The race for sovereign AI

Artificial intelligence is rapidly becoming a strategic technology that governments view as central to economic competitiveness and national security. As governments seek greater control over the systems expected to underpin future economies, a growing number are investing in what has become known as sovereign AI. Rather than relying on foreign cloud providers and commercially available AI systems, sovereign AI initiatives seek to build domestic computing infrastructure, foundation models and data ecosystems that allow countries to develop and deploy artificial intelligence on their own terms.

The race has accelerated rapidly. The Center for a New American Security's April 2026 Sovereign AI Index identified approximately 135 sovereign AI projects announced worldwide between early 2023 and the end of 2025, illustrating how quickly governments have embraced AI as a strategic capability rather than simply another digital technology. Yet the report also reveals a striking contradiction. While countries are investing billions to strengthen their technological independence, 56 percent of tracked infrastructure and foundation model projects involve at least one American technology partner, exposing the continued dominance of U.S. companies across the global AI ecosystem.

The findings suggest that sovereign AI is becoming less about technological self-sufficiency and more about securing the infrastructure, talent and computing capacity needed to reduce strategic dependence.

Defining sovereign AI

The concept of sovereign AI has gained momentum as governments seek greater control over the technologies expected to underpin future economies. The CNAS index defines sovereign AI projects as government-backed initiatives designed to strengthen domestic AI capabilities. These include investments in AI supercomputers, sovereign cloud infrastructure, domestic foundation models and national datasets intended to ensure critical AI capabilities remain under national control.

The pace of those investments has accelerated dramatically. The number of announced sovereign AI projects grew from virtually none at the beginning of 2023 to roughly 135 by the end of 2025. Infrastructure accounts for the largest share of activity, representing 59 percent of all tracked projects, followed by foundation models at 34 percent and national data initiatives at 7 percent.

The global distribution of sovereign AI investment furthermore differs markedly from earlier waves of digital innovation.

According to the CNAS index, the Middle East leads disclosed sovereign AI investment with approximately $41 billion, ahead of East Asia with $25.2 billion and the European Union with $4.3 billion. At the national level, the United Arab Emirates accounts for $33.5 billion in announced sovereign AI investment, followed by Japan with $20.4 billion and Saudi Arabia with $5 billion.

Those figures reflect distinct national strategies rather than a single model for AI development.

In the Gulf, sovereign AI has become closely linked to broader economic diversification efforts. Governments are investing heavily in data centres, compute infrastructure and technology partnerships to diversify their economies and establish regional AI hubs. Across East Asia, countries including Japan, South Korea and Taiwan are building on strengths in advanced manufacturing, semiconductors and digital infrastructure while expanding domestic AI capabilities.

The result is a global competition in which governments are investing not only to accelerate innovation but also to secure greater influence over technologies expected to underpin future economic growth.

Compute has become the foundation

Much of the public conversation surrounding artificial intelligence focuses on chatbots and foundation models. Governments, however, are increasingly concentrating on the infrastructure that makes those systems possible.

The Organisation for Economic Co-Operation and Development’s (OECD) A Blueprint for Building National Compute Capacity for Artificial Intelligence argues that compute has become a fundamental building block of national AI capability and should be evaluated through three dimensions: capacity, effectiveness and resilience. Capacity refers to the availability of computing resources, effectiveness to their ability to support research and commercial deployment, and resilience to maintaining compute despite geopolitical tensions, supply chain disruptions or surging demand.

Building national compute capacity therefore extends well beyond purchasing advanced graphics processing units. Governments must also secure reliable electricity supplies, expand high-speed networking infrastructure, cultivate specialised technical talent, develop sustainable financing models and maintain access to semiconductor supply chains. Weakness in any one of those areas can constrain the entire AI ecosystem.

The CNAS findings reflect that shift in priorities. Infrastructure projects have consistently outpaced investments in foundation models and data initiatives, with the strongest acceleration occurring during 2025.

Independence built on interdependence

Perhaps the report's most revealing finding is that sovereign AI remains deeply dependent on international technology partnerships.

Despite the political emphasis on technological independence, 56 percent of tracked infrastructure and foundation model projects involve an American technology partner. By comparison, only 2.4 percent involve a Chinese partner, while 3.2 percent involve both American and Chinese companies.

The dependence extends beyond infrastructure.

Meta's Llama family underpins 14 sovereign AI model projects, making it the most widely used foundation model in national AI initiatives. France's Mistral supports seven projects, Google's models appear in six, Alibaba's in five, while OpenAI's models support three.

Rather than building entirely independent AI ecosystems, many governments are adapting existing foundation models to national languages, public services and regulatory requirements while investing in domestic infrastructure.

Complete technological independence remains difficult because advanced semiconductors, cloud platforms and frontier models remain concentrated among a relatively small number of companies. Building sovereign AI therefore often means combining domestic investment with carefully selected international partnerships rather than eliminating foreign technology.

The first generation of sovereign AI projects suggests that technological sovereignty is unlikely to mean complete independence from global technology markets. Instead, it is becoming a question of strategic autonomy. Countries may continue relying on international partnerships while seeking sufficient domestic capability to ensure that decisions about how artificial intelligence is developed, deployed and governed ultimately remain their own.