Rent the frontier or build your own.

I keep reading the same article, now in two genres. One is the creator's lament: Europe has been wronged again, an exception has been abused, and a coalition of authors has discovered, five years too late, that the 2019 rules were a license to scrape whatever they could. The other is the user's lament: the best models are walled off, access can be withdrawn from anyone outside the United States, and a handful of labs quietly decide which model handles your request, how much it is allowed to think, and whether you even get the good version.
Different subjects, identical posture.

Someone has done something to us.
Someone else should fix it.

I have spent twenty-five years in music infrastructure, and I know this register. It is the sound of people who mistake being right about a dependency for actually doing something about it.
Both laments are right.

The text and data mining exception became a quiet pipeline for ingesting decades of human work without a trace.
An opt-out you cannot see is not a protection.

A transparency template that discloses the top 10% of domains, with none of the titles, is a concession dressed as a safeguard. And yes, when your intelligence lives hundreds of miles away in someone else's building, that someone can throttle you, downgrade you, retain your data for their own purposes, and lock you out by geography, all without telling you.
Being right about all of it changes nothing on its own.

Dependency is the actual problem

The frontier model debate and the provenance debate are the same debate. In both, the thing that matters most is held by someone else on terms you do not control, and the response has been to complain about those terms rather than to reduce the dependency.

The honest assessment of the computing situation makes this unavoidable. There is not enough data center capacity to serve the world's appetite for AI on equal terms. The announced buildout is running behind schedule, and the economics push the labs toward squeezing margins precisely where the customer cannot see it. A rational person, given equivalent performance, keeps their requests and their data on their own device every time. Not out of paranoia, but because anything that leaves the machine can be priced, prioritized, downgraded, inspected, or stolen. The incentive to own the capability locally is structural and only getting stronger.

This is why the quieter technical story matters more than the weekly outrage. Small models are getting good, and diffusion methods are making them faster by generating in parallel rather than one word at a time, which addresses the memory bottleneck that slows local inference today. Most of what people actually use frontier models for does not require frontier intelligence or frontier prices. Within a year or two, capable models will run on the hardware already sitting in billions of laps and pockets. The point is not the architecture. The point is the direction: control moving back toward the person who owns the device, away from the entity that owns the building.

The reflex that has to go

Europe does not lack intelligence, talent, or legitimate complaints. It lacks the instinct to build before conditions are perfect, because conditions are never perfect, and the people who win do not wait for them.

The traditional reflex is to form a coalition, consult stakeholders, draft principles, and balance interests until the moment has passed. In stable sectors, this protects people, and that is real value.

Applied to a field moving at the speed of capital and computation, the same reflex produces intelligent documents and foreign dependency. Beneath the procedural problem lies a cultural one worth naming plainly.

Across much of Europe, visible ambition draws suspicion before it draws support. You cannot ask founders to compete with trillion-dollar ecosystems and, at the same time, expect them to behave like cautious grant applicants, apologizing for thinking globally.

Humility is useful.
The version of humility that is really fear of being seen to try is not.

Build before you are granted permission

The work I do is in provenance infrastructure, and it exists for the same reason edge AI does: a capability you depend on but do not control is a vulnerability, and the answer is to own it rather than to petition the people who currently hold it. The creator economy's direct problem is provenance, a verifiable, auditable record of what existed, who made it, and when. That record does not appear because a committee wishes for it. It gets built by people who decide to close the gap rather than describe it, usually before anyone has signed off on the attempt.

The practical answer to both sad articles is the same. Stop renting your future and then complaining about the landlord. Move the capability you actually need closer to yourself, whether that is intelligence running on your own hardware or a record of authorship you can prove without asking anyone's permission. Give the legitimate demands of authors, labels, and users the one thing that turns a grievance into a position, namely, ownership of the thing in dispute.

The talent is here. The market is here. What has been missing is the willingness to act before someone else decides you are allowed to.
Less renting. More building.

The complaint will still be valid afterward, and this time you will be standing on something you own.