Why Europe keeps losing the tech champions it grows¶

A plain-language companion to the draft working paper "Sovereignty as a Flow: Capture and the Durability of European Technology" (v0.3, July 2026). The paper sets out the registry, the statistics, and the caveats; this text explains the ideas.

The paper this condenses Read the full draft working paper (PDF)

The short version. Policy debates treat a technology firm's nationality as a fact: this firm is European, that one is American. In reality nationality is a snapshot of a cap table, and cap tables are liquid. We tracked how control of European technology firms actually leaves Europe, across 24 documented captures and a defined universe of 125 prominent firms over twenty years. One structural feature stands out: firms whose governance includes a formal instrument that lets an insider refuse a takeover stayed European in 81 percent of cases, against 59 percent for the rest, a gap the sample is too small to confirm, and the one lock tested by a hostile bid held. The lesson for policy: money buys champions; structure keeps them.

Four doors out¶

Skype, DeepMind, Arm, Nuxeo, Supercell, Wolt, Software AG. Each was, at some point, a European sovereignty hope; none is under European control today. Across 24 such captures, control leaves through four doors.

The front door is acquisition: a foreign buyer purchases the firm outright (Skype to Microsoft, DeepMind to Google, Arm to SoftBank, Nuxeo to Hyland). The second door is the private-equity buyout: a fund takes the firm private, restructures, and typically exits to a strategic buyer, often American (Darktrace, Software AG). The third door is quieter: an infrastructure deal leaves the logo European while moving the firm's stack, computing, or distribution onto a foreign platform (the sovereign-cloud ventures that run an American provider's product; DeepL, Germany's flagship language-AI firm, migrating to Amazon's cloud, is a case to watch until its exit cost is measured). The fourth door is the US listing with a move of incorporation: the founder keeps the keys and moves the company itself to the United States (UiPath, now a Delaware company run from New York). A US listing that leaves the company incorporated in Europe (Spotify, Klarna, Wise) moves the capital and the investors, and we count it as half a step.

The doors differ, and so does what blocks each of them.

The one feature associated with staying¶

We assembled a universe of 125 prominent European technology firms from the last two decades, spanning cloud, AI, fintech, security, enterprise software, semiconductors, consumer platforms, and developer tools, including the captured and the still-independent alike. For each firm we asked one structural question: does its governance contain a lock, some arrangement that lets an insider say no to an unwanted change of control? A foundation or steward-ownership structure, founder shares with extra votes, a golden share, state control. A founder who simply owns the majority does not count: whether that founder will refuse is a matter of intent, and intent is what the outcome shows.

The split points the expected way. Locked firms stayed under European control in 81 percent of the resolved cases; unlocked firms in 59 percent. Put the other way: an unlocked prominent European tech firm had roughly a 41 percent chance of leaving European hands, and a locked one 19 percent. On 102 resolved cases that gap could still be chance (the statistical test gives p = 0.08), so it is a lead worth following. It is not a disguised size effect: the gap holds among firms of similar size.

The exceptions are the most informative part. Four locked firms did end up counted as captured, and each tells the same story from a different side. Software AG's foundation chose to sell its stake to Silver Lake. UiPath and GitLab left through a voluntary US listing: the founders kept control and chose to move. Farfetch went through insolvency: its equity was wiped and the lock died with it. No lock was overridden, and the one tested by a hostile approach held: Ubisoft's founding family fended off Vivendi's creeping stake between 2015 and 2018. A lock guards the front door. It does nothing at the two doors it was never built for: a founder who wants foreign capital, and running out of money. The most tightly locked firm in the set, a state joint venture, resisted every acquirer and went bankrupt anyway. Surviving as European and surviving are different things.

Two more facts round out the picture. A lock is not the only way to stay European: SAP, Adyen, Amadeus and Infineon have ordinary one-share-one-vote governance and stayed European for decades. Their size alone does not explain it, since Arm, Autonomy and Supercell were as large when they were bought. And there is one kind of asset that cannot be bought at all: an open-source commons. When Oracle acquired MySQL, the code was forked into MariaDB and the alternative outlived the acquisition. A proprietary firm is a discrete, purchasable asset; a forkable commons is capture-proof by construction.

What this means for policy¶

The practical readings follow the mechanism, one instrument per door.

Against the front door: policy can reward the lock. Procurement rules and sovereignty labels can credit steward-ownership, foundation ownership, and dual-class structures with a domestic anchor; public co-investment can come with a golden share. Germany's steward-owned technology providers show the model runs at scale.

Against the fourth door: deepen the capital. Firms list in New York because European growth capital and public-market liquidity are shallow. Capital-markets integration is usually filed under finance policy; on these numbers it is also a sovereignty instrument, because it removes the one exit a lock cannot block.

For everything else: a commons. Where a firm will never adopt a lock, European control can still rest on the capability itself if it lives in an open substrate that survives any single firm's sale. Size helps less than it seems: Arm, Autonomy and Supercell were bought from the top of the size range.

And one caution for the subsidy debate: on these numbers, a grant to a prominent unlocked firm funds an asset with about a two-in-five chance of leaving within the period we observe, and the grant itself changes none of the structure that sets those odds.

The small print¶

The universe is a defined and transparent sample of prominent firms, so it licenses the comparison inside the sample and no precise economy-wide rate. The headline association is exactly that, an association with a documented mechanism: profitable firms might both adopt locks and have no need to sell, a possibility our design cannot fully separate out. The absence of overridden locks is close to true by definition, since a lock is an insider's power to refuse; the test that matters is what happens when a bid arrives, and most bids are never made public, so the one public case illustrates the mechanism without testing it. Every case, coding, and computation is sourced in the paper's public dataset.