SILICON VALLEY — With the United States and China holding the upper hand in artificial intelligence, countries across Asia, Europe and the Middle East — South Korea among them — are calling for "sovereign AI." The term is generally taken to mean building an AI governance structure that a nation can control on its own, spanning the pyramid from the semiconductors at AI's foundation, such as processors and memory, through data centers, software and applications. Sovereign AI is rendered as AI "sovereignty" because the underlying concern is that a country should hold decision-making authority over its entire domestic AI industry.
The Lee Jae-myung administration is pursuing an independent AI foundation model project as a national policy initiative and is also moving ahead with a new semiconductor cluster in the Honam region. In July, the president met U.S. AI leaders in San Francisco — including Nvidia Chief Executive Jensen Huang, Broadcom CEO Hock Tan, OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei — and held an AI summit. The gatherings were part of a strategy to realize sovereign AI.
No country, however, has built a complete AI pyramid. The United States, called the strongest AI power, is home to formidable hardware and software companies such as Nvidia, AMD, Amazon Web Services, OpenAI, Anthropic and Google, yet it depends on Samsung Electronics (005930.KS) and SK hynix (000660.KS) for the most advanced memory, including high-bandwidth memory. China, which is contesting U.S. leadership, has its government nurturing domestic players such as Huawei, but it still runs into limits in developing cutting-edge AI models without AI chips from Nvidia and AMD. Full AI independence is that difficult a task.
Sovereign AI is usually imagined as a system in which decision-making authority rests on a self-sufficient base. But no country can fully control AI chips, memory, models and power all at once. Building an AI system that depends on no particular country or company is the ideal of sovereign AI, yet it is hard to achieve in practice. Industry and academia therefore advise combining multiple sources of infrastructure and making use of open ecosystems. The following draws together a panel discussion at AMD Advancing AI 2026, held in San Francisco from July 22 to 23, and a report from the Stanford Institute for Human-Centered AI (HAI).
Thomas Zacharia, senior vice president for global public sector and strategic partnerships at AMD, moderated a panel titled "From Exascale to Sovereign Intelligence: Powering National AI" at AMD Advancing AI 2026 in San Francisco on the 23rd. Exascale refers to computing performance capable of a billion billion operations per second. Joining Zacharia were Trish Damkroger, senior vice president at HPE; Bruno Lecointe, senior vice president at Bull; Stephan Lequena, chief technology officer at Jensy; and Arjun Shankar, director of the National Center for Computational Sciences at Oak Ridge National Laboratory.
Zacharia asked the panelists what AI sovereignty means. "Sovereignty means you have to have ownership of your own systems," Damkroger said. "It includes not just the infrastructure but the data and the outputs of the models built on that data." Lequena said it means "covering the entire life cycle of AI, starting from the infrastructure," adding that "what matters is being able to manage the data while providing access to HPC, or high-performance computing, and to the data." Up to this point, the definitions tracked what most people have in mind.
Then came the argument that AI sovereignty means AI collaboration. "AI sovereignty means openness to collaboration," Lecointe said. "To make predictions in AI you have to use energy, supply power, build data centers and secure the right software, but you also need the right use cases. The only way to achieve that is not to think you can do everything alone." He added: "If there is someone who does it better than you, and they are willing to work with you, you absolutely have to join forces. In AI you end up needing computers, and sovereignty is likely to be something far more open."
Summarizing the panelists' answers, Zacharia asked: "AI sovereignty does not mean that an institution, a company or a country owns it from beginning to end. That is because you want to have the best stack. So isn't this a matter of choice?" Damkroger agreed, replying: "There are different businesses, especially in regulated corporate areas. So the ability to choose the components you want — hardware, software, storage, networking — is absolutely essential." Lequena said, "AI sovereignty is not isolation. No one can handle everything perfectly. It is a mistake to view isolation favorably. The digital economy shows that we absolutely have to cooperate."
Lecointe cited China as the leading example of isolation. China is trying to be self-sufficient in everything from semiconductors to AI software to supercomputers, he said, but only because it lost its options after being isolated internationally by U.S. restrictions and other measures. "In the world, the good example of isolation is China," he said. "They announced the No. 1 computer system on the Top500. They are showing the world that they are trying to do everything themselves because they could not get access to U.S. and European technology. It is because they had no other choice." According to Top500, the global nonprofit, China's LineShine supercomputer took first place for the first time since 2017 on the Top500 list of the world's most powerful supercomputers, released in June at the ISC 2026 conference in Hamburg, Germany.
Lecointe, who oversees HPC, AI and quantum computing at the French technology company Bull, said Europe is falling behind Chinese rivals but is surviving the competition and building AI sovereignty through collaboration. "We have a big competitor in China. That is a worry for us," he said. "We do not have a market as large as China's or the United States', so we have to cooperate and take a different approach. For example, we depend on U.S. chips, but we are not afraid. Sometimes decisions may need to be revised, but I think we can always find a way to cooperate."
As the view spreads that AI self-sufficiency is unattainable in practice and that control over system operations is the core of AI sovereignty, U.S. Big Tech firms are accelerating the commercialization of sovereign AI. Nvidia is the prime example. At an AI conference in Paris in November 2023, Nvidia Chief Executive Jensen Huang began making the case for sovereign AI, saying every region and every country should build its own AI system. In a July report, Stanford HAI said: "At the heart of the AI sovereignty debate is securing domestic access to the physical infrastructure that underpins AI development — data storage and processing facilities and the chips needed to train and run models. Without trustworthy AI infrastructure, many ambitious AI sovereignty strategies are bound to remain aspirations." The report added that "Nvidia has established a substantial presence across North America, South America, Europe, Asia and Africa through its AI factory program."
The same applies to hyperscalers that run large-scale computing infrastructure, including Amazon Web Services, Microsoft and Google. As the cloud has taken its place alongside physical infrastructure as a central pillar of AI, governments and companies are competing to secure cloud capacity for data processing. At the same time, as giant U.S. hyperscalers have come to dominate the cloud market, concerns about vendor lock-in and data leakage have grown, increasing the need for control over cloud environments. Countries are therefore building separate local clouds or working with hyperscalers to create sovereign cloud environments. In particular, as heavily regulated European governments push to expand the use of domestically produced clouds, U.S. providers are developing clouds dedicated to Europe.
Sovereign AI is an important market for AI model developers as well, because models are the applications that actually run on AI infrastructure. The AI services that ordinary consumers encounter directly are models such as OpenAI's GPT, Anthropic's Claude and Google's Gemini. The problem is that major U.S. and Chinese companies dominate AI models too. For governments, having to rely on systems trained on foreign data is uncomfortable.
OpenAI is accordingly seeking to deliver localized model sovereignty through OpenAI for Countries, which supports governments in customizing OpenAI models for local languages and cultures. It helps them use tools based on ChatGPT, OpenAI's chatbot, to integrate public-sector services, for example. The core of sovereign AI, the report concluded, is not eliminating dependency but calibrating interdependence.
