The discourse surrounding "AI sovereignty" has become a central pillar of geopolitical strategy. From Brussels to Washington and Beijing, leaders speak of autonomy as a prerequisite for national security. However, as discussed in the second episode of the State of the Globe podcast miniseries, the idea of total AI independence is largely an illusion. We are not witnessing a race toward self-sufficiency, but rather a deepening of interdependent vulnerabilities. To understand who is winning or losing, we must look beyond the models themselves and examine what Jensen Huang, the Founder and CEO of NVIDIA, calls the "five-layer cake" of AI: energy, chips, infrastructure, models, and applications.
In the early innings of the AI revolution, the focus was almost exclusively on chips, specifically those designed by NVIDIA. However, the bottleneck has shifted. The primary constraint is no longer just the chips, but the energy required to power them. Europe finds itself in a particularly precarious position: we occupy the region with the highest energy prices globally, yet we are trying to compete in a field that is fundamentally powered by cheap electricity. While the U.S. struggles with permitting and infrastructure aging, China is building more gigawatts of power, both renewable energy and nuclear energy, than any other nation. In this layer of the stack, sovereignty is not about code; it is about the grid, and how fast it can be modernized.
A second layer of the illusion is the belief that manufacturing sovereignty is within reach. Currently, 90% of the world’s leading-edge AI chips are manufactured by a single company, Taiwan Semiconductor Manufacturing Company (TSMC), which is based on a single island that is close to China: Taiwan. Even the U.S. and China, for all their rhetoric, are decades away from replicating the precise ecosystem of TSMC.
Furthermore, "chip sovereignty" is a misnomer because the supply chain is a web of thousands of providers. ASML in the Netherlands is the only company capable of producing the extreme ultra-violet lithography machines required for leading-edge chips, yet they depend on over 5,000 suppliers. For example, one critical supplier is Germany’s ZEISS, which provides ASML with high-precision lenses. Without a company like ASML and its intricate web of suppliers, the entire global AI stack, including the American and Chinese frontier model providers, would grind to a halt.
While the U.S. currently leads in closed-source frontier models (like OpenAI’s GPT-4 or Anthropic’s Claude), China has strategically pivoted to dominate the open-source layer. By releasing capable models for free, Chinese firms are ensuring that developers in Southeast Asia, Brazil, and India build their ecosystems on Chinese architectures. This is a digital version of the "Belt and Road Initiative": providing the infrastructure for free today to ensure technological dependence tomorrow.
Where does this leave Europe? We often frame our role as the "world’s regulatory superpower," but regulation in the absence of industrial strength risks becoming a menu from which others choose while we remain hungry for innovation. The recent "watering down" of the EU AI Act through the digital omnibus reflects a growing anxiety that stringent rules may strangle what little innovation we have left in Europe.
Instead of chasing the pipedream of AI sovereignty, which we lack the compute and energy to power anyway, Europe should focus on a strategic pivot:
Ultimately, no single nation can achieve absolute AI sovereignty. Even the U.S. is missing the energy layer, and China is missing the specialized semiconductor manufacturing equipment that ASML provides. The path forward is not through isolationism, but through a coalition of middle powers. Europe can close the gap with the U.S. and China in an AI race where they are the only players that matter, by aligning with like-minded partners Taiwan, Japan, South Korea, and the UK. Governance in the age of AI is no longer just about ethics and privacy, but increasingly about managing a fragile, global tech stack where a producer of extreme ultraviolet lithography (EUV) machines from the Netherlands (ASML), and a foundry from Taiwan (TSMC) form the foundation of the global AI hardware stack, acting as an absolute bottleneck for the entire industry and AI development more broadly.
David Timis is a Senior Fellow in AI Governance at the Global Governance Institute (GGI) and a manager at Generation.org. This commentary is based on the first episode of the GGI podcast miniseries "The Governance of AI."
Photo credit: NVIDIA founder and CEO Jensen Huang in conversation with Larry Fink, chair and CEO of BlackRock, at the World Economic Forum Annual Meeting 2026 in Davos, Switzerland. Image credit: World Economic Forum / Thibaut Bouvier
