Portugal’s Amália AI: €5.5M open-source LLM built on 2 supercomputers

Amália isn't just another LLM—it’s a national AI asset engineered from the ground up on Portuguese high-performance computing infrastructure. The model leverages the Deucalion supercomputer and MareNostrum 5, giving it the scale needed to rival commercial offerings while keeping data entirely within the EU. For AI researchers and developers, the fully open-source release, including training data, means complete transparency and the ability to fine-tune for specific domains—from maritime decision support to cultural heritage applications. This launch marks a technical milestone in the continent’s push for an independent, transparent AI stack.

Portugal took a significant step toward digital sovereignty on July 1, 2026, unveiling Amália, its first open-source large language model (LLM). Developed by a consortium of Portuguese universities and research institutions with €5.5 million ($6.26 million) in EU recovery funds, Amália positions Portugal squarely inside a broader European movement to reduce dependence on U.S. artificial intelligence providers. The launch mirrors earlier national initiatives in France, where the government backed Mistral AI, and in Germany, the home of Aleph Alpha—both created to offer alternatives to models from OpenAI, Google, and Anthropic. Unlike consumer-facing chatbots, Amália is a foundation model designed as a base layer that public institutions, companies, and researchers can fine-tune and deploy for specialized applications. Its code, training dataset, and weights are all released under an open-source license, lowering barriers for uptake across government, banking, insurance, telecommunications, and heavy industry. Prime Minister Luís Montenegro framed the project in strategic terms: “Europe’s strategic autonomy is today, perhaps more than ever, tied to AI. This model will enable us to face the coming decades with greater sovereignty and less dependence.”

“

Developed by a consortium of Portuguese universities and research institutions with €5.5 million ($6.26 million) in EU recovery funds, Amália positions Portugal squarely inside a broader European movement to reduce dependence on U.S.

The model’s initial use cases underscore its utility. A virtual museum guide will enhance cultural tourism, the Portuguese Navy will test decision-support tools, an AI teaching assistant will help teachers plan lessons, and a digital assistant will streamline public service delivery. Each application leverages Amália’s ability to understand Portuguese language and local context—something that generic U.S. models often struggle with. The model run on the Deucalion and MareNostrum 5 supercomputers, national high-performance computing assets that guarantee the necessary scale for training and inference without renting cloud capacity from American hyperscalers. This infrastructure gives Portugal a degree of technological autonomy that many smaller nations lack.

The economic implications are multilayered. For the Portuguese public sector, Amália can accelerate digitization of services, potentially saving costs and improving citizen experience. For domestic businesses, particularly startups and small enterprises, having a free, sovereign LLM removes the need to pay for API access or worry about data leaving the jurisdiction. The banking and insurance sectors, explicitly mentioned by the prime minister, may develop custom models for risk assessment or fraud detection while meeting strict GDPR requirements. By open-sourcing the full stack, Portugal hopes to attract a community of developers and companies that can build a local AI ecosystem, much like the one that grew around France’s Mistral.