What our digital budgets are actually building

We tend to treat digital sovereignty as a matter for governments. The debate revolves around European regulation, public investment, critical infrastructure and geopolitical tension, with companies cast as bystanders who will eventually adapt to whatever framework emerges above them. That reading is misleading.

Sovereignty is also decided in investment committees, at the moment an organisation selects a cloud provider, a collaboration platform, a security solution or an AI model.

Taken one at a time, these decisions serve entirely legitimate goals: better performance, faster delivery, lower cost, quicker access to innovation. Aggregated across a continent, they direct capital flows, accelerate the maturity of certain ecosystems and ultimately determine which industrial capabilities will exist a decade from now. Every technology budget is therefore already a strategic decision. The only question worth asking is which strategy it happens to be funding.

A dependency no single decision created

The Anthropic episode last June illustrates the point precisely. On 12 June, non-US users lost access to the Claude Fable 5 and Mythos 5 models, until the Department of Commerce lifted the measure on 30 June and service resumed the following day. This was not an outage, a cyberattack or a breach of contract. The company was implementing a decision taken by US authorities under export control rules.

What makes the episode interesting is not its duration. Eighteen days is negligible at the scale of an organisation. What deserves attention is what did not fail. No contract was broken, no availability commitment was called into question, and yet service continuity depended on a decision over which neither the supplier nor the customer had any influence.

This reveals no particular weakness at Anthropic. It exposes a structural feature of today's digital economy, which Cigref has described as structural fragility: certain dependencies now sit outside the contractual perimeter altogether and follow from political or industrial choices that user organisations do not control. The phenomenon is not one-sided. In 2025, a trade dispute with the Netherlands led Beijing to suspend semiconductor exports to Europe. Two powers, one instrument, and no negotiation over its use.

A budget is already an industrial policy

We generally treat IT budgets as cost centres or levers of operational performance, when in fact every cloud contract, every collaboration platform and every AI deployment helps fund an industrial ecosystem, its investment capacity and its competitive position. Individually these contracts look inconsequential. Collectively they reach a scale that is anything.

A 2025 study, commissioned by Cigref, estimated that roughly 265 billion euros a year of European enterprise cloud spending flows to American companies, more than eight euros in every ten.

Public institutions have started applying exactly this logic. On 30 July, the European Commission opened a call for proposals to build up to seven AI gigafactories, with applications closing on 12 November 2026 and awards expected in early 2027. Beyond the headline funding, one element deserves attention: eighteen member states have committed to purchasing part of the resulting capacity jointly rather than negotiating their own volumes separately, and France immediately pledged one hundred million euros of orders with the future French recipient. That commitment is arguably more consequential than the subsidies themselves, because it restates something easily forgotten.

An industry does not grow on public investment alone. It grows because it finds customers.

What the private sector has not yet undertaken

Nothing comparable exists on the corporate side. There is no collective effort among large European companies to bring into being a cloud and software ecosystem over which they would have some purchase, even though they are its principal funders and would be its principal beneficiaries.

The standard objection is that European offerings do not all match the maturity of their American competitors. It is a fair point, but it rests on circular reasoning. An offering does not become mature because everyone waits for it to be. It improves through exposure to demanding customers, complex environments and large-scale deployments.

For several years I had the opportunity to work on the development and structuring of technology offerings. An offering never matures on its own. It evolves through contact with its first major accounts, through their requirements, their feedback and their willingness to invest in a long-term relationship. That dynamic is what gradually turns a promising solution into a platform capable of meeting the expectations of large organisations.

This clearly does not mean handing the most sensitive systems to a provider that has yet to prove itself. The workable approach is a graduated one. It begins with mapping projects by actual criticality, an exercise few organisations conduct through that particular lens. It continues by entrusting a European provider with the workloads whose interruption would be tolerable, the ones that allow learning without exposing the business. Where those commitments are multi-year and genuinely demanding, they give the provider what it most lacks: recurring revenue, complex use cases and a reason to invest. The perimeter then widens at the pace of demonstrated maturity, against criteria set in advance rather than improvised under pressure.

This gradualism is not excessive caution but a technical necessity. Migration experience consistently points the same way: moving an architecture built on one hyperscaler's managed services to a European provider is rarely a straightforward infrastructure switch. Starting with the least critical perimeter is also the only way to acquire that migration capability before it is needed on systems that tolerate no error.

The instruments already exist

What makes the inaction more striking is that the necessary tools have already been built, funded and tested, without companies making use of them.

The European Commission has published a Cloud Sovereignty Framework that translates sovereignty into measurable contractual criteria across eight domains, spanning legal and operational dimensions as well as supply chain transparency and technological openness, graded across five levels. The exercise is not theoretical: the framework was used on 17 April 2026 to award a cloud services contract worth up to 180 million euros over six years to four European groupings, including Post Telecom together with OVHcloud and CleverCloud, Germany's StackIT and France's Scaleway. The Commission has stated its intention to make the framework available to all organisations.

Any company therefore has access to a free assessment method, proven on a real procurement, to establish what a supplier's sovereignty commitments actually amount to rather than relying on marketing claims. Alongside it, mapping work on European technology capabilities is under way, led by industrial and academic coalitions that catalogue, layer by layer, what Europe genuinely has and what it lacks.

The intellectual infrastructure is ready. What is missing is neither a framework, nor an inventory, nor a subsidy. What is missing is organised private demand.

The questions I keep coming back to

• Where does our technology spending actually go, not by cost centre but by jurisdiction? The result of that exercise is usually more lopsided than expected, and the share that could shift without measurable loss of quality usually larger.

  
  

• What would happen if one of our critical components became unavailable for three weeks? June made the question concrete, and it calls for a real exercise rather than an answer given around a table.

  
  

• Which is the least critical perimeter in our project portfolio where we could make a lasting commitment to a European provider? That is the most direct contribution an organisation can make to the emergence of an alternative, and the one with the lowest cost of entry.

These questions no longer belong to IT departments alone. They concern executive leadership, procurement and finance, because they bear on capital allocation as much as on technical architecture.

What to watch

For a long time the quality of a technology decision was measured by its cost, its speed of deployment or its return on investment. Those criteria remain essential but no longer suffice. In an environment shaped by geopolitical tension and the growing concentration of digital infrastructure, performance is also measured by what a decision will still make possible tomorrow.

On 12 November, the close of the European call for proposals will reveal which consortia applied, and above all what share of them is genuinely led by European operators rather than by non-European players establishing capacity on the continent. The awards in early 2027 will then show whether the pooled procurement mechanism delivers.

That leaves the question public procurement will not answer on behalf of companies. Eighteen states managed to aggregate their demand in order to bring into being a capability that did not yet exist. Nothing prevents an industry association, a buying group or a handful of large European accounts from doing the same.

The only thing missing is someone, inside our organisations, whose job it is to ask.

You can also find my perspective on this topic in my LinkedIn article.