Blog

Sovereign AI: Why the Answer Lies in the Ecosystem, Not the Models by Andrew Richards

August 10, 2026

There is a great deal of interest in Sovereign AI now. Nations around the world are concerned that if artificial intelligence (AI) is developed and controlled by a single nation, others may lose influence over how AI is deployed and governed within their own borders. In a world increasingly transformed by AI, there is a growing fear that control over critical technologies could become concentrated in the hands of a few countries. 

Governments concerned with AI sovereignty should focus less on the models themselves and more on the ecosystems that support them. Well-designed ecosystems enable innovation, choice, opportunity, education, and economic growth. Poorly designed ecosystems, by contrast, can limit innovation and restrict participation to those already in a leading position. 

My company, Codeplay, was a leader in developing open ecosystems that enable innovation in AI and scientific computing. This work helped the U.S. Exascale Computing Program build supercomputers for AI and scientific research using technologies from multiple hardware vendors. The effort expanded over time and is now managed through the UXL Foundation, a non-profit membership organization. We have also contributed to ecosystem technologies that address issues central to AI sovereignty, including safety, portability, and interoperability. 

Governments around the world should study how the U.S. Exascale Computing Program engaged with open ecosystems through funding, education, and technical contributions. Its objective was not simply to build powerful systems, but to ensure that multiple processor vendors could participate in the supercomputing infrastructure that increasingly underpins AI and scientific discovery. 

Over the past 15 years, advances in AI have been driven largely by deep learning models trained on massive datasets using enormous computing resources. Given this history, it is natural to assume that sovereignty is fundamentally about models and supercomputers. But is it? 

Stepping back, it is worth asking a simple question: what are we trying to achieve? 

There are many reasons governments may want to exercise authority over AI systems operating within their jurisdictions. There is legitimate concern about deepfakes that can deceive individuals and spread misinformation. If a human deliberately deceived others for gain, we would often classify that behaviour as fraud and regulate it accordingly. Similarly, if AI is used to control machines that could endanger people, such as vehicles, robots, or weapons systems, governments have an obligation to establish and enforce appropriate safety regulations. 

Yet none of these concerns are fundamentally about the models themselves. 

A model, in itself, is neither safe nor unsafe. A model does not commit fraud. It performs statistical calculations based on patterns learned from data. What matters is the broader system in which that model operates. 

AI systems include all the components required to deploy and use models in the real world. Critically, these systems connect AI models to devices, services, data sources, and users. It is through these connections that questions of safety, privacy, accountability, and governance arise. Whether the issue concerns the release of misleading information, the operation of potentially dangerous machines, or access to sensitive personal information, the relevant concerns exist at the system level, not the model level. 

This distinction is important because it helps clarify what sovereignty actually requires. 

A critical question for governments is whether AI deployment systems depend on technologies controlled by a single vendor or country. If they do, can a nation truly claim sovereignty over AI? I would argue that it cannot. If a critical component must be imported from another nation and cannot be substituted, then control ultimately rests elsewhere. 

It is tempting to assume that replacing a component with a domestically developed alternative is straightforward. In practice, the challenge is rarely the component itself. The challenge lies in how that component integrates into the broader system. Ensuring that different technologies work together requires open and functioning ecosystems. Without such ecosystems, nations can design as many individual components as they wish, but those components may struggle to interoperate with technologies from other vendors. 

There are many examples of governments getting this right. Modern communications networks are built on open standards and protocols that allow products from different vendors to work together. Examples include GSM for mobile communications and Wi-Fi for wireless networking. Governments often played a crucial role in nurturing and supporting the open standards that eventually became global norms. Countries that engaged early helped shape those standards, while those that arrived later largely adopted decisions made elsewhere. 

Many governments are understandably cautious about selecting technological “winners” because history contains numerous examples of unsuccessful industrial policy. However, creating an open ecosystem is not about choosing winners. Instead, it is about creating conditions in which multiple participants can compete and innovate. 

Every ecosystem requires decisions about interfaces, governance, and technical direction. The defining characteristic of an open ecosystem is that multiple implementations can coexist. Interfaces must be openly documented so that organizations can build compatible technologies. Equally important, there must be a clear intellectual property framework, so participants fully understand the legal implications of contributing to, adopting, and investing in the ecosystem. 

This clarity allows businesses to invest with confidence. 

The challenge for any company operating within an ecosystem is that it can be difficult to justify significant investments when competitors may also benefit from the resulting infrastructure. This is one reason government involvement can be so important. Public investment helps create the foundational capabilities that benefit entire industries rather than individual firms. 

The result is greater resilience and greater sovereignty. Rather than depending on a single vendor outside its control, a nation can support domestic companies that compete successfully in global markets. 

Open ecosystems provide another important advantage: they enable experts from different disciplines to work together. 

If we are concerned about safety, security, or privacy in AI, we need mechanisms that allow specialists in those fields to contribute their expertise. Such experts evaluate and audit systems. While models can be tested, measured, and analysed, governance concerns ultimately require examination of the complete system in which those models operate. 

It is through system-level analysis that experts can identify risks, assess mitigations, and establish appropriate controls. 

This collaborative approach can be seen in organizations such as the UXL Foundation, where members work on issues including testing, security, interoperability, and safety. Specialist groups bring domain expertise to challenges that are directly relevant to national sovereignty. Open ecosystems enable these discussions and evaluations to take place in public and across organizational boundaries. Without such openness, meaningful independent scrutiny becomes far more difficult. 

Governments should support these forms of public evaluation if they want to make meaningful progress on the issues that matter most in the AI era. Academic researchers, regulators, and technical specialists should apply the lessons learned from previous generations of computing systems to modern AI-based systems. 

There is no doubt that moving from public debates about AI to discussions about standards organizations, interoperability, programming models, and system architectures can feel like a dramatic shift. Political conversations about AI often focus on broad societal concerns, while the practical work of building trustworthy AI systems involves highly technical engineering decisions. 

However, if governments genuinely want to achieve Sovereign AI, they must bridge that gap. 

High-level policy objectives eventually need to be translated into technical requirements, standards, and ecosystems. While that can be challenging for policymakers and civil servants, supporting initiatives that connect experts with open technology ecosystems is essential for long-term success. 

Economically, engagement with AI systems ecosystems enables domestic companies to compete in emerging global markets. Despite appearances, many parts of the AI infrastructure stack remain relatively closed. Opening those systems is critical if nations want to develop competitive technologies, including AI accelerators and processors. 

The public benefits are equally important. Engagement with open AI ecosystems allows researchers, engineers, and regulators to conduct meaningful work on safety, security, and privacy. Those efforts produce tangible improvements as real-world AI systems become more trustworthy and accountable. 

The UXL Foundation provides a valuable starting point because it brings together many key components required for building AI systems within a single open-source ecosystem. It also provides a gateway to a broader collection of important standards and technologies, including programming models such as SYCL, intermediate representations such as SPIR-V, and safety-focused standards including MISRA C++ guidance for parallel programming. 

Each of these standards enables specialists to contribute their expertise while remaining part of a larger effort to build open, interoperable, and trustworthy AI systems. 

From the outside, this landscape can appear complex and intimidating. Yet the strength of open ecosystems is that they allow experts to focus on their individual disciplines while contributing to a broader shared objective. 

Relatively modest investments in specialist engagement with open ecosystems can deliver significant public benefits. For governments seeking genuine AI sovereignty, supporting these ecosystems may ultimately prove far more important than owning the models themselves.