Why Europe Keeps Failing to Build Its Own Cloud

In April 2026, the European Commission awarded a tender worth up to 180 million euros over six years for its own sovereign cloud services. Four winners: a Post Telecom consortium with Clever Cloud and OVHcloud, Germany's StackIT, France's Scaleway, and a Proximus-led group whose members include S3NS, which is a joint venture between Thales and Google Cloud. Read that last clause twice. The flagship procurement of European cloud sovereignty, the one meant to prove Europe can stand on its own infrastructure, includes Google inside the tent, wearing a French badge.

I don't bring that up to sneer at the Commission, which was solving a real procurement problem with the vendors that exist. I bring it up because it's the perfect specimen of a twenty-year pattern, and the pattern is the subject of this post. Europe keeps trying to build a thing, a champion, a European AWS, and keeps discovering at the moment of purchase that the thing isn't there. My argument: the failures share one root cause, which is that the goal has been defined as a noun when the deficit is a property. What Europe lacks isn't a homegrown hyperscaler. It's the ability to move, and the initiatives that quietly pursue that property are succeeding while the champion-building keeps failing.

A short history of announcing victory

The modern era of this story opens in 2019, when GAIA-X was announced with Franco-German fanfare as the federated data infrastructure that would give Europe technological self-determination. What followed is well documented and a little painful. Membership swelled, and swelled to include the American hyperscalers whose dominance had motivated the project, which is a bit like a temperance society admitting the distillers for their industry expertise. Working groups multiplied. Specifications shipped. A European AWS did not.

Fairness requires the other half of the ledger, because "GAIA-X failed" is the lazy version. Its policy rules and labeling concepts survived and got absorbed into the Commission's own cloud sovereignty framework, and the vocabulary Europe now uses to grade clouds owes a lot to that work. That's a real legacy. Notice, though, what kind of legacy it is: standards, criteria, labels. Properties. The federation of European alternatives that would compete with AWS on capability never materialized, and the parts of GAIA-X that lived are precisely the parts that weren't trying to build one.

Then came the manifestos, EuroStack most prominent among them, arguing for a full European technology stack from chips to applications. The diagnosis in those documents is often excellent. The prescription is always the same: coordinated investment to build European alternatives, a champion by consortium. And the money, when it finally arrives, looks like April's tender: 180 million euros, six years, four consortia, one of which brings Google. For scale, Microsoft alone announced roughly 80 billion dollars of AI datacenter spending for a single fiscal year. The hyperscalers deploy Europe's flagship sovereign budget about every ten days, as a rounding error in a capex line.

Why champions don't take

It's worth being precise about why the champion strategy keeps missing, because "not enough money" is only the visible layer.

Demand is the quiet killer. European buyers, including public ones, choose US clouds daily because they're better and cheaper for the job at hand, and no amount of summit communiques changes a procurement officer's incentives. A champion needs customers who choose it when it's worse, for years, while it climbs the capability curve, and Europe's fragmented demand never coheres into that kind of patient purchasing. Airbus, the counterexample everyone reaches for, had exactly that: two decades of governments committing to buy aircraft, one product line, captive flag carriers. Cloud is two hundred product lines and nobody's flag carrier.

Supply fragments along national lines. Look at the tender winners again: a Luxembourg-led group, a German one, two French-flavored ones. Each member state wants its own champion, so Europe funds four half-champions per cycle, and the consortium structure that makes the politics work makes the engineering slow. Meanwhile talent gravity does the rest; the engineers who'd build a European control plane are mostly building American ones, from Dublin and Munich, at salaries the consortia can't print.

None of these forces are fixable by announcement. Several are barely fixable at all.

The category error

Now set the failures next to the successes, because Europe has been quietly succeeding at something in parallel, and the contrast is the whole argument.

The Data Act killed exit egress fees years before its switching-charge ban even binds. DORA made vendor-exit strategy a supervised compliance artifact for an entire sector. CADA's tier list is reshaping procurement vocabulary while still a proposal. Every one of those is regulation pursuing a property, the ability to classify, compare, and above all leave, and every one of them moved markets. The hyperscalers restructured billing, built sovereign partitions, and hired compliance armies in response. Nobody restructured anything in response to a federation whitepaper.

That's the category error in one frame. A champion is a thing you must build, fund, staff, and protect against every force in the previous section. Portability is a property you can legislate, standardize, and verify, and it compounds: every workload that becomes cheap to move enlarges the addressable market of every European provider that exists, including the tender's four winners, including the small ones nobody subsidizes, including us. The champion strategy asks Europe to out-build American capex. The property strategy asks it to do what Brussels is actually good at, which is writing rules that make markets contestable, and the last three years suggest that works.

Sequence matters here, and it runs opposite to the intuition. Build the champion first and portability never comes, because the champion's backers need lock-in too, just pointed at a different flag. Build portability first and capacity has a reason to grow: workloads that can move create demand that Scaleway and StackIT and OVHcloud can win incrementally, product by product, without needing to be AWS on day one. The property creates the market the champion was supposed to conquer.

The steelman: a right to move needs somewhere to go

The strongest objection to everything above: portability without capacity is its own theater. If every serious workload's realistic destinations are three American clouds, then the freedom to switch is the freedom to circle, and Europe's dependency survives with better paperwork. Capacity is real sovereignty; datacenters, chips, and power are the hard part, and CADA's own drafters agree, which is why the proposal pairs its tier list with a push to triple European datacenter capacity. There's also a fair jab at people like me: a portability layer is conveniently the thing a small infrastructure vendor can build, so of course we'd rank it first.

Concede the structure of that, and one more thing besides: some champion-building has worked at the component level, and a Europe with Mistral in the model layer and a serious chip effort is better positioned than one without. But the objection proves timing, and the argument was about sequence all along. Capacity built behind a locked-in market sits idle, which is the GAIA-X-era lesson; capacity built into a contestable market fills, which is what the Data Act era is starting to show. Europe should absolutely pour concrete. It should just stop expecting the concrete to conquer anything before the switching costs fall.

Our own position in this is unhidden: we're a small European vendor whose product thesis is a deterministic, portable model of infrastructure across clouds, so the property-first argument is also our book, discounted as always. The test I'd offer is simple and falsifiable. Watch which European providers grow over the next three years: the subsidized consortia, or whoever makes leaving a hyperscaler boring. My bet is on boring. Where's yours?

Related: CADA: Europe Just Wrote a Tier List for Clouds, the property-first strategy in regulatory form. More about what we're building at light-cloud.com.