Building or Buying AI? What the TDM Debate Means for Europe
As Europe approaches the 2026 review of the Copyright Directive, we are making a choice that goes far beyond copyright.
We are deciding whether Europe will be a builder in the AI century—or merely a client.
Whether the next unicorn is built in Berlin, or rented from Boston or Beijing.
That was the core message I brought to the SME Connect discussion on “TDM Reform and Creativity – What Would It Mean for Europe’s AI Ecosystem?” It is also a message that has suddenly become very concrete. A new legislative proposal in France shows exactly what is at stake—and how quickly well‑intentioned regulation can undermine Europe’s technological sovereignty.
If we break the current Text and Data Mining (TDM) framework, we do not just tweak copyright law. We legislate away Europe’s ability to build competitive AI.
From Brussels to Paris: When Copyright Law Tests Europe’s AI Future
A trans‑partisan group of French senators has recently introduced the “IA et création” bill. Its stated goal—protecting authors and creators—is legitimate and important. But the mechanism proposed, a so‑called “presumption of exploitation,” should set off alarm bells across Europe.
Under this approach, AI developers would be legally presumed to have infringed copyright unless they can prove otherwise. In practice, this reverses the burden of proof and creates a guilty‑until‑proven‑innocent framework for AI training.
For anyone who understands how AI models are trained, the implications are obvious: proving a negative across billions of data points is not a compliance exercise—it is an operational impossibility. Compared to competitors in the US or Asia, European startups would face crushing legal friction before they even write a line of code.
This should not be looked at as a French curiosity. It is a preview of what happens if Europe abandons the logic underpinning Article 4 of the Copyright Directive.
TDM Is the Floor, Not a Loophole
There is a persistent narrative in Brussels—and now in national capitals—that the current TDM system is too permissive, and that Europe should move from today’s opt‑out model to some form of opt‑in or presumed infringement.
That would be a fatal mistake.
The opt‑out mechanism in Article 4 is not a loophole. It is the de minimis floor for innovation. Training a meaningful AI system requires analysing vast, fragmented corpora at scale. Today’s framework allows developers to mine data unless rights are explicitly reserved in a machine‑readable way. That balance works.
Replace it with permission‑first logic—or worse, a reversed burden of proof—and you do not rebalance copyright. You create prohibitive transaction costs that only the largest incumbents can survive.
Imagine a small European startup building an AI system to optimise renewable energy grids. They need decades of meteorological data, technical journals, and infrastructure logs—millions of documents from thousands of sources, many defunct or untraceable.
Under opt‑out, they innovate.
Under presumed infringement, they drown in legal uncertainty.
Who survives that world? Not European SMEs. Big Tech does.
When Europe’s AI Champion Sounds the Alarm Bells
The reaction to the French proposal from Mistral AI, Europe’s flagship foundation-model company, has been swift and refreshingly straightforward.
As reported by Contexte, Mistral warns that the bill would “irreversibly compromise Europe’s ability to remain in the global AI race, by calling into question the very conditions under which AI models can be trained and deployed.”
The company points to the €2 billion it has just invested in France to train its models—real infrastructure, real jobs, real sovereignty—and cautions that the proposed law risks “reducing these efforts to nothing.”
This is a direct warning about the viability of Europe’s AI industrial base.
When the CEO of Europe’s most prominent AI company personally engages with parliamentarians to explain that a law threatens the foundations of sovereign technology, policymakers should listen carefully. Sovereignty is not achieved through declarations of intent; it is built—or dismantled—through the legal conditions that determine whether AI infrastructure can exist in the first place.
The Wrong Benchmark
When alternatives are discussed, the comparison often defaults to the United States. That is misleading.
The US relies on fair use: flexible in theory, chaotic in practice. A growing wave of AI‑copyright litigation is creating a fog where only companies with massive legal budgets can operate. That is not an SME‑friendly environment.
If Europe wants inspiration, it should look to Japan and Singapore.
Japan’s concept of non‑enjoyment cleanly separates consumption from analysis. Training models is treated as production, not reading for pleasure, and is permitted regardless of commercial intent.
Singapore’s Copyright Act goes further by preventing contracts from overriding its computational data analysis exception. Lawful innovation cannot be quietly blocked by terms of service.
This is what statutory certainty looks like—and it is no coincidence these jurisdictions are pulling ahead.
Innovation Arbitrage Is Happening Now
Capital is mobile. Talent is mobile. Regulation is not neutral.
If Europe becomes a permission‑first—or presumption‑of‑guilt—jurisdiction, while Asia offers clarity and the US offers capital, we are effectively taxing our own innovation.
A venture capitalist deciding where to invest €10 million understands this instantly:
- In Paris, a startup must reserve a large share of funding for legal defence and compliance.
- In Tokyo, that same money goes to compute, engineers, and products.
The result is predictable. Europe loses investment, IP, talent—and eventually tax revenue.
When Cultural Pressure Writes Industrial Policy
One of the deepest problems in the TDM debate is who is at the table.
The loudest voices belong to Big Tech and Big Content: Silicon Valley giants negotiating with major publishing, music, and film studios. Their conflict is about licensing revenue.
Missing are SME founders, researchers, and cultural heritage institutions.
The result is a one‑size‑fits‑all approach where mining medical journals for cancer research is treated like streaming a blockbuster movie.
That is not balance. It is policy distortion.
Protecting cultural creation is essential. But if Europe protects heritage by preventing innovation, the future of both is lost.
The Forgotten Majority: SMEs as AI Users
This debate is not only about who trains models. It is also about the millions of European SMEs who simply want to use AI tools to remain competitive.
Fragmented or overly restrictive copyright rules create three immediate risks:
- Liability traps, where SMEs cannot assess whether using global AI tools exposes them to claims.
- “Lobotomised AI”, where Europe receives geoblocked or watered‑down models trained on limited datasets.
- Supply‑chain instability, where essential AI services can disappear overnight due to injunctions.
You cannot build a business workflow on an API that might be illegal next quarter.
A Choice Europe Cannot Defer
The French debate is a warning shot, not an outlier. It shows how quickly good intentions can translate into rules that hollow out Europe’s AI ambitions.
The 2026 review of the Copyright Directive is a decisive moment. Europe must resist letting a conflict between Big Tech and Big Content dictate its industrial future.
If we want AI to be built in Europe—not merely consumed here—three principles are non‑negotiable:
- Defend opt‑out as the absolute baseline for TDM.
- Shift policy attention from training inputs to harmful outputs.
- Choose statutory certainty over litigation and presumption of guilt.
Europe has the talent, the capital, and the ambition to lead.
Let us ensure we do not also have the regulation to fail.
Get the article’s core concepts in a unique auditory summary—listen to the AI-generated track below!
This song was created entirely using artificial intelligence tools. If you enjoy this experiment, keep an eye out for each song illustrating our December articles.
Written byCaroline De Cock, LL.M., Head of Research
