Europe, AI, and Waiting: Can the EU Win by Holding Back?

Czech economist Tomáš Sedláček has floated a deliberately contrarian idea: in the race for artificial intelligence, the winner may be whoever waits. On this reading, Europe's instincts — GDPR, caution, institutional slowness — are not handicaps but latent advantages that pay off once the technology matures. It is a seductive argument, and an easy one to wave away when it leans on stretched analogies. So instead of arguing about it rhetorically, we put it through a scenario analysis: more than 50 variants modeled, fewer than 30 of which earned a meaningful probability weight. The result is more interesting than either the optimists or the cynics expect. Europe really could gain by waiting — without ever building a frontier model — but only if several conditions hold at the same time. A simplified map of the paths is published at vibecoding.cz/euro-ai; what follows is the summary.

The thesis, stress-tested

"Win by waiting" sounds like a license to do nothing. The analysis says the opposite. Passive free-riding — following the market, buying whatever the leaders ship — lands Europe in the category it most fears: the AI colony, the point at which it starts to suffer economically because it never built competitive AI capability. Waiting only works as a strategy, not as an excuse. To avoid colonization, Europe has to succeed on three fronts simultaneously, plus a fourth that holds the other three together over time.

Three fronts at once

1. A real AI substrate of its own. This does not require a homegrown frontier model. It requires data centers for inference — the everyday running of AI — that can operate frontier models of any origin, and over which Europe has decisive control. That control points toward open-source / open-weight models rather than dependence on a foreign API.

2. Adoption where Europe is already strong. AI has to be absorbed into the sectors where EU firms remain competitive — pharmaceuticals, machinery, energy, automotive, defense — so the continent's existing industrial advantages compound rather than erode.

3. The ability to enforce its own rules and build trust in AI. This is the front Europe is closest to today, through the EU AI Act and GDPR — provided that trust can be turned into a certifiable, auditable, exportable property rather than just internal cost.

The fourth, binding condition — a talent absorber. For the first three to last, Europe must capture educated people and keep engineering capacity permanently on hand: attract foreign experts, or train and retain its own. Without it, the benefit of the other three drains away with each generational turnover. Together these form an absorption mechanism that accumulates durable, domestic AI assets instead of exporting the value abroad.

The EU AI Act and Global AI Regulation: What Tech Companies Should Watch in 2026

The playbook: what "active waiting" actually demands

Waiting, done deliberately, is a portfolio of moves — not a pause:

• Formally quit the frontier-model race and reallocate the money. This does not mean blocking private European labs; it means abandoning the dream of a state-funded frontier model.
• Make AI adoption a duty, always paired with building the "substrate" — the compute and data layers that serve AI: inference capacity, a skills/data layer, operations and integration. In practice: vouchers for small and mid-sized firms, AI in public administration, conditionality in public tenders — steady pressure to adopt, while building the layer where the margin actually accrues.
• Build the substrate on inference and data, not on training. Training is the most expensive, least defensible part; running and fine-tuning models is where Europe can keep capacity at home and minimize the outflow of data and margin.
• Fund verticals with domain data. Public money flows into applications built on top of commodity models in fields where Europe holds the data and the market: machinery, pharma, automotive, defense, energy.
• Run the talent absorber. Return grants, university spin-offs, a domestic-operation condition on subsidized projects, a European-team share in every large deployment — so each project leaves behind people who can build the next one themselves. (This is how China absorbed EU–US industry.)
• Shape the agent and protocol layer now. Standards for agents, identity, authorization and payments are being written today; here, waiting is not an option.
• Export trust — conditionally. Turn the AI Act and GDPR into a certifiable feature, while continuously measuring whether outside players actually adopt European standards or whether the rules merely raise internal costs.
• Pre-define both a buy trigger and reversal triggers. A measurable condition under which Europe buys capability at scale (inference price below a set threshold plus demonstrated commoditization of the model layer), alongside a set of WATCH triggers that cancel the waiting logic outright.

Six colonization channels to monitor

Waiting is only safe if the downside is instrumented. These are the channels that signal the situation is deteriorating; tripping any one should force a strategy review.

WATCH triggers that void the whole logic

These were judged marginal, but if any occurs it would demand an immediate pivot, because each breaks the core assumption that the model layer is commoditizing on a predictable timeline.

Bottom line

Waiting is a strategy with a built-in expiration date: it holds only while the model layer commoditizes and no WATCH trigger fires. Read honestly, the data says Europe's wait turns into an advantage only as a conscious, expensive, four-role portfolio that adds absorption to adoption and polices the six drainage channels. Even then it most likely wins on sovereignty, wins conditionally on economics, and wins only uncertainly on long-run competitiveness. The passive version — free-riding, following the market — ends, on this analysis, in the AI-colony scenario. The probabilistic and time dimensions here are deliberately simplified into three levels; the underlying paths are open for readers to judge for themselves.

Explore the Euro-AI scenarios

Sources and further reading