The Sovereign AI Paradox: Why Countries Can't Go It Alone
Countries around the world are racing to build sovereign AI capabilities, seeking greater national control over their AI technologies and ensuring economic benefits by investing billions in domestic data centers, national AI models and computing infrastructure. But behind these ambitious efforts is the uncomfortable reality that the AI systems that countries want to control depend on a tightly interconnected global network of technology providers, semiconductor manufacturers, cloud platforms and communications infrastructure that are largely out of their control. At the 2026 Aspen Cyber Summit in Washington, D.C., international cybersecurity and technology leaders questioned whether countries can ever achieve true AI independence.
Governments increasingly view AI as critical to economic competitiveness, national security and technological leadership. But even countries spending heavily on domestic capabilities remain dependent on foreign-developed models, specialized chips, globally distributed manufacturing and infrastructure that can be disrupted by cyberattacks, geopolitical tensions or physical conflict.
This raises a fundamental question about what sovereign AI truly means. Does sovereignty mean building and controlling everything within a nation’s borders, or does it mean managing foreign dependencies, maintaining access to critical technology and ensuring resilience when that access is threatened?
Why Sovereign AI Still Depends On Foreign Technology
“Sovereign AI sounds great as a tagline, but what exactly are you sovereign about?” explained David Koh, Senior Advisor at Singapore’s Ministry of Digital Development and Information.
Koh questioned whether most countries have the resources to compete at the frontier of AI model development. Training and maintaining competitive foundation models requires substantial investment, specialized expertise, and access to advanced computing infrastructure.
The problem isn’t just the considerable cost of developing domestic models. There is the question of whether countries would accept potentially less capable systems for applications involving national security or other critical interests when more advanced alternatives are available.
This tension is already playing out in Europe. As I recently reported in Forbes , German search engine Ecosia moved away from French AI company Mistral in favor of European infrastructure provider Melious AI , which offers access to a range of open-weight models, including Chinese-developed alternatives.
Ecosia cited better performance, lower costs, and greater flexibility as reasons for the switch. The decision illustrates the dilemma Koh was describing. Countries and companies may want sovereign AI capabilities, but achieving the performance and economics they need can mean relying on technology developed elsewhere.
“So you have to get plugged into leading-edge models,” Koh said.
That creates a dependency from the outset. Countries can customize foreign-developed models, build domestic applications, and run AI systems in their own data centers, but the underlying technology may still be controlled by providers outside their borders.
“Now if someone decides to turn it off for whatever reason, then what happens to your country?” Koh asked.
The dependency extends beyond software. AI infrastructure relies on semiconductors designed by companies such as Nvidia, chips manufactured by companies such as TSMC, and semiconductor manufacturing equipment from companies including ASML.
Countries may operate domestic data centers and deploy open-weight AI models, but that doesn’t eliminate their reliance on international technology supply chains.
For most countries, building an entirely independent AI technology ecosystem is neither realistic nor necessarily the best use of national resources.
Australia’s Approach To Managing AI Dependencies
For Dr. Tobias Feakin, Managing Director of Protostar Strategy and Australia’s former Ambassador for Cyber Affairs and Critical Technology, the sovereign AI debate comes down to a question of acceptable dependency.
Speaking at the Aspen event, Feakin described the challenges facing governments across the Asia-Pacific region as they attempt to translate Sovereign AI ambitions into functioning national capabilities.
Many countries lack the resources to establish their own advanced AI infrastructure or develop competitive foundational models. They must make decisions about which international providers to rely on and how much control they need over the technology.
China’s growing availability of lower-cost, open-weight models further complicates these choices. Countries must evaluate performance, pricing, access, geopolitical considerations, and technology dependence when selecting which systems to adopt.
As Feakin explained, the issue is not simply whether countries use foreign technology, but which dependencies they are willing to accept.
This creates a difficult balancing act. A country seeking to minimize its reliance on U.S. technology providers may increase its dependence on Chinese systems, or vice versa. Deploying an open-weight model domestically can provide greater control over its operation, but still leaves dependencies on chips, infrastructure, model development, and software ecosystems.
For governments trying to strengthen their AI position, sovereignty becomes a question of deciding which capabilities are worth building domestically, which can be obtained through international partnerships, and what alternatives exist when access is disrupted.
In a follow-up conversation, Feakin pointed to research by the Tech Policy Design Institute , which has developed a methodology for assessing national AI capabilities and dependencies.
Rather than treating sovereignty as something countries either possess or lack, the institute’s Expanding AI Sovereignty to AI Agency framework examines what it calls AI agency, the ability of a country to influence its technological future, maintain strategic choices, protect national interests and capture economic value.
The framework evaluates 103 AI capabilities across six categories: infrastructure and resources, data assets and lifecycle management, models and applications, innovation and adoption, skills, and governance. It broadens basic concepts of sovereignty into four dimensions: control, access, choice, and leverage.
Control concerns the ability to govern capabilities directly. Access involves securing international partnerships and technology availability. Choice addresses resilience and the ability to switch between alternatives, and leverage reflects the ability to use national strengths to influence outcomes within an interconnected global technology ecosystem.
The institute’s interactive AI Agency assessment , developed in consultation with more than 250 experts, applies this methodology to Australia.
Based on its December 2025 assessment, Australia received an overall AI agency score of 10.5 out of 15. Its domestic AI maturity was classified as emerging, but it received high ratings for sovereignty and agency under the institute’s methodology. Australia’s assessment illustrates how countries can offset technological weaknesses with strengths elsewhere in the AI ecosystem.
The assessment identified strengths in areas including critical minerals, strategic datasets, computer vision model development, and international engagement. On the other hand, it identified particularly weak national capabilities in manufacturing and packaging AI accelerators.
This shows how countries can balance their weaknesses and strengths when it comes to AI. Australia may have limited domestic capabilities in some foundational AI technologies, but it possesses strengths that can provide influence and strategic options within the wider AI ecosystem.
How Estonia Protects Its Digital Independence
Estonia provides another example of how technology sovereignty can be strengthened without requiring everything to remain within national borders.
“For Estonia, digital sovereignty has never meant that we need to build everything ourselves. It’s about choices, capabilities, and also control,” said Ambassador Helen Popp, Estonia’s Ambassador-at-Large for Cyber Diplomacy, during the Aspen panel.
Estonia’s approach reflects decades of preparation for potential threats to its digital infrastructure, particularly given its geographic proximity to Russia.
Estonia established a Data Embassy in Luxembourg , under an agreement signed in 2017 and ratified in 2018, allowing critical government data and information systems to be maintained outside Estonia’s borders. The core idea is to preserve the continuity of government services in the event of a major disruption or attack affecting domestic infrastructure.
Instead of viewing foreign-based infrastructure as a loss of sovereignty, Estonia treats geographic distribution among trusted partners as a way to protect its ability to continue operating independently. This approach to sovereignty is more about resilience and diversification than domestic ownership.
A country may be better served by maintaining access to multiple technology providers, distributed infrastructure, and trusted international partnerships than by concentrating all its technology investments and resources domestically.
There are cultural dimensions, too. Popp pointed to the importance of protecting Estonia’s language, culture, identity, and democratic values as AI adoption expands. For Estonia, sovereignty involves maintaining the ability to make national decisions about technology, not necessarily developing every component independently.
The Infrastructure Risks Sovereign AI Can’t Escape
The Aspen discussion highlighted another vulnerability often overlooked in national AI strategies: physical infrastructure.
Feakin pointed to the increasing strategic significance of undersea communications cables in the Indo-Pacific, where damage and outages can threaten the connectivity supporting national economies.
Countries can invest in domestic AI data centers, computing capacity, and cloud services, but those investments remain dependent on reliable international connectivity and other physical infrastructure.
Cables, electricity supplies, telecommunications networks, and semiconductor supply chains can become points of failure regardless of how much domestic control a country has over its AI software.
Singapore’s experience illustrates the cybersecurity dimension of this problem.
Koh described how Singapore responded to a sophisticated cyber intrusion targeting its major telecommunications operators. The scale and complexity of the attack required extensive government involvement and cooperation with private telecommunications companies.
The lesson, he argued, was that neither governments nor private companies could manage such threats independently.
For sovereign AI initiatives, the implication is that national control over AI resources offers limited protection if the communications networks, power systems, or infrastructure supporting those resources remain vulnerable. Resilience requires understanding the entire chain of technological and physical dependencies, not merely the location of a country’s data centers.
The discussion at Aspen suggests that governments need to focus as much on the dependencies beneath their sovereign AI ambitions as on the capabilities they are building.
Owning a data center does not mean controlling the chips inside it. Hosting an AI model locally does not guarantee independence from its developer. Keeping data within national borders does not protect the networks and infrastructure required to access it. And building everything domestically may create vulnerabilities of its own if it concentrates critical infrastructure and eliminates alternatives.
If the objective of sovereign AI is technological independence, few countries are likely to achieve it. If the objective is reducing critical dependencies while preserving control, resilience, and freedom of choice, that is a more achievable goal.