Generated by Codex with GPT-5

This summary covers The Economist’s June 20th, 2026 By Invitation article by Judith Dada, listed in the contents as Judith Dada on the stakes for Europe in AI and published under the headline How Europe must respond to America's AI warning shot.

The essay argues that Donald Trump’s order forcing Anthropic to block non-Americans from its newest frontier models should jolt Europe out of complacency. For the first time, a government had compelled the withdrawal of publicly released advanced AI from foreign users. Whether the ban lasts is less important than what it revealed: Europe depends on American firms, American infrastructure and American political tolerance for access to the most powerful AI systems.

Dada’s answer is not digital isolation. She rejects both resignation and protectionism. Europe should try to build its own frontier labs, but it should also admit that catching up may take years and may fail. A serious strategy therefore needs two tracks at once: make Europe a more credible home for AI development, and build enough bargaining power to secure access to American frontier systems on better terms.

Bargaining From Strength

The first proposal is diplomatic. Europe should gather what Dada calls AI middle powers into a coalition that has more leverage together than any member has alone. The relevant assets are scattered: Dutch lithography through ASML, Korean memory chips, Japanese semiconductor equipment, Canadian and Nordic energy, German industrial data, British AI safety expertise and European scientific capacity.

None of those strengths is enough by itself. A single chokepoint can be bypassed, weakened by technical change or overwhelmed by American scale. But bundled together, they become a negotiating position. The point is not to recreate the entire AI supply chain inside Europe. It is to make access to frontier models something Europe can bargain for rather than something it receives by permission.

Compute As Strategic Infrastructure

The second proposal is physical. Europe has only a small share of global AI compute, while America has the overwhelming majority. Dada argues that Europe should aim for a much larger share within five years, which would require tens of gigawatts of new data-centre capacity, faster permitting and far more electricity generation.

That is politically awkward because data centres consume power, create few direct jobs and make AI’s disruption more visible. But the essay treats compute as strategic infrastructure, closer to ports, grids and defence capacity than to an ordinary corporate investment. Without it, Europe cannot train, host, test or deploy advanced models at meaningful scale. It will remain a customer, not a shaper, of the technology.

Dada also cautions against confusing sovereignty with self-sufficiency. Europe cannot build all the needed facilities alone, especially when capital and advanced chips flow first to American buyers. The practical move is to make Europe the best place for American AI firms to build, while requiring durable access agreements for science, cyber-defence and military uses.

Labour Markets And A Story Voters Can Believe

The third proposal is institutional. Dada argues that Europe’s labour rules can slow adoption of new technology and weaken the market power that gives the continent influence. Her preferred model is not American-style insecurity, but Danish-style flexicurity: companies can reorganize and adopt new tools, while workers receive retraining and generous support when jobs change.

This is where the essay moves beyond industrial policy. Europe cannot ask voters to accept more data centres, faster corporate disruption and looser labour rules with only a fear-based argument about falling behind. It needs a positive account of what AI can do for European lives: better work, stronger science, more capable defence, protected welfare states and safer technology shaped by democratic societies.

The closing point is that sovereignty matters because safety and agency are linked. A Europe that depends entirely on decisions made in Washington or San Francisco will have less say over how powerful AI is built and used. The Anthropic ban may be amended or withdrawn, but the dependency it exposed will remain. Europe still has assets, markets and allies. The question is whether it can organize them before AI power hardens into another field where it negotiates from weakness.