There is a particular kind of power in being the country — or the continent — that everyone else has to obey. Europe has it. The General Data Protection Regulation reshaped how the entire internet handles personal data; the EU AI Act is now the most ambitious attempt anywhere to govern how machine intelligence is built and deployed. When Brussels writes a rule, Silicon Valley reads it. That is real influence, and it should not be dismissed.
But influence over the rules is not the same as influence over the thing being ruled. And on that second measure — who actually owns the chips, the data centres, the energy and the capital that turn AI from a paper into a product — Europe is not a leader. It is, on most counts, a long way behind. A provocative new report imagines where that gap leads if nothing changes. Critics call it alarmist. The numbers underneath it are not.
The gap, quantified
Start with the rawest figure. Analyses cited by Cybernews put Europe’s share of global AI compute at roughly 5 percent, against about 80 percent concentrated in the United States. That is not a rounding error or a temporary dip; it is a structural imbalance in where the world’s most strategically important machines physically sit.
The capital flows explain why the gap is widening rather than closing. Per the same Cybernews analysis, US tech firms spent over $400 billion in 2025, much of it on AI infrastructure, and American hyperscalers poured more than $200 billion into AI data centres in the first quarter of 2026 alone. Read that again: a single quarter of American data-centre spending dwarfs anything Europe has announced as a multi-year ambition. When one side is investing at that velocity and the other is debating frameworks, the lead compounds.
This is where the real critique bites. Europe has not been slow on AI in every sense — it has been fast on regulation and slow on capacity. The AI Act arrived before the gigawatts did. Rules are being written for an infrastructure base that Europe largely does not own and is not building quickly enough to own. That sequencing — govern first, build later, if at all — is the heart of the problem. You can set the speed limit for a road, but if the cars are all manufactured and parked somewhere else, your authority is mostly theoretical.
The ‘irrelevance’ warning
The report at the centre of this debate, surfaced by Cybernews, sketches a near-future scenario in which Europe slides toward economic and political marginalisation — ‘irrelevance’ is the provocative word — if it continues to under-invest in AI infrastructure while over-regulating its use. The picture it paints is one of dependence: European businesses, governments and citizens running their most important workloads on American (and increasingly other) compute, paying rent to platforms they neither control nor can easily leave.
Critics have, fairly, pushed back. The scenario is a worst case, and worst cases are designed to alarm. Europe is not a technological backwater — it has world-class research, formidable engineering talent, and in chip-equipment maker ASML, a genuine chokepoint that the entire global AI supply chain depends on. To call the continent ‘irrelevant’ in any literal sense is overstated.
But there is a kernel of truth that the pushback should not be allowed to bury. The value of the report is not its precise forecast; it is that it crystallises the build-versus-regulate debate into a single uncomfortable question. If a region can write the rules but cannot build the systems, who actually holds the leverage when the two interests collide? The alarmism is a delivery mechanism. The underlying anxiety — that Europe is optimising for the part of the value chain that matters least — is legitimate, and the people calling it merely alarmist rarely have a convincing answer for the 5-percent figure.
Why building beats only regulating
The case for building is not anti-regulation. Good rules matter, and Europe’s instinct that AI needs guardrails has been vindicated repeatedly. The argument is about sequence and balance. Rules without capacity produce dependence, not sovereignty. A continent that regulates an industry it does not host ends up exporting its standards while importing its technology — and the importer is always the weaker party in that exchange.
So what does building actually require? Three levers, none of them legislative:
- Talent. The researchers and engineers exist in Europe, but too many of them leave for ecosystems where compute, capital and ambition are abundant. Retaining and concentrating talent is a precondition for everything else.
- Energy. AI data centres are, fundamentally, energy infrastructure. The race for compute is increasingly a race for power — cheap, reliable, ideally clean. Europe’s energy costs and grid constraints are a competitiveness issue dressed up as an environmental one, and they cannot be regulated away.
- Capital. The hyperscaler capex numbers above are the clearest signal of all. Europe lacks the deep pools of patient, risk-tolerant capital that fund infrastructure at American scale. Fragmented capital markets across member states make this worse, not better.
The prize for getting these right is the ability to export technology, not just standards. Standards are a defensive asset — they shape behaviour on your terms. Technology is an offensive one — it generates the wealth, the jobs and the strategic weight that let you keep writing standards in the first place. Europe has spent a decade perfecting the defensive game. The report is a reminder that defence alone does not win.
The India read
None of this is only Europe’s problem, and that is precisely why it matters from Delhi or Bengaluru. India is, on the compute question, another have-not — a vast market and a deep talent pool racing to build domestic capacity it does not yet possess at scale. The IndiaAI Mission, the push for sovereign compute and subsidised GPU access, and a wave of domestic data-centre investment are all bets on the same proposition Europe is being warned about: that hosting the machines, not just using them, is what confers real agency.
The choice in front of India is the cleaner version of Europe’s dilemma: rule-taker or builder. A rule-taker consumes models, chips and cloud capacity designed elsewhere, governed by frameworks shaped elsewhere, and absorbs whatever terms the builders set. A builder may start smaller and slower, but it accumulates leverage — over pricing, over access, over the direction of the technology itself. India has the demographic and demand-side advantages to make the builder path viable in a way that few other regions can claim.
The honest caveat is that ambition has to be realistic. India will not out-capex American hyperscalers, and pretending otherwise produces vanity projects rather than capacity. The realistic ambition lies in the layers where India’s advantages are real: applied AI built on top of strong domestic compute, public digital infrastructure that the world already studies, multilingual and cost-efficient models tuned for emerging markets, and talent that does not have to emigrate to build at the frontier. That is not the same as winning the raw-compute race. It is something more durable — building enough that you are never purely at someone else’s mercy.
Europe’s predicament is a warning precisely because Europe is not weak. If a continent with ASML, elite universities and enormous wealth can find itself hosting 5 percent of the world’s AI compute, no one should assume that influence over the rules will protect them from irrelevance in the build. The countries that take that lesson seriously will spend the next decade pouring concrete and securing power, not just drafting clauses. The ones that do not will keep writing rules for a future they no longer own.
