Flagship project 14 of the Plan IA360, presented on 21 September 2026, provides for launching the Alianza Iberoamericana de la Inteligencia Artificial ("Ibero-American Artificial Intelligence Alliance") on 3 November 2026, at the Foro Digital Iberoamericano ("Ibero-American Digital Forum"). As of today it doesn't exist as an operational structure: the plan doesn't list countries or define its governance. The three decisions that follow don't depend on that date: they can be taken beforehand, and it's worth doing so.
The plan sums up its commercial logic in a phrase worth taking seriously: Spain will project its regulatory leadership «de modo que cumplir antes permita vender antes» ("so that complying earlier allows selling earlier"). The alliance is built on three pieces — an Ibero-American data and interoperability space, models «en español, portugués y lenguas originarias, con ALIA como base» ("in Spanish, Portuguese and indigenous languages, with ALIA as the base"), and shared computing — but none of the three settles, on its own, the concrete decisions already facing any business that today sells or wants to sell an AI product or service in Latin America. Those decisions are three, and they don't wait for November.
First decision: which model, not just which provider
The first question isn't which AI provider do I contract, it's what type of model does this use case need. A large generalist model isn't always the best option for serving customers in Spanish: sometimes a smaller one, cheaper to run, trained or fine-tuned with a focus on Spanish, is the better fit, rather than a generic one trained mostly in English and adapted afterwards. ALIA is, according to the plan itself, «una de las primeras familias de modelos abiertos impulsadas por un Estado» ("one of the first families of open models backed by a state") and the first trained in Spanish and in Spain's co-official languages; the alliance will build on that base to also reach Portuguese and the region's indigenous languages, something the plan announces but doesn't yet describe in detail.
Choosing the right model for each use case — budget, target latency and data residency included — is a technical decision worth taking case by case, without depending on which specific provider sells the model. Prioritising which use cases to tackle first, by expected return and risk, and ordering them into an 18-month roadmap, is what a maturity diagnosis and roadmap does, not a generic catalogue comparison.
Second decision: Spanish isn't a single language
The plan talks about «modelos en español» ("models in Spanish"), in the singular, as if it were a single variant. It isn't. An assistant trained on peninsular Spanish can sound forced or outright confusing to a customer in Mexico, Argentina or Colombia: everyday vocabulary changes, the expected level of formality changes, even expressions that are neutral in one country sound wrong in another. Selling with AI across several markets in the region means checking, country by country, whether the assistant's tone and vocabulary fit whoever is going to read or hear it, not assuming "Spanish" is a box you tick once.
Third decision: where the customer's data lives
The third decision is the one least thought about until it's already too late: where a customer's data in another country is stored and processed when the conversation passes through an AI system. When the assistant answers by drawing on the business's own documentation — catalogue, terms, order history — it should do so by citing the exact source, something that depends on the system's technical design, not on good intentions. This is where Summum IA's RAG and internal AI search comes in: an index over the business's own documents, with the source system's permissions carried over and storage chosen according to the customer's residency requirements (EU or the business's own infrastructure).
Three questions before selling AI outside Spain
- Is the chosen model tested with the Spanish variant of the country you're selling into, or only with Spain's?
- Is a customer's data in another country stored separately from anyone else's, with clear rules on who can query it?
- Is there already a use case proven in one market before replicating it in another, or is it being assumed that what works in Spain will work the same way abroad with no adjustment?
Frequently asked questions
Do I have to wait for the Alianza Iberoamericana to sell with AI in Latin America?
No. The three decisions on this page — model, language variant and data residency — can be settled today, regardless of whether the alliance ends up materialising on the terms the plan announces.
Is it compulsory to use ALIA to benefit from the alliance?
The plan doesn't say. ALIA is the declared base the alliance's models will be built on, but there's no obligation to use it today to operate in the region.
Does the AI model change depending on the country you're selling into?
Not necessarily the model itself, but it should be adjusted or tested with each market's Spanish variant, because a tone that's right in one country can sound wrong in another.
What happens with data protection if my customer is in another country in the region?
It depends on each country's regulatory framework, which isn't the same as the European one. What does depend on the system's technical design is that one customer's data doesn't get mixed with another's, something that gets resolved in how the AI system is built, not after an incident.
Flagship project 14, Alliances area, of the Plan IA360 (official La Moncloa PDF). The rest of the Plan IA360, including the plan's full international outreach and the other projects in the Alliances area, is in the complete guide to the Plan IA360.
Content last verified: 22 September 2026.