The 1% number
Ninety-nine percent of the European Commission’s own cloud runs on American providers. One percent on a European one: OVHCloud. The body that writes the rules for European digital sovereignty cannot itself host its email outside the reach of a foreign government.
The most ambitious digital policy programme in the world, designed to free Europe from precisely this dependency, cannot free its own back office from it.
This is what the political conversation about “European Big Tech” is up against. Not “we need to catch up”, we are far past catch-up. We are at the point where the policy machine cannot even self-host. Every argument about Mistral, EuroStack, sovereign cloud and European champions has to start here, with the one percent.
And once you start there, a different question opens up. Not “how do we build a European version of Microsoft,” but “what would actually fix this?”
The two roads
There are two roads in front of European digital policy.
The first road is the one being marketed loudly right now. Pick winners. Funnel public money and procurement into “European champions” — Mistral, Aleph Alpha, EuroStack — and hope for a continental Microsoft, a continental Oracle, a continental OpenAI. Brand them with a flag. Call it sovereignty.
The second road is quieter and a lot less photogenic. Use public power: procurement, funding, mandates, regulation, to scale the commons that the engineering layer of the world already runs on. Open source. Federated standards. Public infrastructure that no one owns and no one can buy.
These roads do not end in the same place. The first one ends where every concentration story ends. The second one is the only road that, when it has been tried at scale, has actually worked.
I am going to make the case for the second one. But to do that honestly I have to start by being clear about where the first one goes.
Where the champions road ends
The “European champion” pitch is not new. It has been tried in cloud, in social, in search, in chips. What is new is that we now have, in 2026, a fresh and very visible American example of where this thinking ends when you let it run all the way to its conclusion.
In January, one day after Donald Trump’s second inauguration, Larry Ellison stood at a White House podium next to Sam Altman and Masayoshi Son and announced Stargate — a $500 billion AI infrastructure deal between Oracle, OpenAI, SoftBank and the United States government. WIRED would later call Ellison a “shadow president.” His family now controls Paramount, CBS and MTV, and the Skydance bid for Warner Bros adds CNN, HBO and the HBO library on top of one of the world’s largest databases. The year before Stargate, Ellison gave a talk in which he described an AI surveillance future where “citizens will be on their best behavior”. He was not a critic of that future. He was a vendor.
Oracle is not an accident of history. The company started in 1977 on a CIA database contract — the project was literally called “Oracle.” Forty-eight years later it is fused to a US administration, with a half-trillion-dollar infrastructure programme and effective ownership of large parts of American mass media.
It is reasonable, looking at this, to ask: is “let us build one of those, but European” really the safer bet?
Look at Palantir. The same Foundry platform now sits inside the UK’s NHS Federated Data Platform on a £330m, seven-year contract, while the same parent company runs ICE’s ImmigrationOS deportation-targeting tool in the United States, drone strike targeting via Project Maven, and predictive policing in the West Bank. The UK’s CLOUD Act exposure means the NHS contract cannot override US federal law — when Washington asks for the data, Washington gets the data. In March, UK science minister Patrick Vallance told MPs that future deals would be done “differently.” A graceful way of admitting the road has a destination, and Britain does not want to arrive at it. The Netherlands has been on a parallel script: the Dutch Marechaussee was quietly running border screening on Palantir while the Justice Minister told parliament otherwise — I wrote about that and the wider Dutch sovereignty mess recently.
Or look at Próspera, on the Honduran island of Roatán. Backed by Peter Thiel, Marc Andreessen, Balaji Srinivasan and Brian Armstrong via Pronomos Capital, it is a literal corporate jurisdiction. Companies pick their own regulatory framework — Singapore banking, Japanese biotech — or opt out. One percent business tax. Bitcoin as legal tender. Private arbitration replaces courts. After Honduras’s Congress unanimously repealed the law that created it, Próspera filed at ICSID for up to $10.8 billion in damages — roughly two-thirds of the entire Honduran national budget.
Stargate, Palantir, Próspera. They are not three unrelated stories. They are three layers of one arc: scale, then media capture, then state fusion, then — at the limit — replacement of the state itself. Corporate-state fusion is the historical hallmark of authoritarianism. I will leave the F-word in your hand.
This is the road. A European Oracle is still an Oracle. A European Palantir is still a Palantir. The flag does not change the chemistry.
And lest anyone imagine this dynamic stops at the Atlantic: ASML — Europe’s industrial crown jewel, and the lead investor in Mistral — has already openly threatened to leave the Netherlands over Dutch policy on expat taxes and migration, and is still publicly pressuring The Hague on fiscal consistency a year later. The government’s response was a cross-ministry rescue programme nicknamed “Operation Beethoven.” That is what concentration buys, even at home: a champion large enough to hold its host state to ransom over its own future.
Why it probably fails anyway
Set the values question aside for a moment. Even on its own terms, the champions road is a poor bet.
Mistral, the supposed European OpenAI, is around 2% of the global LLM market. It sits at the bottom of capability benchmarks. Its compute runs on American GPUs and US-owned cloud — a hardware-layer dependency I have argued the EU can only really fix by getting back into silicon. Its largest shareholder, after a €1.7bn round at an €11.7bn valuation, is ASML. The “proof we can do scale” looks an awful lot like a single private-capital syndicate hedging.
Aleph Alpha is in a similar shape. So is most of the EuroStack roster. The price-tag estimates for a real European stack run from €300bn at the optimistic end (Bria/Timmers) to €5tn for a full one. The timeline is a decade, minimum.
Meanwhile, Estonia and Greece have signed OpenAI deals to deliver education AI to schoolchildren. The European Parliament uses Anthropic’s Claude. The European Commission’s own cloud, again, is ninety-nine percent American. Even the believers do not buy European.
European cloud market share has actually fallen — 27% in 2017, around 15% today. US hyperscalers sit at roughly 65%. The gap is not closing. It is widening, in the segment supposedly being defended.
Cecilia Rikap at UCL’s Institute for Innovation and Public Purpose has made the same case at much greater length: replicating the hyperscaler model in Europe would be a sub-optimal solution; what Europe actually needs is a public-led, democratic digital value chain. She is right, and the math is not subtle. We are not going to out-Oracle Oracle. We are not going to out-OpenAI OpenAI. Not in this decade and not on this budget. Pretending otherwise is wishful thinking.
What winning already looks like
There is a road that has worked. We are on it. We have been on it for more than thirty years and most of the policymakers have not noticed.
In 1991 a Finnish student named Linus Torvalds released a hobby kernel named Linux. Today Linux runs 100% of the world’s TOP500 supercomputers, and has every year since 2017. It runs 96.4% of production Kubernetes clusters. 96.3% of the top one million web servers. Over 90% of public-cloud workloads. Microsoft Azure itself runs 61.8% Linux virtual machines. Google Cloud runs 91.6%. Android, built on the Linux kernel, ships on roughly 70% of the world’s smartphones.
The Linux ecosystem is now a roughly $22bn market, projected at $100bn by 2032, growing at 20.9% CAGR. 78.5% of professional developers use it as a primary or secondary OS.
Linux did not win by becoming Microsoft. Linux won by being structurally different from Microsoft. It has no owner to capture, no shareholder to please, no kill switch a foreign government can pull. You cannot sue it into a deportation contract. You cannot put its CEO on a White House stage. There is no CEO. The only thing it has is Linus being Linus (bless you buddy).
That is the point. The thing that makes Linux unbeatable is not its technical brilliance, real as that is. It is its non-capturability.
Every concentration story we just walked through: Stargate, Palantir, Próspera, depends on something a small group can own. The commons cannot be owned. That is the whole, durable, exportable insight of the open source movement, and it is already sitting in the EU’s hand, deployed at planetary scale, waiting for a public strategy worthy of it.
Ideology is not a dirty word
There is going to be an objection, so let us get to it. “You are being political. Tech policy should be neutral.”
Tech policy is not neutral and has never been neutral. Every choice of standard, every procurement clause, every funding line is a values choice. “Let the ecosystem choose itself” is not neutrality — it is anarchism for billionaires, and it is the explicit ideology of Thiel, Andreessen, Ellison and the network of investors behind the cluster I just described. They are happy to call it ideology, in their books and on their podcasts. It is only on policy panels that they suddenly become humble engineers.
I am going to be honest about my own frame. I think the question of European digital policy is, at root, a question about whose values get baked into infrastructure that 450 million people depend on. That is a democratic question, not a technical one. And I think the answer should be: not Larry Ellison’s, not Peter Thiel’s, and not the values of whoever Mistral’s largest shareholder happens to be in five years either.
That is the ideology. I would rather state it than dress it as neutrality.
The role only governments can play
Here is the part that often gets confused.
ASML putting €1.7bn into Mistral is private capital doing what private capital does. I have no objection. Investors invest. That is allowed and there is nothing to legislate against in a coordination round between two European companies. But it is not a public strategy.
Public money and public mandates have a different job. Private capital chases concentration, because concentration is where the returns are. The unique thing governments can do, and that nobody else can do, is fund the things private capital structurally will not: maintenance, public goods, federated infrastructure, the unglamorous middle of the dependency graph that everything else sits on top of.
That means several concrete things. Public Money, Public Code as binding law. Procurement that weights open source and disqualifies extraterritorial-law exposure for critical systems — the Palantir case proves you cannot contract around the CLOUD Act. A real EU-level Sovereign Tech Fund, scaling Germany’s STA model fifty- to one-hundred-fold, with multi-year horizons that let maintainers plan a career instead of a grant cycle. Authoritative EU-jurisdiction mirrors of npm, PyPI and the major container registries, so a single American court order cannot pause a continent — Packagist, already run out of Europe by Composer’s own maintainers, is the proof that this is doable. Improved CRA legislation for individual maintainers, Open Source Program Offices. Fork rights in perpetuity for any code paid for with public money. A small, fast “fork-and-fund” unit that can stand up the next Valkey or OpenTofu within weeks of the next Redis-style relicense.
None of that picks a national winner. None of it is a champion. All of it pours public power into the only road that has actually worked.
The window
This is a policy window, with a deadline. The Cloud and AI Development Act is in front of the Commission. The EU Open Source Strategy consultation is open this quarter. Germany’s Sovereign Tech Agency is hiring. The French/German digital sovereignty task force reports later this year. France is moving 2.5 million government desktops off Windows. Schleswig-Holstein is eighty percent migrated to LibreOffice. Denmark is in. They are each rebuilding the stack separately — a coordination problem I sketched a fix for recently — and even uncoordinated, the political tolerance for disruption is the highest it has been in a generation. It is not going to stay that way.
We can spend the window building the second copy of Oracle. Or we can spend it on the only road that does not end where the first road ends.
A European Oracle is still an Oracle. Linux did not beat Microsoft by becoming Microsoft. Whose values get baked into the infrastructure 450 million people depend on is a choice that is being made, with public money, this year.
I would rather we made it on purpose. If you work in EU policy, public-sector IT, an OSPO, a foundation, or you are a maintainer Europe quietly depends on: I want to hear what would change your mind, and what is in your way. Find me on LinkedIn or get in touch. I am writing a series.
One last thing
Since the rest of this piece has been Europe-shaped: the commons does not respect borders. Once Europe funds a kernel, a registry, a federated identity wallet, an open model, that infrastructure is available to every democracy that needs it. Brazil, India, South Africa, Canada, Japan, the smaller republics still working out where to host public services. None of them want to depend on Stargate either, and none of them have €300bn for their own EuroStack. What they can plausibly use is what Europe is in a position to build. Funding the commons is not a parochial European project — it is a gift, deliberate or otherwise, to every democracy that currently has no vendor at all.