Amalia: Portugal releases an open-source AI model for sovereignty
Amalia's July 1, 2026 launch is a concrete sovereign-AI signal: Portugal is moving from a research project to an open national building block focused on European Portuguese, reusable by public administration, universities, companies, and researchers.
1. What changes with the launch
Reuters reported on July 1 that Portugal had launched Amalia as its first national open-source AI model, with the training dataset and source code released under an open license. Portuguese coverage adds that the launched version starts around 9 billion parameters, with a stated path toward 22 billion and public funding raised to EUR 7 million through 2027.
The official project website frames Amalia as open-source and public AI. Its objectives are explicit: promoting European Portuguese, preserving cultural representation, supporting sovereignty over citizens' data, and stimulating national AI research.
2. Why this is a real sovereign-AI signal
Sovereignty is not only about building data centers or localizing data. It also requires models, benchmarks, datasets, and adaptation capacity that reflect a territory's language, use cases, and constraints. Amalia directly addresses that point by making European Portuguese better covered through an open, auditable, and improvable foundation.
The AMALIA technical report highlights how underrepresented European Portuguese remains in existing large models and releases native evaluations for that linguistic variant. The European signal is important: each country that structures an open model layer reduces dependence on generic models and improves its room for maneuver on sensitive public uses.
3. Practical reading for AI Belgium, AI France, and Odoo Enterprise
For AI Belgium, AI France, Odoo Belgium, Odoo France, and Odoo Enterprise programs, Amalia confirms a key point: a sovereign AI strategy must combine infrastructure, model, data, governance, and use cases. The first domains cited by the project - education, culture, media, science, and public services - are close to the needs of regulated organizations that want to industrialize AI without losing control.
The right decision is not to copy Amalia as-is, but to copy the method: define critical languages and corpora, choose licensing and hosting compatible with internal constraints, then test AI assistants on traceable business workflows before scaling them.
Identify the linguistic and business datasets that must remain locally controlled before industrializing AI assistants across ERP, support, HR, and public-service workflows.
Frame a sovereign trajectoryRead the official Amalia website