Skip to main content

Spain ranks 7th globally on AI — Brussels says the hard part starts now

Ilustrační obrázek
Barcelona's Forward Tech Summit put a familiar European tension on stage: the continent has written the most detailed AI rulebook in the world, and now it has to show that companies can still move quickly inside it. Daniel Calleja, the European Commission's representative in Spain, argued that the answer is investment — in data centres, connectivity and people — rather than looser rules. And the rules themselves stopped being voluntary in August.

The summit, organised by the Catalan employers' federation Foment del Treball, brought EU officials, local government and industry together in a city that has quietly become one of southern Europe's busiest technology hubs. According to Diari ARA's report from the event, Calleja called for faster adoption of artificial intelligence across the Spanish economy and for more spending on the unglamorous parts of it: data centres, connectivity and the infrastructure that makes AI services usable at scale.

What Calleja actually said — and what he didn't

The Commission's representative did not present European regulation as a brake. He presented it as the framework that makes investment bankable: common rules, shared risk management, one market. That is the same argument Brussels has been making since the Draghi and Letta reports painted a sober picture of European competitiveness, and it is the argument that European businesses keep testing against reality.

The political subtext is hard to miss. Europe is not short of AI users; it is short of AI infrastructure it controls. Most of the models a Spanish hospital, law firm or design studio reaches for are trained and hosted outside the EU, and every new data centre on European soil is partly an argument about who sets the terms.

The rules stopped being voluntary on 2 August

This is the part of the story that changes everyday practice. The voluntary codes of practice and the enforcement grace period for general-purpose AI providers are gone. Since 2 August 2026, national competent authorities and the EU AI Office can enforce the AI Act's obligations directly, and the Article 50 transparency duties apply: if people are talking to an AI system, or looking at synthetic images, audio or video, they have to be told.

In plain language: the question for a European company is no longer "will there be rules?" It is "can I show what my system does, where the data goes and who is accountable?" The European Commission's own AI Act pages remain the reference point for what applies when.

Spain's scoreboard: seventh in the world, average at home

Spain has a better story to tell than its self-deprecating tech discourse suggests. Stanford University's Global AI Vibrancy Tool currently places the country 7th globally — ahead of several economies with far larger research budgets. Two of the EU's AI Factories are hosted on Spanish soil, at the Barcelona Supercomputing Center (BSC) and the Galicia Supercomputing Center (CESGA), giving researchers and startups access to serious compute inside the EU.

The awkward number is adoption. Spain's overall uptake of AI sits at or just slightly above the European average — respectable, but not what you would expect from a country ranked seventh in the world for AI activity. That gap between capability and daily use is exactly what Calleja was pointing at. The Spanish government's own AI Atlas, published in September 2026, tries to map the same terrain: what exists, what is funded, what is still missing.

The two objections nobody in Brussels can wave away

Barcelona's deputy mayor, Jordi Valls, backed adoption "with European values" while naming two concerns that will not disappear. The first is dependency: building a European AI stack on top of a handful of American corporate oligopolies is a strategic risk, not just a procurement choice. The second is energy and water — data centres are physical installations, and Catalonia has already had droughts sharp enough to make cooling water a political subject. Larger EU data centres are already required to report their energy performance under European efficiency rules; those reports are about to become considerably more interesting.

What an AI bill actually looks like for a Spanish SME

Speeches about infrastructure are abstract. Invoices are not. Take a mid-sized Spanish company that pushes 50 million input tokens and 10 million output tokens per month — plausible for a customer-support and document-processing setup. Using published list prices per million tokens, the monthly bill splits like this:

ModelInput / 1MOutput / 1MMonthly (USD)
DeepSeek-V4.1-Flash (off-peak)$0.15$0.60$13.50
Mistral Large 3$0.50$1.50$40.00
Meta Muse Spark 1.3$1.25$4.25$105.00
GLM-5.3$1.40$4.40$114.00
Grok 4.6$2.00$6.00$160.00
GPT-6 Astra$10.00$50.00$1,000.00
Claude Fable 5.1$10.00$50.00$1,000.00

That is a spread of roughly 74× between the cheapest and the most expensive option for identical token volumes. Top-tier output pricing has fallen from around $150 per million tokens in 2025 to $50 today — still three times the European open-weight alternatives. And these are pre-tax list prices in dollars: a Spanish buyer pays domestic VAT (21% IVA) on top, and if you need EU data residency or a signed data-processing agreement, the cheap US-hosted tier may simply be off the table.

This is where the Barcelona argument becomes concrete. Mistral Large 3 is French, priced at the low end, and deployable on EU infrastructure. Two Spanish AI Factories exist for workloads that need sovereign compute. The infrastructure question Calleja raised is not an abstraction about cables — it is the reason a company in Valencia can pick a European model without apologising for it.

The investment question behind the speeches

None of this settles the harder trade-off. Europe can fund data centres, enforce transparency and still lose the productivity argument if small firms never get past the pilot stage. The honest read of the Forward Tech Summit is that Brussels has finished legislating and started counting: energy, connectivity, adoption rates, cost per token. Those are the numbers that will tell us in two years whether the European approach worked — and unlike a summit speech, they cannot be phrased carefully.

Do the AI Act's transparency rules apply to a small European business, not just to OpenAI or Google?

They can. Article 50 duties target the people deploying AI systems as well as the providers. If your chatbot talks to customers, or your marketing uses generated images or voice, users generally need to be informed and synthetic content generally needs to be marked. Since 2 August 2026 these obligations are enforceable by national authorities, not just guidance on paper.

Is a cheaper non-European model actually cheaper once compliance is included?

Not always. Token prices are only one line item. Add VAT, the legal work around data transfers, and possibly a European hosting tier, and a model listing at $0.50 per million input tokens can end up competitive with a $0.15 one — before you count the risk of having to migrate later if the legal situation changes.

Where can a Spanish company get EU-based AI compute today?

Two of the EU's AI Factories are hosted in Spain, at the Barcelona Supercomputing Center and the Galicia Supercomputing Center, and they exist specifically to give researchers and businesses access to European compute. Commercial EU-hosted inference from providers such as Mistral is the more common route for ordinary business workloads.

Discussion

No comments yet — be the first to share your thoughts.
X

Don't miss out!

Subscribe for the latest news and updates.