In previous articles, I argued that most economic actors, from consumers to SMEs and large corporations, hardly care about digital sovereignty. They are buried in their own tech sediment, getting out of there is costly and offers little help with their immediate business problems. Overall, digital sovereignty is too abstract and is therefore largely delegated to governments and policymakers.

But if digital sovereignty is primarily a political challenge, it needs a political narrative. A narrative reduces complexity, makes the issue communicable, and can align different interests around a common understanding of the problem and what should be done about it.

What Makes a Political Narrative Work?

A political narrative is effective when it does more than tell a compelling story. It provides a simple framework for understanding what is happening, why it matters, and what should be done about it.

• Reduce complexity: Make complex politics understandable.
• Create meaning: Explain why things happen and matter.
• Shape perception: Define the problem and make certain solutions appear logical.
• Connect to values: Link issues to fairness, freedom, security, or solidarity.
• Create identity: Define who “we” are and what we stand for.
• Motivate action: Show that change is necessary and possible.

“Make America Great Again” is a textbook example for an effective narrative. It reduces complexity by turning diverse economic, social, and geopolitical problems into one story of American decline. It creates meaning by identifying causes and actors, from political elites to globalization and immigration. It connects this diagnosis to values such as strength, security, and sovereignty, creates a collective identity around the idea of forgotten or “real” Americans, and, crucially, offers agency: decline is not inevitable. America can become “great again.”

This article is an attempt to develop such a narrative for European digital sovereignty in the age of AI and geopolitical change. But I want to add one constraint: realism. A convincing narrative should be grounded in how technology, economics, and geopolitics are likely to develop. More importantly, as that future unfolds, reality itself should reinforce the narrative. New technological breakthroughs, investment decisions, market developments, and geopolitical events should become evidence that the underlying story is right.

The Tech Narratives Of Today

The dominant technological narrative today comes from the United States, and it could hardly be clearer: America must win the AI race. The premise is that leadership in AI will translate into economic prosperity, national security, and geopolitical power. Whoever leads in AI will shape global standards and reap the economic benefits.

At its core, the narrative rests on a remarkably simple assumption: scale wins. More compute produces more capable AI, more capable AI creates economic and geopolitical power, and therefore the country that can build the most compute wins the race. The logical consequence is brute force scaling: more chips, bigger data centers, more electricity, and more capital.

And America is putting enormous amounts of money behind this narrative. Stargate alone was announced with a commitment of $500 billion over four years for AI infrastructure, starting with $100 billion. By 2026, the scale had grown much larger: the world’s nine largest cloud providers were expected to spend around $830 billion in capital expenditure in 2026 alone, much of it driven by AI infrastructure. Meanwhile, the major AI hyperscalers had accumulated roughly $1.5 trillion in purchase commitments, on top of around $1.5 trillion in lease obligations. The message is hard to miss: if scale wins, America intends to outscale everyone.

The European narrative, on the contrary, is: “We will become an AI Continent.” Von der Leyen proves it with the announcement of €10 billion in investments over 6 years into AI infrastructure, accompanied by announcements of billion euro fines against Google and moves to make Microsoft Azure and AWS gatekeepers under the Digital Markets Act. So while the American narrative puts massive infrastructure buildout on the table as proof, the European narrative combines a much smaller infrastructure push with regulation as a demonstration of digital sovereignty.

Both narratives have their weaknesses

But let’s start with the European one. Von der Leyen, interestingly, lets Europe be judged by a metric largely defined by the competing American narrative: AI infrastructure buildout. If the 2023 to 2026 model of ever larger compute clusters is indeed the winning scenario, Europe is not becoming more sovereign by investing €10 billion. It is falling further behind. Even if the broader ambition to mobilise €200 billion succeeds, Europe would still be competing in a game whose rules are being set elsewhere.

Proving the effectiveness of the European narrative through regulation is equally problematic. First, because it is reactive, negative proof: sovereignty is demonstrated by constraining somebody else rather than by showing what Europe itself can achieve. Second, because transformative technologies are Pandora’s boxes. Once they are out, economically useful, and spreading through society, they become extraordinarily difficult to contain. Regulators can slow them down and shape their consequences, but history suggests that technological reality can eventually overwhelm the original regulatory ambition. Just remember the Index Librorum Prohibitorum. Starting in 1559, the Catholic Church spent more than 400 years trying to control which books Catholics could read. In 1966, it finally abandoned the Index as enforceable Church law and entrusted the matter to the “mature conscience of the faithful.”

But the American narrative is beginning to show cracks as well. Its scaling logic requires extraordinary amounts of capital to keep flowing into infrastructure. McKinsey estimates that meeting global AI compute demand could require $5.2 trillion of data centre investment by 2030. At the same time, the IEA already sees bottlenecks in electricity, grid connections, chip manufacturing and high bandwidth memory, and expects data centre electricity consumption to roughly double between 2025 and 2030.

The economic tension is even more interesting. The infrastructure required to produce AI is becoming extraordinarily capital intensive, while its fundamental output, tokens, is becoming radically cheaper. Stanford found that the inference cost for GPT 3.5 level performance fell more than 280 fold in roughly 18 months. Paul Kedrosky goes further and describes tokens as an inherently deflationary commodity, with quality adjusted prices falling by roughly 70 to 90 percent annually. This creates a peculiar economic tension: enormous, increasingly long lived capital commitments are being made to produce a commodity whose price is collapsing. Or, in Kedrosky’s terminology, the AI buildout risks creating a duration mismatch between long lived infrastructure and a deflating commodity.

The European Narrative Opportunity

While the American narrative had the upper hand over the last few years, evidence from the real world is beginning to open space for a convincing European alternative.

The positive evidence seems increasingly convincing: the revenue trajectory of frontier AI companies suggests that AI is not another crypto like fad, but the next major platform shift. Benedict Evans has described these shifts as changes in the underlying technological paradigm that occur every decade or two. They do not replace everything that came before, but they change what gets built next and how it gets built. In that sense, the American narrative probably gets one big thing right: demand for AI, and therefore for tokens, is going to be enormous.

But the experience of the last few years also suggests that AI has very different economics from the software paradigms that preceded it. Software scales extraordinarily well. You can build a product for a few million euros and distribute every copy at almost zero marginal cost. Combined with network effects and accumulated technological lock in, this created the extraordinary margins and market power of the American software industry that we know today.

AI increasingly looks different. Producing intelligence requires chips, data centres, electricity, cooling, networks, land, and capital. Marginal costs matter again. At the same time, AI is becoming an exchangeable commodity: models offer increasingly similar capabilities, switching is easy, prices are collapsing, competition is intense, and the powerful network effects that protected the software platforms of the last era are largely absent. In that sense, AI may bring us back towards the economics of earlier technological frontiers such as railways, electricity grids, telecommunications, and industrial infrastructure. These were technologies where competitive advantage depended not only on code and scalability, but on engineering, manufacturing, physical infrastructure, natural resources, and the ability to deploy enormous amounts of capital.

And AI is not the only technological frontier moving in this direction. Look at what else is becoming strategically important: satellites, robotics, quantum technologies, power grids, batteries, semiconductor fabs, defence systems, and the infrastructure required to connect all of them. Much of this requires material engineering and research, large upfront investments, physical supply chains, and actual factories. In short: a lot more atoms, relatively fewer bits, and a lot more capital going into both.

And then there is an interesting irony: While the physical side of technology is becoming more important, software itself is being commoditized as well. AI coding tools are rapidly reducing the cost of turning an idea into working software. Thanks to the enormous American AI infrastructure buildout and heavily subsidised access to frontier models, Europeans get access to that capability without having to finance the infrastructure behind it themselves. One could provocatively say: America is subsidising Europeans to prompt themselves out of their software lock in, while simultaneously flooding Siemens with orders for gas turbines to power the data centres doing the prompting.

This is where the European narrative opportunity begins. The AI platform shift may be enormous without reproducing the economics that made America dominant during the software era. Instead, AI and the technological waves around it could shift value back towards capital intensive infrastructure, energy, manufacturing, industrial engineering, and physical technology, areas in which Europe still possesses substantial capabilities. At the same time, AI is beginning to commoditise one of the great sources of American technological advantage: the ability to deploy sticky software globally and sell it at huge margins.

The European opportunity, then, is not necessarily to beat America at the AI game America is currently playing. It may be to recognise early that the game itself is changing.

Own The Physical Frontier

The key to our next technology narrative lies in a simple shift. The last technological era rewarded software: global scalability, near zero marginal costs, powerful network effects and enormous margins. America dominated that game. The next era is becoming physical again. AI requires massive infrastructure while commoditising intelligence and software. Robotics, energy, semiconductors, defence, space and quantum depend on engineering, factories, research, resources and capital. The technological frontier is moving towards European strengths.

The short version of that narrative should therefore circle around something like “Back to Atoms,” “The Physical Frontier,” or “The Age of Engineering.” My favourite is “The Physical Frontier”: it does not suggest going backwards. It says that the technological frontier itself is moving into the physical world.

The details of the narrative therefore could be:

• Reduce complexity: The technological frontier is moving from bits back towards the physical world.
• Create meaning: As software and intelligence become abundant, the scarce physical capabilities around them become more valuable.
• Shape perception: Europe should stop judging its technological strength by the platforms it failed to build and start looking at the capabilities the next technological era will depend on.
• Connect to values: Prosperity, resilience and sovereignty come from the ability to build what the world needs.
• Create identity: Europe is not yesterday’s industrial continent. It is the engineering continent for the next technological era.
• Motivate action: Double down on Europe’s physical strengths: energy, grids, robotics, semiconductors, defence, space, advanced manufacturing and research.

And Europe does not need to give up on AI. It only needs to give up the ambition of being permanently at the LLM frontier. Being a few months behind the most capable model matters far less if intelligence is rapidly commoditising. What Europe does need to own is the harness that controls how LLMs are integrated into applications, workflows, industries and infrastructures, and turn increasingly abundant intelligence into economic value.

Most importantly, reality can prove this narrative right. Every cheaper token commoditises intelligence. Every coding agent weakens software lock in. Every new data centre increases the value of energy and infrastructure. Every new robot increases the value of engineering and manufacturing.