Digital Sovereignty as a Collective Endeavor
Digital sovereignty is not about isolation. It is the capacity to make autonomous choices while building coalitions with like-minded partners. Europe aims for a third way that combines innovation with safeguards, openness with resilience, and competitiveness with fundamental rights.
China vs. US: Two Different Races
China’s rise in AI has been faster than most Western observers expected. Xi Jinping’s 2015 goal to become an innovation leader by 2030 is well on track. Western audiences often believe the US is winning, but that perception relies on US-centric models. In reality, 50% of AI developers are Chinese, and 70–80% of employees at US firms like xAI are Chinese. The 2025 DeepSeek moment — a small team with limited compute challenging US models — demonstrated China’s engineering ingenuity under chip restrictions. Today, 40% of frontier models are Chinese, and China leads in AI patents.
The Four A’s and I’s: Contrasting Strategies
The US and China follow fundamentally different AI strategies, captured in four contrasts:
- Accessibility: China open-sources models (DeepSeek, Qwen); the US builds IP moats.
- Affordability: China targets low-cost, high-volume markets (Global South); the US relies on subscription models to recoup huge investments.
- Applicability: China goes vertical (industry-specific, optimized stacks); the US goes horizontal (hyperscalers, general-purpose).
- Augmented vs. Imposing: China’s government promotes collaboration and AI for good; the US aims for technological supremacy and winner-takes-all.
Regulation also differs: China treats code as law (engineering mindset, binary danger assessment, accountability of all actors); Europe treats law as code (legal responsibility, provider-centric, risk levels). The conclusion: China is democratizing AI; the US is monopolizing AI. Europe should not fear China as an iceberg but work with it, learning from its speed and adaptability.
Silicon Valley’s View: Extreme Acceleration
AI is becoming infrastructure, like electricity. The shift is from experimentation to execution. Three forces drive this: massive capital inflow, extreme competition (model iteration in weeks), and growing pressure for profitability. The hype cycle is resetting — AI must now produce measurable outcomes, not just slides. Agentic AI enables small teams to achieve massive output (e.g., one person + 30 agents doing what 10 engineers did in months). The concept of vibe coding has evolved into vibe working: AI becomes a digital co-worker, not just a tool. Startups like Open Clo reach billion-dollar valuations with tiny teams. For governance, this raises urgent questions about accountability, taxes, and productivity models. Europe can lead in trust and operability rather than compete on speed.
European Commission’s Perspective: Excellence, Trust, Engagement
The European Commission’s AI strategy rests on three pillars:
- Excellence: investment, R&D, AI factories and gigafactories.
- Trust: the AI Act provides guardrails.
- International Engagement: bilateral networks, UN, OECD, G7.
The AI Act addresses complexity, opacity, autonomy, and risks to safety and fundamental rights, while preventing fragmentation of the single market. The third way argues that trust is necessary for adoption, and adoption yields benefits. Europe leads by example and collaborates internationally. However, the Commission counters any overly rosy picture of China: China’s cooperation in the UN comes with a price — it aggressively pushes its values against like-minded countries. Meanwhile, the US is not regulation-free; state-level fragmentation (e.g., California) is a problem. The EU’s uniform rules are an asset.
UK Government’s Approach: Fastest Adoption, Sovereign Edge
The UK aims for the fastest AI adoption in the G7 via a productivity dividend. Its concept of sovereign edge, not wall means ensuring access to frontier tech, compute, data, and skills while avoiding dependency on a single supplier. Sovereignty is practical, not self-sufficiency. UK–EU collaboration is a strategic imperative — shared infrastructure, standards, and supply chains. The UK has joined EuroHPC and invested £7.8M. Three near-term priorities:
- Infrastructure: compute via AI Research Resource and AI Growth Zones.
- Regulatory innovation: sandboxes, AI Growth Lab.
- People and skills: foundational skills for 10 million workers, advanced fellowships.
The UK co-founded the AI Security Institute, established the binding Council of Europe AI Convention, and leads in AI assurance. Coalition-building is essential.
OECD’s Evidence and Partnerships
The OECD provides three dimensions of support:
- Shared principles: the OECD AI Principles are embedded in the EU AI Act and the Council of Europe Convention.
- Evidence base: the oecd.ai platform offers real-time data on compute, VC investment, and more.
- Partnerships: GPAI integrated into OECD, 46 members.
Key data points: US and China dominate compute access; emerging economies have only 23% of capacity, mostly not AI-enabled. 61% of global VC goes to AI, with 75% to US-based companies. The EU is a net beneficiary (receives more than it supplies). Investment is shifting from research to infrastructure and deployment. The G7 Code of Conduct has seen 25 organizations report on risk management — companies find value in transparency for internal governance and client trust. OECD’s global outreach includes co-creation workshops in Latin America, Southeast Asia, and Africa — not just exporting frameworks but building together.
Civil Society: The EU’s AI Act as Global Public Good
A US professor noted that the EU’s inferiority complex is misplaced: the AI Act provides a global public good and incentivizes US companies to improve. There is no evidence the AI Act slows innovation. International convergence is underway — AI laws exist in China, South Korea, Vietnam, several US states, Canada, Japan, Brazil, and more. The EU should build on this momentum. Without access to the EU market (and other large markets), AI corporations cannot be profitable. The Brussels effect can become Brussels + Brasília + Pretoria + Hanoi. Three recommendations:
- Strong enforcement of the AI Act (especially GPAI rules from August).
- Actively support partners passing similar legislation.
- Use emerging convergence to create a large regulated market.
Belgian Industry: Strategic Interdependency and Physical AI
China’s 15th Five-Year Plan explicitly aims for full technological self-reliance, including semiconductor full stack. Europe must act, not debate. The concept of strategic interdependency means not building a complete isolated EU stack, but deciding which nodes of the AI value chain to own (e.g., data, some LLMs for democracy), which to build with trusted partners (like-minded democracies), and which to safely source from allies (US, even China — with learning). Physical AI (world models for robots, industrial machines, ports, energy grids) is a still-open playground. Large language models are over-indexed. Europe has a strong robotics sector (e.g., 90% market share in service robotics like milking robots). The window of opportunity is now; in five years it closes.
Panel Discussion: Talent, Collaboration, Open Source, and Sustainability
Talent shift: More tech workers are leaving the US for the EU (political climate, layoffs). Europe must seize this opportunity with easier immigration and better incentives.
UK AI Security Institute: A successful model — pay flexibility, high-profile hires (e.g., Jade Leung from OpenAI), positioned as a place to move the dial on AI governance.
EU–UK collaboration: Should be deeper. The UK has technical expertise; the EU has regulatory implementation. The AI Safety Report (IPCC-style) should be better shared.
Open source/open weight models: Pros include sovereignty, privacy, adaptability. Cons include safety risks if capabilities are high (e.g., child safety, anthropomorphism). The EU already funds Open Euro LLM; the OECD is cataloging open models.
Sustainability: Smaller models (physical AI) are more energy-efficient than large LLMs. Transparency on water/energy usage is needed. The EU funds frugal AI research.
Communication strategy: Europe is bad at hype. It needs to better communicate AI as augmentation, not replacement. The UK’s 10 million worker upskilling and Future of Work Unit are good examples.
AGI / Manhattan Project question: The UK focuses on building sufficient compute, competing in areas of strength (chip design, cybersecurity), and equipping workers. The EU supports infrastructure but does not develop frontier AGI directly — emphasis is on adoption and trust.
Closing: Europe as Water, Not an Iceberg
Europe must act as water around the iceberg — adaptive, forward-looking, transformational — rather than protecting the past. Key tensions remain: security vs. openness, regulation vs. speed, owning nodes vs. cooperating. The panel’s consensus is that Europe’s influence comes from credibility, coalition-building, and practical deployment — not from scale alone.
The concepts covered here are best seen in practice. The video goes through each step in detail and is worth watching.
