Portugal's AMALIA AI Model Treats Foundation Tech as Public Good

Portugal has launched an open-source artificial intelligence foundation model that frames AI development as a public infrastructure project rather than a commercial venture, according to analysis published by the Peterson Institute for International Economics.

On July 1, Portugal introduced AMALIA (Automatic Multimodal Language Assistant with Artificial Intelligence), a 9-billion-parameter model developed by a consortium led by NOVA University Lisbon. The project received €5.5 million in public funding, with an additional €1.5 million committed through 2027, as detailed by PIIE senior fellow Monica de Bolle.

Unlike consumer-facing chatbots, AMALIA serves as foundational technology that public institutions, universities, companies, and researchers can adapt for specific applications. The model is open source and complies with EU data usage and privacy regulations.

Why it matters

The Portuguese approach represents a fundamentally different philosophy for governing critical AI infrastructure. By treating a foundation model as a public utility—comparable to water or electricity systems—Portugal has secured national autonomy over technology that will increasingly underpin economic activity. This contrasts with the US model, where the most capable systems remain proprietary, access is sold by subscription, and terms can change at corporate discretion.

The public good economics

AMALIA meets the technical definition of a public good: one person's use doesn't diminish availability for others, and no one can easily be excluded. A hospital fine-tuning AMALIA for clinical needs doesn't prevent a university from using it for education or a court from applying it to legal proceedings.

This creates a market failure. Private developers won't build such systems because they can't capture enough of the distributed benefits. Portuguese, the world's fifth most-spoken language with 260 million speakers, accounts for only 2 percent of web data in common training sets. Where private Brazilian developers have addressed Portuguese, they've produced closed commercial products.

Inverting the American model

The United States historically seeded general-purpose technologies through public investment—the internet began as a Pentagon project, GPS remains government-operated, and basic computing science flowed from federally funded labs. These kept foundational layers open for private sector innovation.

With AI, that sequence reversed. The foundational layer is privately owned, and public institutions queue as customers. Meanwhile, data centers receive tax abatements, subsidized land, and priority utility access, with grid upgrades partially financed by ratepayers who also bear noise pollution and higher electricity costs. The inputs are socialized while outputs remain proprietary.

Risks and the real test

Public provision carries risks. AMALIA might age as the frontier advances, future budgets might not materialize, and users might default to other models rather than contribute improvements. The critical measure over the next two years is whether anyone actually builds on this foundation.

For roughly €7 million—less than a highway interchange—Portugal has secured a public stake in technology its economy will increasingly depend on. The country has chosen to invest in autonomy rather than rent from foreign corporations on their terms.

Details of Portugal's AMALIA project were first reported and analyzed by Monica de Bolle at the Peterson Institute for International Economics.