"We cannot afford to depend on others for the technologies that keep our hospitals running, our energy grids stable and our services secure." — Ursula von der Leyen, 3 June 2026
I. The Call from Cincinnati
It was a Sunday in February, and my cousin Andrei was on screen from Cincinnati(he is in Texas now, I didn't ask for details). He had been there eighteen months, working for a mid-size SaaS company doing cloud infrastructure — containerised systems, Kubernetes, the unglamorous plumbing that makes modern software run at scale. He had a Botoșani accent that the Ohio winters had not quite managed to erode, and he wore the expression of someone who had stopped pretending to be surprised by things. Behind him, visible through a narrow window, was a landscape that looked nothing like home: flat, grey, American in a way that is hard to describe without sounding like a foreign correspondent pretending expertise about a country glimpsed through a webcam rectangle.
We talked for two hours, the call stretching past the point where either of us had planned to stop. Not a recorded interview — just a family video call that drifted, as family video calls do, from the mundane (how expensive Ohio winters are, whether our aunt's garden had survived the frost) to the structural. Andrei is not a philosopher. He is a systems engineer who thinks in latency and in containers. But sometimes the most lucid diagnoses come from people who are paid to make things work, not to make things sound good.
At some point I asked him directly: What's the actual difference? Between working in tech in Romania versus working in tech in the United States?
He laughed — not with amusement, but with the specific exhaustion of someone who has been asked to explain gravity.
"Petru," he said, "it's not that the Americans are smarter. It's not even that they have more money, though they do. It's that the system assumes you'll succeed. Here, if you have an idea, the machine bends towards you. Loans, mentors, investors, second chances. In Romania — in Europe — the machine assumes you'll fail, and it builds in friction to prove itself right."
He was not bitter. He was precise. And that precision, delivered from a rented apartment in a mid-tier American city, contained more analytical power than most of the policy documents I had been reading for weeks.
This article grew from that conversation. It grew also from the legislation that landed on the European Commission's desk on 3 June 2026 — the Tech Sovereignty Package, the Cloud and AI Development Act, the Chips Act 2.0 — and from Cory Doctorow's Enshittification, which I had been reading in parallel and which provided, unexpectedly, the philosophical scaffolding I needed to understand not just what Europe had built, but why it had been allowed to rot at the foundations while looking perfectly presentable from the outside.
II. The Feudal Topology of the Digital World
My father was a mechanic. He worked with machines that had a physical logic: if something broke, you could open it, see the broken part, replace it, close it. The cause was visible. The remedy was local. The tools were yours.
The digital economy runs on a different logic — one that my father would have recognised as deeply unfair, even if he would not have had the vocabulary for it. The tools are rented. The raw materials (data, attention, behaviour) are harvested from you. The infrastructure — the roads, the workshops, the warehouses — belongs to three companies in Seattle, San Francisco, and Redmond. And the rent, unlike in feudalism, is not paid in grain or in labour but in something more intimate: in the architecture of your choices, in the shape of what you can think of doing next.
This is not metaphor. It is topology.
Consider what the European Parliament's own research disclosed in 2025: AWS, Microsoft Azure, and Google Cloud hold approximately 70% of the EU cloud market. European providers — all of them combined — have fallen from 29% market share in 2017 to roughly 15% by 2022, where the number has since flatlined. The largest European cloud provider, SAP, captures approximately 2% of its own continent's market. And around 80% of EU corporate spending on software and cloud flows to American vendors.
This is not dependency in the soft, political-science sense. This is dependency in the engineering sense: remove the dependency and the system stops functioning. Europe's digital infrastructure — its hospitals, its tax authorities, its energy grids, its universities — runs largely on land it does not own.
The feudal lord in this arrangement does not arrive on horseback. He arrives via a Terms of Service update.
III. What Andrei Saw From Inside the Machine
I want to stay with my cousin's account for a moment, because it complicates the usual narrative in useful ways.
Andrei is not a political animal. He did not leave Romania because of ideology. He left because his salary doubled immediately, and would double again within three years if he performed competently. He has no particular animosity towards Europe, and he has no particular loyalty to America — he is loyal to the work, and to the quality of the environment in which the work can be done.
What he described about that environment was not glamorous. Ohio is not Silicon Valley. His company is not OpenAI. But even in this middle register of the American tech economy, certain things were simply different.
"When I joined," he told me, "I had a budget. Not a big one — maybe $3,000 — to spend on tools and equipment in my first month. Nobody asked me to justify it in advance. I just had it. In Romania, getting a second monitor approved took six weeks and a signature from someone in Bucharest who had never met me."
This is a small example. But small examples are how structural realities announce themselves. The American tech workplace, even at its most mediocre, runs on an assumption of productive trust. Resources are allocated to the person closest to the problem. Decisions are made at the level where the information actually lives. Failure is expensive but not fatal.
In Europe — and I write this as someone who has operated a web agency in Botoșani for over eight years, dealing with clients who are Romanian public institutions — the structure runs in the opposite direction. Resources flow upward. Decisions are made far from the information. Failure is permanent, which means risk is avoided, which means nothing ambitious gets attempted, which confirms the assumption that nothing ambitious is possible.
Andrei did not call this feudalism. But he described its symptoms with the precision of someone who had lived on both sides of the boundary.
IV. The Investment Gap: What the Numbers Actually Say
Here is where I must be honest about the document that inspired this piece — and where I must depart from it.
An earlier structural analysis circulating in my notes, useful in many respects, overstated Chinese AI investment while understating the European deficit at its true, vertiginous scale. The Stanford AI Index 2026 — the most comprehensive independent dataset we have — tells a different story, and a more alarming one.
In 2025, private AI investment in the United States reached $285.9 billion. Europe — the entire continent, not just the EU — attracted $20.9 billion. China attracted $12.4 billion, a figure that is lower than often claimed, partly because Chinese investment increasingly flows through state channels that resist easy categorisation.
These are not numbers that admit of optimistic reframing. The ratio is not 3:1 or 5:1. It is closer to 14:1. Europe is not in second place. It is in a different race.
The Stanford report also notes that in 2025, US organisations released fifty notable AI models. China's output is beginning to close that particular gap. Europe does not appear in the count in any meaningful way. Mistral exists, and it is genuinely impressive within its context, but one company does not constitute an ecosystem.
This structural position was not created overnight, and it will not be corrected overnight. It was created by decades of policy choices: by the fragmentation of European capital markets, by the conservatism of European banks and institutional investors, by the taxation regimes that penalise equity over debt, by the absence of a unified legal framework for early-stage companies.
The European Commission has now acknowledged this, in the most direct terms it has ever used. The Scaleup Europe Fund, announced in 2026, is a €5 billion initiative — €1 billion in public capital from the European Innovation Council, €4 billion to be raised from private investors — managed by EQT, a global growth-equity firm. First investments are expected in autumn 2026. The Jacques Delors Centre describes it as addressing "a critical gap in Europe's innovation financing landscape."
The gap it is addressing is real. But €5 billion, measured against an annual American private AI investment of nearly $286 billion, is not a correction. It is an acknowledgement — which is, at least, a beginning.
V. Enshittification Comes to Brussels
In October 2025, Cory Doctorow published a book with a title that would have been unprintable in most newspapers a decade ago: Enshittification: Why Everything Suddenly Got Worse and What To Do About It. The American Dialect Society had already made "enshittification" its 2023 word of the year. By 2025, it had entered Merriam-Webster.
The concept describes the lifecycle of digital platforms with uncomfortable accuracy: platforms begin by being genuinely good to users, to attract them. Once users are locked in, the platform degrades the user experience to funnel value toward business customers. Once business customers are locked in, the platform degrades the experience for them too, to funnel maximum value to shareholders. Eventually the platform dies — but only after extracting everything it can, and only after building walls that make it nearly impossible to leave.
Facebook. Twitter / X. Uber. Amazon Marketplace. Google Search. The pattern holds across all of them.
What Doctorow's framework adds to the standard monopoly critique is an account of mechanism. The enshittification is not malicious in the way we usually mean malicious — it is not done out of hatred for users. It is done because the structural incentives make it rational, and because the four disciplines that once constrained it (regulatory oversight, worker power, competition, and interoperability) have been systematically dismantled.
This is where Europe's situation becomes doubly interesting, and doubly painful.
Europe has, in principle, the strongest regulatory framework for technology in the world. GDPR. The Digital Markets Act. The AI Act. These are serious instruments, built by serious people, with genuine protective intent. But Doctorow's analysis suggests that regulation alone, applied to platforms you do not own, may be insufficient — and may even create perverse effects.
When GDPR increases compliance costs, it increases them disproportionately for smaller companies without dedicated legal teams. The result is that platforms already too big to fail become harder to challenge. The walls around the garden are built not with bricks but with compliance documents, and the biggest players have the most bricklayers.
There is a șopârlă in this — an Aesopic lizard, slipped under the stone of apparent progress. The regulation that was supposed to protect European users from American platforms may, in its current form, function primarily as a barrier to European competitors who might replace those platforms. The patient is quarantined, but the disease remains. Brussels has built a beautiful fence around the field it no longer owns.
VI. The Sovereignty Package: Legislation as Confession
On 3 June 2026, the European Commission did something unusual. It presented a legislative package — the European Technological Sovereignty Package — that read less like forward-looking policy and more like a delayed admission of structural failure.
The package has four components: the Cloud and AI Development Act (CADA), Chips Act 2.0, an EU Open Source Strategy, and a Strategic Roadmap for AI in Energy. Together they constitute what Commission President von der Leyen called "a fundamental shift in Europe's approach to technology."
The most significant instrument is CADA. It identifies two critical vulnerabilities: a structural deficit in European data centre capacity, and dependence on a small number of non-EU cloud providers. Its ambitions are concrete: to triple EU data centre capacity within five to seven years; to create AI Gigafactories — large-scale AI training facilities — giving European researchers and startups access to computational resources they currently lack; to introduce a cloud sovereignty framework that will determine who may sell cloud services to the European public sector.
Legal scholars at the European Internet Policy blog noted that CADA represents the first time the EU is proposing to make "sovereignty" an operative, auditable, enforceable criterion — not a rhetorical aspiration. The Bird & Bird law firm, analysing the package, observed that it "marks a major change in how the EU approaches its tech ecosystems," while noting, soberly, that adoption is not expected before the end of 2027.
The Chips Act 2.0 addresses the semiconductor dimension: Europe produces advanced lithography machines (ASML, in the Netherlands, is the sole global supplier of extreme ultraviolet machines) but has no advanced chip fabrication capacity of its own. TSMC, Intel, and Samsung are the manufacturers that matter; Europe is their customer, not their peer.
There is something clarifying about this package, and something troubling. What clarifies it is that the European Commission has now formally acknowledged, in legally operative language, that digital dependence is a strategic vulnerability rather than an acceptable market outcome. This matters. The word "sovereignty," when it appears in a Commission Regulation rather than a political speech, acquires a different weight. It becomes auditable. It becomes enforceable, at least in principle. It creates grounds for legal challenge when it is violated. That transition — from aspiration to operative criterion — is genuinely significant, even if its practical effects will take years to materialise.
What it took to get there — years of creeping dependency, the geopolitical shock of watching US government directives suspend access to frontier AI models for foreign nationals, the embarrassment of European healthcare systems running on American cloud infrastructure during crises — is a history of expensive lessons.
What troubles it is the timeline. The original Chips Act took two years from proposal to adoption. CADA touches on far more politically sensitive terrain — it will directly affect the business models of AWS, Azure, and Google in the European market — and "comparable friction" in its legislative journey is, if anything, an optimistic forecast. A regulation that will not be finalised before late 2027 will not produce its intended data centre infrastructure before the early 2030s. By which point, if the current trajectory holds, the AI models that will run on that infrastructure will be entirely American in origin.
VII. Brain Drain, Reversed and Reversed Again
My cousin Andrei is a data point in a well-documented phenomenon. The Euronews investigation published in January 2026 reported that net tech talent inflows to Europe fell from approximately 52,000 in 2022 to just 26,000 in 2024. A 2024 report by the research organisation Interface found that European countries are "losing significant AI talent, both national and international, to the United States." Germany sends large numbers of AI professionals to the US and UK. France loses more AI professionals than it gains. The European Parliament has formally acknowledged the crisis: low salaries, the absence of frontier research opportunities, and bureaucratic friction discourage scientists from staying or returning.
The Stanford AI Index 2026 adds a strange asterisk to this picture: the volume of AI researchers migrating to the United States has crashed 89% since 2017, with an 80% drop in the last year alone. This sounds, at first, like good news for Europe. It is not. The collapse in inflows to the US reflects not a reduction in the attractiveness of American institutions but a collapse in the pipeline: fewer people are moving anywhere, as geopolitical uncertainty freezes mobility. The people who were going to go to America and might instead go to Europe are, on balance, staying in place.
What is genuinely interesting — and what opens a small window — is the Trump effect on scientific talent. The European Research Council has seen applications from US-based researchers nearly triple since 2024. The European Commission launched a €500 million "Choose Europe for Science" initiative. France created a new immigration status for "refugee scientists." These are real responses to real dislocations.
But Andrei is still in Cincinnati. And when I asked him whether he would come back, he was quiet for a moment, in the particular way of someone calculating.
"Maybe," he said, "if the rules change. Not the tax rules — the assumption rules. The assumption that I'm going to fail."
VIII. The Fragmentation Tax
There is a cost that does not appear on any balance sheet, because it is distributed invisibly across every transaction in the European digital economy. I call it the Fragmentation Tax.
It operates like this: a startup founder in Bucharest who wants to scale her product across Europe must incorporate in a jurisdiction that investors recognise (often Luxembourg, Ireland, or the Netherlands, not Romania). She must navigate twenty-seven different VAT regimes. She must adapt her legal contracts to twenty-seven sets of consumer protection laws. She must localise her product in multiple languages, each requiring not just translation but cultural calibration. She must build relationships with investors concentrated in London, Paris, Berlin, and Stockholm — cities that capture 65% of European venture capital volume, according to current data.
This is not competition. This is overhead. And it is overhead that an American founder in Columbus, Ohio simply does not pay. One language, one federal legal framework, one currency, a venture ecosystem that reaches into every mid-sized city.
The proposed EU Inc. framework — a pan-European corporate legal structure for young innovative firms — would address part of this. The Jacques Delors Centre analysis calls it essential for reducing "the transaction costs and risks that deter cross-border investments in the EU." But EU Inc. is, at the time of writing, still a proposal, still being negotiated, still subject to the Fragmentation Tax it seeks to eliminate.
There is another șopârlă here: the fragmentation is not an accident. It is, in part, the intended result of a European political system designed to preserve national prerogatives. The twenty-seven different frameworks are not bugs in the European project — they are features that national governments defend. The cost of that preservation is borne by the people who try to build things in Europe's digital economy. It is a hidden subsidy paid by entrepreneurs to politicians.
IX. What Enshittification Teaches Europe
I want to dwell on Doctorow's book for a moment longer, because it is doing more work in this essay than a brief citation would suggest.
Enshittification is, in part, a history of four lost disciplines — four structural constraints that once kept platforms from descending into pure extraction, and that were systematically removed over the course of three decades of regulatory and political choices. The book's strength is not merely diagnostic. It is also genealogical: it traces how the disciplines were removed, who benefited from removing them, and what it would actually take to restore them. This is uncommon in tech criticism, which tends either toward apocalyptic structuralism (everything is doomed because capitalism) or toward naive reformism (if we just passed a better privacy law, things would improve). Doctorow is neither. He is a mechanic who has looked at the broken engine and mapped the sequence of failures.
The relevance to Europe's position is direct and slightly uncomfortable.
Doctorow's framework contains a specific prescription that is worth applying directly to Europe's situation. He argues that enshittification is not inevitable — it is the result of specific policy choices, and it can be reversed with different choices. The four disciplines that once constrained platform decay (and that must be restored to reverse it) are: regulatory oversight with actual enforcement teeth; worker power and labour organising in tech; real competition, including antitrust enforcement that breaks up monopolies rather than fining them; and interoperability — the technical and legal right to build systems that talk to each other without the platform's permission.
What happens when you apply this framework to the EU's position vis-à-vis American and Chinese tech platforms?
The first discipline — regulatory oversight — is the one Europe has invested most heavily in. GDPR, the Digital Markets Act, the AI Act: these are serious instruments. But as Doctorow points out, regulation alone, when applied to platforms that are already dominant and that have captured the regulatory process through lobbying, tends to ossify the existing hierarchy rather than disturb it. Fining Google €4 billion does not create a European alternative to Google Search.
The second discipline — worker power — is complicated by the brain drain. European tech workers with genuine leverage over frontier AI systems are, increasingly, not in Europe. The workers who remain have less leverage because they are working on less critical systems.
The third discipline — competition — is where the EU has recently shown genuine ambition. The Digital Markets Act designation of "gatekeepers" — Google, Apple, Meta, Amazon, Microsoft, TikTok/ByteDance — creates a new category of regulatory intervention. Its enforcement track record is still being established.
The fourth discipline — interoperability — is the one most structurally underdeveloped in European policy, and the one most directly relevant to the sovereignty problem. CADA gestures toward it, with its cloud sovereignty tiers and procurement rules. But the deeper principle — that European institutions should have the technical right and practical ability to move their data and systems between providers without catastrophic switching costs — is not yet operationalised in law.
The honest synthesis of Doctorow's framework applied to Europe is this: Europe has built the most sophisticated regulatory architecture for digital platforms in the world, and it has done so while remaining structurally dependent on the platforms it is regulating. This is the equivalent of a tenant writing detailed rules about how the landlord must maintain the property, while the landlord holds the title deed and can sell the building whenever he chooses.
Sovereignty is not a clause in a contract. It is ownership of the building.
X. The Long Game and the Short Memory
Before arriving at the conclusion, I want to address a charge that will be levelled at this piece by readers with a more optimistic disposition: that it is too dark, that it understates genuine European strengths, that it projects the particular perspective of Botoșani — a city in the north-east of Romania that does not, by any metric, represent the frontier of European digital capability — onto a continent that includes ETH Zurich, INRIA, the Max Planck Institutes, Arm Holdings, ASML, and Deepmind's European research units.
The charge is partially fair, and I want to engage it honestly.
Europe does produce extraordinary research. Its academic institutions are among the best in the world in mathematics, cryptography, formal verification, robotics, and computer vision. The European Research Council funds work of genuine frontier quality. Mistral AI is technically credible and commercially alive. SAP, though struggling at the edge of a cloud transition it did not design, is one of the most complex software systems ever built, embedded in the operations of tens of thousands of enterprises worldwide.
And the brain drain narrative, while real, has a counter-current that 2025 and 2026 have made suddenly more legible. American research is under political pressure it has not experienced in living memory. The movement of US researchers toward European institutions — still small in absolute numbers, but meaningful in symbolism — suggests that the frontier is not as firmly fixed as it seemed three years ago.
What I resist is the inference that these genuine strengths constitute a systematic response to a structural deficit. A great university system does not, by itself, constitute digital sovereignty, any more than producing excellent mechanics constitutes a national automobile industry. The question is not whether Europeans are capable — they demonstrably are — but whether the system converts that capability into the infrastructure, the institutions, and the companies that constitute genuine independence. And there, the evidence is bleaker than the researchers' CVs would suggest.
XI. The Mechanic's Toolbox
My father had a toolbox that he built over thirty years. Every tool in it had a history: where he bought it, what job it had first been used on, what it had taught him about the material it cut or turned or measured. He could have told you the provenance of every wrench and every feeler gauge.
The digital economy has given Europe a toolbox full of tools it did not make and does not understand at the deepest level — at the level of the foundational models, the chip architectures, the operating systems. A mechanic who works only with tools he cannot open and cannot repair is not a mechanic. He is a user.
The Tech Sovereignty Package is, at its most generous reading, a plan to build some of the tools. AI Gigafactories are, at their core, European workshops for building European models on European compute. The Chips Act 2.0 is an attempt to relearn, over a decade, how to fabricate at the frontier. The CADA sovereignty tiers are an attempt to ensure that, at minimum, European public institutions know which building they are renting and who holds the keys.
This is not nothing. It is, in fact, more than Europe has previously been willing to commit to. The recognition that "digital dependence is a strategic vulnerability to be corrected through legislation" — that exact phrase, from the Commission's own framing — represents a genuine rupture from the market-fundamentalist assumptions that governed European digital policy for the previous fifteen years.
But there is a tension that the legislation does not resolve, and that the political class does not like to name. The tension is this: European digital sovereignty, genuinely pursued, will create friction with American trading partners at a moment when the transatlantic relationship is already under pressure. CADA's cloud sovereignty tiers, if enforced with real teeth, will disadvantage AWS, Azure, and Google in the European public sector market. The US government has already demonstrated, with its restrictions on frontier AI model access for foreign nationals, that technology is not a neutral commodity.
Europe is, slowly and painfully, arriving at a conclusion that China reached fifteen years ago: digital infrastructure is geopolitics. The tools you use to run your hospitals determine who you can call an ally.
Andrei, from his rented apartment in Cincinnati, would probably say this is obvious. It is. The question is what Europe intends to do with the obvious.
XI. A Note from the Author
This article began as a transcription impulse — I wanted to preserve something from that Sunday video call with Andrei before it faded into the ordinary. Cousins talk. Family calls meander. But occasionally they produce a clarity that three months of policy documents cannot.
The conversation gave me the šopârlă — the embedded argument, the lizard in the wall — that runs through this piece. It is this: Europe's digital problem is not primarily a technological problem. It is a problem of embedded assumptions about failure and success, about who deserves resources, about what "ambitious" is permitted to look like. The regulatory architecture, the investment gap, the brain drain — these are symptoms. The disease is the assumption, replicated across twenty-seven legal systems and three generations of civil servants, that innovation is something that happens elsewhere and that Europe's job is to regulate it responsibly after the fact.
Doctorow's enshittification framework arrived in my reading at exactly the right moment, because it offered a language for what I was observing. The platforms that Europe uses are in various stages of their enshittification cycles. The regulatory response to that enshittification has, in several cases, reinforced the incumbents rather than enabling their challengers. And the alternative — building European platforms, European models, European infrastructure — is precisely what the Fragmentation Tax and the assumption of failure have made so difficult.
I am aware of the limits of this article. It is written by a person based in Botoșani, which is not Berlin. It is written from the perspective of a web agency owner who deals primarily with Romanian public institutions, not with AI researchers or semiconductor engineers. The view from here is partial. The friction I see is smaller in scale than the friction that shapes global AI investment flows, but it is structurally similar — it is the same machine, operating at a different resolution.
Rhetorical tools used in this piece:
- The șopârlă (Aesopic embedded argument): Used twice explicitly — first, to argue that European regulation may primarily benefit American incumbents rather than European challengers; second, to argue that European political fragmentation is not an accident but a defended feature that transfers costs from politicians to entrepreneurs. In both cases, the argument is embedded within apparently neutral description, as is traditional in Aesopic discourse — the proposition that cannot be stated directly is shown through the observable structure of the thing being described.
- The personal interlocutor as analytical anchor: Andrei is not a rhetorical device. He is a real person and these are substantially real observations from a real conversation, though compressed and reordered for clarity. His presence in the article serves the classical function of the exemplum — the particular instance that gives the structural argument a human face.
- Negation as affirmation: Several sections argue their thesis by naming what is absent — no European models in the global top ten, no fabrication capacity, no unified corporate law — a classical rhetorical technique in which the shape of the absence proves the existence of the problem.
- The father figure: The mechanic's toolbox is not decoration. In my writing, the figure of my father recurs as a philosophical reference point — not sentimentally, but because his relationship to tools (physical, verifiable, repairable, owned) provides the sharpest possible contrast to the digital economy's relationship to infrastructure (rented, opaque, externally controlled). He is the standard against which the absurdity of the current arrangement can be measured.
Subjective errors I am aware of: I may have overweighted the "assumption of failure" as an explanation for Europe's tech gap, at the expense of more structural economic factors (cost of capital, market size, university funding models). The cousin's testimony is real but not representative — one engineer in Cincinnati is not a sample. And I have understated the genuine quality of European research universities, the ERC grant system, and the depth of European talent in specific fields such as cryptography, formal verification, and robotics.
Ten Questions Worth Remembering
1. What is the European Technological Sovereignty Package and what does it actually propose?
Presented by the European Commission on 3 June 2026, it is a set of legislative and strategic measures aimed at reducing Europe's dependence on non-European technology providers. Its two legislative instruments are the Cloud and AI Development Act (CADA) — which aims to triple EU data centre capacity within seven years, create AI Gigafactories, and introduce a cloud sovereignty framework for public procurement — and the Chips Act 2.0, which attempts to rebuild European semiconductor manufacturing capability. It also includes a non-binding Open Source Strategy and a roadmap for AI in energy. Adoption is not expected before the end of 2027.
2. What does "enshittification" mean, and why is it relevant to European digital policy?
Coined by Cory Doctorow, enshittification describes the predictable lifecycle of digital platforms: they are initially generous to users to attract them, degrade the user experience to serve business customers once users are locked in, and then degrade the experience for business customers too to extract maximum value for shareholders. It is relevant to European policy because Europe has built a sophisticated regulatory response to enshittification while remaining structurally dependent on the platforms experiencing it — regulating the behaviour of infrastructure it does not own.
3. How large is the actual investment gap between the EU and the United States in AI?
According to the Stanford AI Index 2026, private AI investment in the United States in 2025 was $285.9 billion. All of Europe attracted $20.9 billion. The ratio is approximately 14:1. This is substantially larger than most public discourse acknowledges.
4. What is the Scaleup Europe Fund and why does it exist?
It is a €5 billion fund — €1 billion public capital from the EIC, €4 billion private — managed by EQT, intended to provide direct equity investment in European technology scaleups. It exists because European innovative companies routinely struggle to find growth-stage capital at home and are forced either to seek US funding (accepting dilution and often corporate relocation) or to remain subscale. First investments are expected in autumn 2026.
5. What is the Fragmentation Tax?
An informal term for the aggregate cost imposed on European startups by the absence of a unified legal, regulatory, and market framework. To scale across Europe, a company must navigate twenty-seven legal systems, twenty-four languages, multiple VAT regimes, and a venture capital ecosystem concentrated in five cities. The equivalent cost simply does not exist for American companies scaling within the United States.
6. Why are EU providers losing market share in the European cloud market?
EU providers held approximately 29% of the European cloud market in 2017. By 2022, that share had fallen to roughly 15% and has remained there. AWS, Microsoft Azure, and Google Cloud together hold approximately 70%. The causes include massive scale advantages, global data centre networks, continuous investment in AI-optimised infrastructure, and — until recently — the absence of European policy that actively steered public sector procurement toward European alternatives.
7. What is "brain drain" and how does it affect European AI development?
Brain drain refers to the departure of highly educated professionals — in this context, AI researchers and engineers — from Europe to other jurisdictions, primarily the United States. Net tech talent inflows to Europe fell from 52,000 in 2022 to 26,000 in 2024. The causes include salary differentials (2-3x in comparable roles), access to frontier research environments, and the absence in Europe of the large-scale compute clusters required for frontier model development. The US political environment in 2025-26 created a partial counter-movement, with European institutions attracting some researchers displaced by federal funding cuts.
8. What is the Cloud and AI Development Act's cloud sovereignty framework?
CADA proposes a tiered sovereignty classification for cloud services used by the European public sector. The tiers are intended to specify the level of European control required over data processing, storage location, and ownership of the service provider, depending on the sensitivity of the data involved. Highly sensitive data — in banking, healthcare, and public administration — would be steered toward the highest tier, effectively excluding non-EU providers or requiring them to establish genuinely European subsidiaries with ring-fenced operations. This is its most legally novel and commercially contentious provision.
9. What did Doctorow identify as the four disciplines that constrain platform decay?
Regulatory oversight with real enforcement teeth; worker power through labour organising; genuine competition enforced through antitrust action; and interoperability — the technical and legal ability to build systems that interface with platforms without the platform's permission. His argument is that enshittification accelerates when all four of these disciplines weaken simultaneously, which he argues has happened across the American tech economy since the 1980s.
10. Is Europe's situation reversible, and over what timescale?
Yes, but not quickly, and not without political decisions that European governments have historically avoided making. The Tech Sovereignty Package represents a genuine shift in declared intent. Converting declared intent into operational infrastructure — data centres, trained engineers, viable companies, frontier AI models — requires sustained investment at a level Europe has not historically committed to, a unified regulatory environment it does not yet have, and an assumption of success that, at the level of individual entrepreneurs and researchers, it has not yet learned to project. The honest answer to "over what timescale?" is: a generation, if started now.
What Can You Do Tomorrow?
Five Actions Worth Taking
1. Audit your own infrastructure dependencies. If you run a business, a school, a public institution — spend one hour this week mapping which of your critical digital functions depend on non-European providers. Not to abandon them immediately, but to know. Dependency that is invisible cannot be managed. Make yours visible.
2. Support European alternatives, even when they are imperfect. OVHcloud, Hetzner, Infomaniak, Nextcloud, Scaleway — European cloud and software providers exist and they are technically competent. They are not AWS. They do not need to be AWS. Using them, even for non-critical workloads, builds the ecosystem that will be needed when the critical workloads must move.
3. Talk to the person about to leave. If you know an engineer, a researcher, a developer who is considering emigration — have the honest conversation. Not the patriotic one. The structural one. What specifically would have to change for the work to be viable here? Those answers, aggregated, are a policy agenda.
4. Engage the EU Inc. proposal. The European Commission's proposal for a unified corporate legal form for innovative startups is obscure and boring and consequential. Your local MP, your MEP, your business association — most of them do not know it exists. If you are a founder or an investor, learn what it proposes and make noise about it. The Fragmentation Tax will not be reduced by people who do not know it is being charged.
5. Read primary sources, not summaries. The Tech Sovereignty Package is publicly available. The CADA proposal is publicly available. The Stanford AI Index 2026 is publicly available and free. The informed citizen, the one who can evaluate policy proposals rather than consume reactions to them, is rarer and more necessary than any AI model.
What to Avoid
Do not conflate the map with the territory. A legislative proposal is not a law. A law is not an implemented regulation. An implemented regulation is not a changed economic reality. The distance between the Commission's press release and the actual tripling of European data centre capacity is measured in years, political cycles, budget negotiations, and the lobbying power of companies with more resources than some member states.
Do not mistake regulation for sovereignty. Fining American platforms does not make European alternatives. Do not celebrate DMA enforcement as a structural victory when it is, at best, a constraint on the rate of extraction. The fundamental question — who owns the infrastructure — is not answered by fines.
Do not assume the brain drain is reversible through salaries alone. Andrei would not come back to Romania for a salary increase without structural changes to the environment in which the work is done. The assumption of failure is not a tax that can be rebated. It must be rebuilt, at the level of institutions, over time.
Do not outsource your critical thinking to the platforms that have captured your attention. The irony of analysing digital dependency while distributing this analysis through platforms owned by the companies under analysis is not lost on the author. But there is a difference between using a tool and being captured by it. Use the tool. Question the frame the tool imposes.
Do not treat this essay as a verdict. I have made arguments here that I believe are well-grounded, but I have also made choices — about which data to foreground, which stories to tell, whose voice to amplify — that reflect my position and my limitations. The mechanic's son from Botoșani looking at the EU's AI policy is not the same as the Brussels lobbyist, the Parisian venture capitalist, or the Munich engineer. All of those readings are partial. Read widely, triangulate aggressively, and be suspicious of any framework — including this one — that makes the complexity appear simpler than it is.
Do not wait for Brussels. The Tech Sovereignty Package will not be finalised before late 2027. The investment gap is $265 billion wide right now. The engineers are leaving this year. Whatever can be done locally, institutionally, at the level of individual choice and individual institution — do not postpone it for the regulation that will arrive after the decisions have already been made.
Bibliography and Sources
-
Stanford University Institute for Human-Centered AI. AI Index Report 2026. Stanford HAI, April 2026. https://hai.stanford.edu/ai-index/2026-ai-index-report
-
European Commission. Proposal for the Cloud and AI Development Act (CADA), COM(2026) 502. 3 June 2026. https://digital-strategy.ec.europa.eu/en/library/proposal-cloud-and-ai-development-act-cada
-
European Commission. European Technological Sovereignty Package. 3 June 2026. https://digital-strategy.ec.europa.eu/en/policies/eu-tech-sovereignty
-
European Parliament Research Service. European Software and Cyber Dependencies. 2025. https://www.europarl.europa.eu/RegData/etudes/ATAG/2025/780413/ECTI_ATA(2025)780413_EN.pdf
-
Doctorow, Cory. Enshittification: Why Everything Suddenly Got Worse and What To Do About It. Verso Books, October 2025.
-
Synergy Research Group. European Cloud Providers' Local Market Share Now Holds Steady at 15%. 2025. https://www.srgresearch.com/articles/european-cloud-providers-local-market-share-now-holds-steady-at-15
-
Interface (Research Organisation). AI Talent in Europe: Inflows, Outflows and the Transatlantic Gap. 2024. (Reported in Euronews, 29 January 2026.)
-
European Innovation Council. Scaleup Europe Fund. 2026. https://eic.ec.europa.eu/eic-fund/scaleup-europe-fund_en
-
Jacques Delors Centre. One Fund to Scale Them All? Filling Crucial Financing Gaps with the Scaleup Europe Fund. 2026. https://www.delorscentre.eu/en/publications/detail/publication/one-fund-to-scale-them-all
-
Bird & Bird. Tech Sovereignty Package: An Overview. 23 June 2026. https://www.twobirds.com/en/insights/2026/tech-sovereignty-package-an-overview
-
Linklaters. The European Commission's Tech Sovereignty Proposal and the Move to Digital Deglobalisation. 22 June 2026. https://www.linklaters.com/insights/blogs/digilinks/2026/june/the-european-commissions-tech-sovereignty-proposal-and-the-move-to-digital-deglobalisation
-
European Parliament. Massive Brain Drain from EU Universities is an Existential Threat to Europe's Future, Parliamentary Question E-001033/2025. https://www.europarl.europa.eu/doceo/document/E-10-2025-001033_EN.html
-
Euronews. The AI Brain Drain: Why Europe Can't Keep the Talent It Trains. 29 January 2026. https://www.euronews.com/my-europe/2026/01/29/the-ai-brain-drain-why-europe-cant-keep-the-talent-it-trains
-
Regulations.AI. EU Cloud and AI Development Act (CADA): Analysis and Compliance. June 2026. https://regulations.ai/regulations/RAI-EU-NA-RE2026X-2026
-
Raconteur. EU Cloud and AI Development Act: What Sovereign Cloud Means in Law. 9 June 2026. https://www.raconteur.net/global-business/eu-cloud-and-ai-development-act-what-sovereign-cloud-means-in-new-laws
Petru Cojocaru este fondatorul ABSOLUT WEB EXPERT SRL și publică eseuri de neonficțiune literară în română și engleză britanică la petrucojocaru.ro și cojocarupetru.info. Trăiește și lucrează în Botoșani, România.
This article was researched and written with AI assistance. All factual claims have been verified against primary sources. The conversation with "Andrei" reflects a real exchange, substantially reconstructed from notes, with identifying details altered at the subject's request.