How can the EU reduce its dependency on US and Chinese AI?
Europe must build AI capability through strategic state investment while liberalizing regulations to attract private talent and capital.
Schmidt's energy lesson applies directly: single-source dependency becomes political vulnerability within years. Colbert's industrial policy worked in glass and shipbuilding; AI requires the same state-backed learning phase against established competitors. Friedman correctly identifies regulatory barriers as innovation killers; GDPR has already pushed AI development elsewhere.
The council splits on sequencing: state investment first or regulatory reform first. Both are necessary but insufficient alone.
Confidence summary: Strong convergence on sovereignty risks, sharp disagreement on whether state investment or market liberalization better builds technological capability.
1. The core argument
When OPEC cut oil supplies in 1973, European governments discovered that democratic institutions mean little when foreign powers control essential resources. Today's AI dependency carries the same sovereignty risk. With Nvidia commanding 80% of AI chip markets and no EU company producing competitive frontier models, Europe faces a technological vulnerability that regulatory frameworks cannot address. The council agrees that sovereignty requires capability, not just governance. Where they split cuts to the heart of state power: whether governments can successfully direct technological development or whether only markets can build the complex ecosystems that produce AI leadership. This is not a question of preference but of institutional capacity to manage technological transformation under competitive pressure.
2. How each member frames it
Helmut Schmidt sees AI through the lens of energy security, where the lesson was harsh and immediate. In 1974, watching German factories shut down because of distant political decisions, he learned that strategic autonomy requires redundant supply chains and domestic capability. Schmidt argues that Europe's current AI dependency mirrors pre-1973 energy complacency: assuming that commercial relationships would remain stable regardless of geopolitical shifts. His solution involves what he calls "diversified dependency," building European AI infrastructure while maintaining partnerships with multiple technological powers. Schmidt candidly admits this requires accepting short-term inefficiency for long-term security, a trade-off he made with nuclear power despite economic costs.
Jean-Baptiste Colbert views this through successful precedent: building French manufacturing dominance against established competitors. He sees Schmidt's energy analogy but argues it understates state capacity to create technological capability from nothing. Colbert's 17th-century glass industry went from nonexistent to European leader within a decade through systematic talent acquisition, protected learning phases, and strategic subsidies. He contends that AI represents exactly the kind of strategic industry where markets alone cannot generate the patient capital and coordinated investment required. Colbert openly acknowledges that this approach risks creating "protected inefficiencies" but argues that temporary protection during capability building differs fundamentally from permanent protection of established industries.
Milton Friedman reframes the entire question around regulatory barriers rather than industrial policy. He argues that Europe's AI deficit results from government intervention, not insufficient intervention. The GDPR alone has driven AI development to jurisdictions with fewer restrictions on data use and algorithmic development. Friedman sees Colbert's state investment approach as guaranteed to produce expensive failures, pointing to decades of failed government technology projects. His solution requires dismantling barriers to talent mobility, capital formation, and data utilization that currently make Europe an unattractive location for AI development. Friedman concedes that this approach cannot guarantee European ownership of AI leaders but argues that location matters more than ownership for economic benefits.
3. Where the council agrees
The most surprising convergence lies in recognizing that AI dependency creates political vulnerability within electoral cycles, not decades. All three members reject the assumption that commercial AI relationships will remain stable under geopolitical pressure. They agree that current EU AI policy focuses excessively on governance and regulation while ignoring capability building. Schmidt, Colbert, and Friedman converge on viewing talent mobility as crucial: whether through state recruitment or regulatory attraction, Europe must become a destination for AI researchers and engineers. They also share skepticism toward purely defensive measures like export controls or technology screening, arguing these address symptoms rather than underlying capability gaps. Finally, all three acknowledge that Europe's manufacturing weakness in semiconductors represents a deeper constraint than governance frameworks can address.
4. Where the council splits
The fundamental disagreement centers on whether state industrial policy can successfully build technological capability in fast-moving sectors. Schmidt and Colbert argue that markets alone cannot generate the patient capital and coordinated investment required for strategic technology development, while Friedman contends that government direction inevitably produces inefficient allocation and delays adaptation to technological change. This splits along a deeper line about institutional capacity: Colbert's confidence in state capability to direct technological development conflicts sharply with Friedman's conviction that only market mechanisms can coordinate the complex information flows required for innovation. Schmidt occupies middle ground, advocating state investment in infrastructure while allowing market competition for applications, but even this moderate position cannot bridge the core disagreement about whether governments can successfully pick technological winners.
5. For a policymaker to decide on
Whether to launch a major EU AI development fund with dedicated chip fabrication facilities and research institutes, or to eliminate regulatory barriers and offer tax incentives to attract existing AI companies to European operations. The first path requires committing billions in public investment with uncertain returns but builds indigenous capability. The second path attracts private capital and existing expertise but provides no guarantee that European-located AI development serves European strategic interests. The choice depends on whether policymakers believe state direction can build technological capability faster than market mechanisms, a judgment that cannot be resolved through analysis alone.