Airfields to bring back abundance
Let us begin far from Brussels. Between 1942 and 1945, the Pacific War turned the islands of Melanesia into logistical bases for the American and Japanese armies. To the people who lived there, the spectacle was astonishing: runways cut out of the jungle, towers going up, planes landing laden with tinned food, tools and medicine. Abundance on a scale never seen before descended from the sky, and nobody could be seen making any of it.
Then the war ended, the bases closed and the cargo stopped. Some communities set out to bring it back by reproducing what they had observed: runways cleared in the bush, control towers built of wood and bamboo, headphones carved from coconuts, parades with wooden rifles. On the island of Tanna, in what is now Vanuatu, the John Frum movement, which predated the war and was transformed by it, has been waiting ever since for an American benefactor to return. Anthropologists gave these movements a name: cargo cults.
In 1974, the physicist Richard Feynman drew on the phenomenon to coin the term “cargo cult science”, an expression that has since entered everyday language.
“They’re doing everything right. The form is perfect. […] But it doesn’t work. No airplanes land.”
The analogy belongs to a physicist rather than to the anthropologists, who have shown how much these movements also owed to colonial rule. Its value here lies in the mechanism it highlights. The islanders had no way of seeing the industry that produced the cargo, so they reproduced the only part of it they could see. When the cause of prosperity is invisible, people imitate its forms. And then they wait.
Our bamboo runways
This mechanism describes European technology policy, starting with its announcements of technological ambition. Our runways are the “AI gigafactories” promised under the InvestAI plan, €20bn within an overall investment target of €200bn; the €109bn of private investment announced at the Paris summit in February 2025; and the “sovereign clouds” that run American hyperscalers’ technology under a local flag. These announcements will be worth only what they eventually deliver: capacity in use, paying customers, a viable business model. Until then, they are runways. Our control tower is the AI Act, the world’s first comprehensive regulation of artificial intelligence, adopted in 2024 to direct traffic the continent was not generating; the summits, Bletchley, Seoul and Paris, are its liturgy. Regulating is not the same as imitating. But most of our political energy goes into the outward forms of power, and very little into what produces it.
The numbers tell a different story. Not one European company ranks among the world’s ten most valuable technology firms. In 2025, the European Union attracted around 6% of global venture capital invested in AI, against 75% for the United States. Three American providers control 70% of the European cloud market. A few planes have landed. There is no scheduled service. The parts exist, but nobody has assembled them.
A cargo cult also has a tell-tale signature: failure never disproves the belief; it merely calls for the ritual to be refined. Every new assessment of Europe’s decline produces another summit, another report, another plan. As early as September 2024, the Draghi report put the additional investment Europe needs to stay competitive at close to €800bn a year across the full range of policy priorities; it is widely cited but implemented only to a limited extent. And in June 2026, weeks before the deadline, Parliament and Council postponed the application of the AI Act to stand-alone high-risk AI systems until December 2027, and until August 2028 for those embedded in products, in the name of simplification and competitiveness. The rules had got so far ahead of the businesses they were meant to govern that those businesses had to be given more time.
A standard is not an industrial policy
There is no law of history that says regulation must always come after industrial strength has been established, and Europe’s protections do not need to wait for champions to emerge before they can be justified. But a standard is not an industrial policy, and GSM is the proof. It made Nokia, Ericsson and Alcatel because the standard came coupled with operators placing orders and equipment makers filling them. GDPR achieved its purpose, protecting Europeans’ data, and spread abroad through access to the European market; it created no industry, because that was never its aim. A standard backed by production capacity and commercial demand creates market leaders. A standard on its own creates a carefully swept runway.
Regulation can also disadvantage domestic challengers. Compliance costs are largely fixed: marginal for a firm valued at $3 trillion, a substantial burden for a Series A start-up. Regulation designed to contain the giants becomes their moat, and a barrier to entry for the newcomers it claimed to nurture.
The invisible engine
What the islanders could not see, we can name: the chain that produces value, as distinct from its outward signs. It has four links, and we know every one of them.
Capital comes first. OpenAI’s March 2025 round, $40bn in a single raise, exceeds annual venture capital investment in France and Germany combined; Anthropic raised $13bn six months later. Yet Europe’s savings are among the largest in the world, and around €300bn of them leave each year, mostly for the United States. Financing foreign growth buys a claim on its future income, so the return is not the problem. Where the money ends up is. We finance abroad the companies, jobs and decision-making centres that we fail to build at home, for want of financial plumbing: cautious banks, undersized pension funds, twenty-seven legal systems and no exit market. Our unicorns list on Nasdaq, and it would be wrong to blame them: a company goes where the capital is. It is our job to make sure the capital is here too.
Energy comes next. Electricity shortages are not universal across Europe. France draws from its nuclear fleet low-carbon, dispatchable and abundant power that most large economies envy; the Nordic countries have a comparable advantage in hydropower. What Europe lacks is the ability to turn that advantage into productive capacity: grid connections that take years, permits that drag on, and a fragmented market in which average industrial prices remain two to three times American levels. A data centre is, before anything else, a twenty-year electricity contract.
Then comes computing power itself. Measured by annual capital expenditure, the four large American hyperscalers invested close to $400bn in 2025, mostly in data centres, and have announced more than $700bn for 2026; InvestAI remains an investment target, not money spent. In terms of installed computing capacity, a single xAI site in Memphis houses several hundred thousand GPUs, while Jupiter, Europe’s exascale supercomputer inaugurated at Jülich in September 2025, brings together around 24,000: a scientific achievement, but roughly an order of magnitude smaller by GPU count.
Finally, people. Leading American laboratories pay their researchers several times what their European counterparts earn; employee share ownership is still taxed less favourably in parts of Europe than in the United States; entrepreneurial failure still carries a stigma here; and the fragmentation of the single market holds back the growth of software companies. DeepMind was founded in London and sold to Google as early as 2014; Hugging Face has French founders and its headquarters in New York. Much of the research comes from Europe; much of the commercial value is captured in America.
Europe’s own successes confirm the diagnosis. ASML holds a global monopoly on EUV lithography, proof that Europe can lock down a critical chokepoint in the global supply chain; but that monopoly is the product of three decades of patient capital in hardware, a sector whose long cycles suit the way our institutions allocate capital. Mistral AI, Europe’s best hope in models, raised €1.7bn in September 2025, of which €1.3bn came from ASML under a strategic partnership. That alliance between two European industrial companies is exactly the ecosystem Europe says it wants; it also tells us that the €1.3bn cheque was written by an industrial company, not by a fund. Europe already has electricity, companies, engineers and capital. What it lacks is a system that connects paying customers with productive capacity and retains the resulting value in Europe. The problem is coordination, not a complete absence of resources. That makes it more tractable.
What waiting costs us
Waiting is not free. AI profits are concentrated among a handful of companies; in October 2025, Nvidia became the first company to pass $5 trillion in market value. European investors may own shares in these companies. But the larger industrial question is where the businesses, jobs, intellectual property and strategic decisions are located. Today, that place is not Europe.
The regulatory lever is wearing out too. The Brussels effect depends on a market that global firms cannot afford to bypass, and as Europe’s relative weight declines, the cost of staying out falls; it is already visible in the features delayed or withdrawn from the European market. Regulation without an industry behind it eventually regulates no one.
What if AI is a bubble?
Some will object that the cargo may be a bubble, and that the wise course is to let the Americans burn their capital. The objection deserves better than a shrug. Today’s valuations will no doubt correct at some point, and no one should finance any asset at any price. But it confuses two questions. The first is price, and that is a matter of financial discipline. The second is capacity, and that cannot be fixed after the fact. When a technology bubble deflates, it leaves behind infrastructure of very uneven residual value: the fibre overbuilt in 1999 carried the video and cloud services of the following decade; a badly negotiated power contract becomes a liability; a team disperses. Those who bought the assets of 2001 for a fraction of their cost had the balance sheets, the customers and the engineers to run them.
If the AI cycle turns, what matters is whether Europe can acquire productive assets, operate them and retain control over them. That requires sites with grid connections, local customers, skilled teams and investment vehicles. Europe has all these elements, but they remain fragmented; they do not yet form a whole.
A formula Europe already knows
There is, however, one decisive difference from Tanna, and a reason not to despair: we know what the invisible chain looks like, because we have assembled it before. Airbus, Ariane and GSM share several foundations: demand secured in advance, through orders from governments, airlines or operators; patient capital; energy and factories; engineers; and rules written as the capability took shape. The formula has not aged. Europe has simply stopped applying it.
Applying it to AI does not mean repealing our laws. Protecting citizens, from their data to product safety, does not depend on the existence of champions. Claiming to set the world’s standards does: that ambition has to be earned through industrial capacity, and the compliance burden must be proportionate to the maturity of the companies that carry it.
The next step is to channel European savings into risk-bearing investment, with measurable results: prudential treatment that favours private equity, pan-European pension funds and a single European company statute, so that a fraction of the €300bn that leaves each year stays.
Computing must also be treated as energy policy: make full use of nuclear and hydropower at last, sharply reduce permitting and grid-connection times, secure long-term electricity contracts, and site capacity where low-carbon electricity is available at competitive prices.
And public-sector buyers must become early customers, as the Pentagon was for semiconductors and GPS while DARPA funded the research that led to the Internet. In defence, health and public administration, Europe commands enormous demand that it does not direct towards European suppliers.
Finally, the measure of success has to change, and this is where the cargo cult returns one last time. Counting gigawatts connected, GPUs deployed and funds raised, rather than summits held, would be progress, but it would still measure inputs rather than outcomes: an under-used cluster, a funding round at an inflated valuation and a premature listing are runways of another kind. Feynman defined scientific integrity as the refusal to fool oneself about what works. Landings are results: customers who pay and come back, capacity used at a viable rate, services delivered at a competitive cost, and value, in intellectual property, skills, profits and decision-making power, that stays in Europe once the cycle has passed.
Count the planes that actually land.
On Tanna, the cult has never ended: every year, the islanders still watch for the cargo to return. At least the followers of John Frum never claimed to set the rules of global aviation. Europe does. It is up to us to decide what we want to be in this story: the keepers of an immaculate runway on which almost nothing lands, or a continent that applies the industrial capabilities it already possesses to the next technological challenge.