar investment on infrastructure.
Absolute independence in AI is a luxury reserved for a select few. For smaller and developing states, the real test is deciding which parts of their digital sovereignty they are willing to trade, and with whom, to guarantee a seat at the AI table. Navigating that compromise with intention, rather than accepting passive dependency, will determine who thrives in the emerging digital order.
Devmini Bandara is an Attorney-at-Law and Factum’s Manager: Legal, Research and Policy, with interests spanning law, public policy, governance, digital rights and institutional reform.
Prihesh Ratnayake is a Data Scientist working as Factum’s Research Specialist in AI and Data Science with expertise in tech policy, platform accountability and digital rights.
Factum is an Asia Pacific-focused think tank/consultancy on Diplomacy, Tech-Plomacy, Digital and Energy Futures accessible via www.factum.lk.
The views expressed here are the author’s own and do not necessarily reflect the organization’s.
