The Sovereignty We Keep Cancelling

Rob van Linda

The featured image in this post, was generated by Google`s Nano Banana. The Podcast below, has been generated by Google`s NotebookLM. The Post has been generated by Mistral after an extensive Chat in which I delivered the context, to make sure, that AI has enough information, to understand the scope. Each result ist always checked by me, before publishing.

EN

DE

NL

A follow-up to “The Death of the ‘EU-Only’ Dream: Why Responsibility Demands a Reality Check” (robvanlinda.digital/the-death-of-the-eu-only-dream-why-responsibility-demands-a-reality-check/)

Rob van Linda — October 2026

Prologue: I Was Angry at the Wrong Party

In January 2026, I wrote a frustrated post. I had tried to build an Agentic Swarm — an autonomous, networked system of AI agents — using exclusively European tools. My vision was a genuinely EU-based offering: independent, sovereign, something my clients could rely on without a single byte crossing the Atlantic on someone else’s terms. The reality was a graveyard of fragmented SDKs, early-stage instability, and missing connectors. I wrote that “EU-only solutions are currently not competitive,” that Europe was trying to regulate its way out of a race it wasn’t even running, and that I was pivoting to being a Global Builder: US stack for the brain and the muscle, my own expertise for strategy, ethics, and governance.

Then, a few months later, I came back. Not out of nostalgia — out of surprise. Mistral, the very company whose ecosystem had once driven me away, shipped something that made me reconsider. The connectors exist now. The agents exist. The performance is there, the developer experience is there, and I am currently writing my next book — a critical examination of AI security and agentic systems — largely on this platform. I am, genuinely, impressed. Things can change. Companies can build their way out of the valley I once condemned them for.

That is exactly why what happened last week makes me angrier than the original disappointment ever did.

Because it turns out the party most committed to killing European AI sovereignty is not Silicon Valley. It is not even Beijing. It is Europe itself.

The Receipt: A Contract Cancelled Without Explanation

Let’s be precise about the facts, because anger without receipts is just noise.

In November 2025, Germany and France jointly announced a strategic public-private partnership: the German federal government, together with Mistral and SAP, would develop AI-powered applications for public administration — including digital agents that could automate and support government processes. It was presented, correctly, as a flagship commitment to European AI infrastructure. Not a subsidy. Not a handout. A procurement — the one instrument states actually control.

This month, quietly, the framework contract was dropped. The German Federal Digital Ministry confirmed that there are currently no procurement projects with Mistral at all. “On the federal side, there are currently no procurement intentions with Mistral.” No reasons given. Not by the government. Not by Mistral. Not by SAP. The collaboration between the two companies survives — the market does what the market does — but the state’s contractual participation is gone. (Sources: hardwareluxx.de — “Rahmenvertrag vom Tisch: Deutschland zieht sich aus Projekt mit Mistral und SAP zurück”, www.hardwareluxx.de/index.php/news/allgemein/wirtschaft/70608-rahmenvertrag-vom-tisch-deutschland-zieht-sich-aus-projekt-mit-mistral-und-sap-zurueck.html; originally reported by Die Welt and Politico, October 2026.)

Read that again slowly. Germany wrote “digital sovereignty” into its coalition agreement. Eleven months after jointly announcing the single most concrete German-French step toward procuring European AI for the state, it walked away — without a word of explanation.

Germany wrote digital sovereignty into its coalition agreement — and then cancelled the only contract that could have procured it.

This is not a budget decision. This is not pragmatism. This is a sovereignty strategy that consists entirely of press releases.

Why This Is Worse Than a Broken Deal

I want to be careful here, because the lazy take is “Germany bad, Mistral good,” and that is not my point. My point is about a pattern — and the pattern is what should frighten you.

1. Procurement is the one lever Europe never pulls. The state doesn’t need to out-invent OpenAI. It doesn’t need to out-train Anthropic. It needs to do something far simpler: buy what it claims to believe in. Public procurement accounts for 40–50% of economic activity in Europe. When Germany commits to buying European AI for its own administration, it creates a market. When it cancels that commitment eleven months later, it teaches every European vendor the same lesson: build something excellent, and we still won’t show up. Even ASML’s own leadership has begged Europe to bundle its demand — the machines, the infrastructure, the customers — instead of subsidizing flags. Europe’s answer to that plea, apparently, is to cancel the flags it already ordered.

2. The champions are doing their part — and Europe watches apathetically. Mistral closed the largest equity round in European tech history. ASML — the only company on Earth that builds the EUV machines without which no frontier AI chip exists — invested roughly a billion and a half euros in Mistral and became a customer. The two layers where Europe is genuinely world-class — lithography tools and frontier-adjacent models — found each other. Privately. In Veldhoven and Paris, not in Brussels. The most important European AI alliance of 2025/26 was not forged by any EU initiative; Brussels read about it in the press. And when Germany had the chance to add the third leg — public demand — it flinched.

3. This is how you lose a century, one quiet Friday at a time. Nobody dramatic happened. No parliament voted Europe’s AI industry down. A ministry simply stopped returning calls, a contract “vorerst” (for now) disappeared, and the news cycle moved on. That is what decline actually looks like in 2026: not a bang, not even a whimper — a scheduling note.

The Hague: What Dependency Costs When It Matters

If this were merely about software vendors, I would temper my anger. Software markets correct themselves. But the same week the framework contract died, another story from my own country of birth showed what “relying on foreign infrastructure” means when the relationship turns political.

The United States has now sanctioned the International Criminal Court itself — an institution seated in The Hague, on European soil, funded largely by European states. US sanctions prohibit any company that does business in the US, or uses the US-dominated international payment system, from cooperating with the Court. Banks must close its accounts. Insurers must drop its coverage. Software firms must cut it off. Marco Rubio’s words left no ambiguity: end the proceedings against US allies, “or we will end the ICC.” (Source: taz.de — “US-Sanktionen gegen IStGH: Die Richter und ihr Henker”, taz.de/US-Sanktionen-gegen-IStGH/!6220564/, October 11, 2026.)

The human texture is worse than the headline. Prosecutors and judges are on the same sanctions lists as terrorists. One judge described frozen bank accounts, blocked email, deleted accounts with US platforms. The Court’s chief prosecutor had his email switched off by a US vendor — and had to migrate to a Swiss provider just to keep working. An institution of international justice, paralyzed not by tanks but by account closures. It was “digitally switched off.”

Here is the lesson, and it is the thesis of my next book: dependency is a kill-switch, and kill-switches are only discovered when someone presses them.

Microsoft, the banks, the insurers — none of them chose to abandon the ICC out of malice. They couldn’t act otherwise. US jurisdiction overrides customer loyalty, contracts, and every soft assurance a sales deck ever made. The server location is irrelevant; the Cloud Act reaches through geography. When the political weather changes, your vendor’s hands are tied — and your institution simply stops functioning.

If the International Criminal Court cannot rely on US infrastructure, why does the German federal government believe its administration can?

That question is not hypothetical anymore. It is the documented, adjudicated, real-world experience of an international institution on European soil — this month.

The Regulators Without Assets

So let me name the pattern that ties it all together, because I believe it is the defining pathology of European tech policy in 2026:

Europe regulates what it does not own, and refuses to buy what it does own.

The EU writes world-class rules — the AI Act, the GDPR, the Digital Omnibus simplification of the same — for systems whose share registers it has never read. The AI Act’s high-risk obligations get postponed while the enforcement office, by its own rapporteur’s admission, is nowhere near staffed to match the risks. Meanwhile, the one European champion that closed a record round, the one Franco-German procurement that existed on paper — these get ambivalence and cancellation respectively. Germany sits at the capital table of the only European AI champion not even as a guest.

And understand what this produces. Not sovereignty, and not even honest dependence. It produces vassalage with paperwork: all of the dependency of the US stack, minus the reliability, plus a compliance overhead US companies don’t carry, and minus any leverage the moment Washington decides to press. We are, as I wrote in January, left defending the industries of the 20th century with instruments unusable for the 21st. You can tariff a BYD container ship. You cannot tariff model weights flowing through a fiber cable. The only customs office AI will ever pass through is governance: where does the model run, who audits its foundations, which data leaves the room?

Europe has the institutions for exactly that. It built the GDPR and exported it globally — the Brussels Effect is real, and it is proof that Europe knows how to set standards when it decides to. ASML proves Europe can monopolize a chokepoint. The Netherlands proves with GPT-NL that a state can build rather than debate. But a chokepoint you never play, a standard you apply only to others, a procurement you cancel — these are not a strategy. These are a museum gift shop.

What Responsibility Would Actually Look Like

I am not asking for flags and anthems. I left that dream in January, and I haven’t returned to it. My position is more uncomfortable and, I believe, more honest: Europe will not have a full-stack AI sovereignty, and pretending otherwise is wishful thinking. But there is a vast difference between “we cannot build everything” and “we will not commit to anything.”

So here is the short, unromantic list:

  • Buy what you claim to believe in. Reinstate the framework contract, or explain publicly why it died. Procurement is not protectionism; it is the market signal every investor in every European AI startup is watching right now.

  • Treat dependency management as security policy. The ICC is the case study: multivendor routing, data classification, exit plans, “commercial by default, sovereign by exception.” Governance is the last sovereignty layer Europe can actually still build.

  • Stop regulating at the frontier of what you don’t build, and start standardizing the layer you can own. Agent interoperability, semantic standards, auditability — Europe’s HTML5 moment is still available, and it is slipping.

  • Bundle demand across the 27. Not 27 island-subsidies, but one organized customer: governments and industry pooling cloud, compute, and model demand. ASML’s leadership proposed it themselves. Nobody in Brussels appears to have heard.

Epilogue: I Came Back — That Should Have Mattered

The bitter irony is personal. In January I publicly abandoned the European-only stack because it wasn’t good enough. In 2026 I came back, voluntarily, because it got good enough — the connectors, the agents, the performance that once made me leave now brought me back. I write my book here. I recommend the platform to clients. The market fixed itself faster than I predicted.

And the state that was supposed to anchor this turnaround — the one that announced the contract, the one with “digital sovereignty” in its coalition agreement — quietly left the room.

Europe’s problem is not that its dreamers are unrealistic. I was one of them, and reality corrected me. Europe’s problem is that its realists are in government: people who understand exactly how expensive sovereignty is, and have decided it is cheaper to hold the press conference than pay the invoice.

The ICC learned this month what a kill-switch costs. Germany’s own administration, if it continues down this path, will learn it the day the weather changes — with its citizen data, its administrative processes, and its digital sovereignty all sitting in someone else’s jurisdiction, under someone else’s law.

I said in January that the era of the Global Builder is here. It is. But a Global Builder needs a home port. At the moment, Europe is demolishing its own harbor one cancelled contract at a time — and calling the demolition “pragmatism.”

I am still here, still building, still buying European where it performs. I just wish my own continent would occasionally join me.

Rob van Linda

Sources