How can the EU benefit from the US-China AI race?
Europe cannot win the AI spending race, but it can make its approval the price of entry to safety-conscious markets worldwide.
Prebisch and Lee converge on the same mechanism: whoever sets the standard collects the rent. The EU AI Act, enforced with market muscle, turns European approval into a prerequisite for global sales. Schmidt's 1973 energy crisis argument sharpens this: regulatory leverage only holds if the underlying market cannot be bypassed. The EU's 450 million consumers make that bypass costly enough to matter.
Hayek and Deng both warn that the Act risks codifying yesterday's risks while US and Chinese labs build tomorrow's systems. Deng's answer from Shenzhen in 1979 is the most practical: carve genuine sandboxes, let firms deploy under monitored conditions, and rewrite the rules as data arrives. A standard built on live deployment evidence is enforceable; one built on anticipated harms is not.
Confidence summary: Strong convergence on the certification-as-leverage mechanism; genuine split on whether the EU's regulatory architecture is agile enough to sustain it.
1. The core argument
The least obvious thing about the US-China AI race is that the EU's best move is not to enter it. The race rewards capital concentration and risk tolerance at a scale Europe cannot match. What Europe can do is position itself as the jurisdiction whose approval unlocks every market that does not want American recklessness or Chinese opacity. That is not a consolation prize. Standard-setters collect rents. The question is whether the EU AI Act, now two years in force, is the instrument to do that, or whether it will freeze the rules at the moment of its drafting while the technology outpaces every classification it contains. The answer depends on one thing the council could not fully resolve: whether Brussels is capable of writing a regulatory framework that updates itself faster than AI capabilities change. If it can, Europe becomes indispensable. If it cannot, it will have regulated what others invented and taxed what it did not build.
2. How each member frames it
Raúl Prebisch reaches for the commodity analogy not as rhetoric but as a structural diagnosis. In the 1950s Latin American exporters transferred value upward because they accepted the price and standards set by the centre. What the cards left out is Prebisch's candid acknowledgment of the trap's mirror image: regulatory standards without productive capacity behind them collapse into formalism. His challenge to Schmidt is genuine, not rhetorical. Europe must use the AI Act to build firms capable of competing at the frontier, not merely to certify systems built elsewhere. Without that industrial depth, the standard becomes a toll booth on a road Europe did not pave.
Helmut Schmidt reframes the question as one of structural dependency, which is sharper than it sounds. His 1973 argument about energy was not that Germany should build more pipelines; it was that dependence is a political condition, not a technical one. Applied here: what matters is not whether Europe leads in AI research, but whether the AI Act creates dependencies that competitors cannot escape cheaply. The European Monetary System analogy is instructive. It worked not because Germany's economy was the largest, but because exit costs were prohibitive. Schmidt would accept regulatory imperfection in exchange for irreversibility. He would reject any sandbox design that allows firms to operate indefinitely outside the Act's scope.
Friedrich Hayek is the council's structural dissenter, and his challenge is the most uncomfortable one to dismiss. His Nobel lecture argument was not about AI; it was about the epistemological hubris of any authority that claims to know enough to specify outcomes. The AI Act's risk classifications, conformity assessments, and prohibited-use categories were written for a technology that has since changed substantially. Hayek does not oppose safety rules in principle; he opposes locking them in before the knowledge base exists to write them reliably. His candid limit is that he offers no institutional alternative. His challenge to Deng is the brief's most productive tension.
Deng Xiaoping takes Hayek's epistemological point and converts it into a design principle. The Shenzhen insight was not ideological; it was practical. Nobody in 1979 knew what a functioning market economy looked like inside a communist state, so Deng built a space where the answer could be discovered. His application here is direct: regulatory sandboxes with genuine deployment rights, mandatory data collection, and scheduled rule revision would allow the AI Act to evolve from evidence rather than from anticipation. What his card omitted is the institutional condition: Shenzhen worked because Deng had the authority to override objections from the centre. European sandboxes will require member states to cede regulatory discretion, which is a political problem, not a technical one.
Lee Kuan Yew synthesises the group without papering over its tensions. Singapore's survival was not built on size or spending; it was built on indispensability. The certification role he describes for Europe provides market leverage without requiring frontier AI spending. His sharpest point, left off the card, is that certification credibility is fragile. Singapore maintained its position by being scrupulously reliable. If the EU certifies systems that subsequently cause serious harm, the certification brand collapses. That means Deng's sandbox-and-update mechanism is not optional; it is the condition under which Lee's strategy remains viable.
3. Where the council agrees
The most surprising point of agreement is that regulatory divergence from the US is an asset, not a liability. After Washington rolled back Biden-era AI safety requirements in 2025, the EU became the primary rule-setter for safety-conscious markets by default. All five members, including Hayek, accept that this creates a real opportunity. The disagreement is about how to exploit it, not whether it exists. Beyond that, the council agrees that the EU's leverage rests on market size, not AI capability. The 450 million consumer market makes bypass expensive enough to matter. It agrees, further, that a standard which cannot update will decay, and that the CHIPS Act supply-chain gaps represent a secondary lever Europe should use to deepen its semiconductor position, since certification without fabrication capacity is ultimately dependent on the players it seeks to regulate.
4. Where the council splits
The actual line runs between Schmidt and Hayek on one side and Deng on the other, with Prebisch watching both. Schmidt accepts regulatory imperfection as the price of irreversibility: get the dependencies locked in, even if the rules are imprecise. Hayek and Deng both reject that, but for different reasons. Hayek doubts the rules can be made precise enough to be useful; Deng thinks they can, but only after deployment evidence exists. Schmidt would say that waiting for evidence concedes the field to the US and China in the interim. Deng would say that locking in bad rules is worse than operating with provisional ones. Neither is wrong. This is a genuine strategic trade-off between commitment and adaptability, not a technical question with a correct answer.
5. For a policymaker to decide on
The concrete choice is this: enforce the AI Act as written, accepting that its risk classifications will lag the technology in exchange for the credibility of a stable, non-negotiable standard; or amend it now to create genuine regulatory sandboxes with mandatory revision cycles, accepting some short-term uncertainty in exchange for rules that stay grounded in real deployment data. The first option builds the brand faster. The second keeps the rules accurate longer. A policymaker must decide which failure mode, irrelevance or inaccuracy, costs Europe more.