The global competition for supremacy in artificial intelligence (AI) is dominated by the major technology behemoths in the United States vs. their rivals in authoritarian China. But there is a third path forward being pursued by countries in Europe that view AI as a public, not a private good. One of the most interesting developments in Europe occurred on July 1, when Portugal launched AMALIA (Automatic Multimodal Language Assistant with Artificial Intelligence). Its goal, like the French company Mistral AI SAS, is best understood as a means of achieving some independence from the terms set by a handful of US companies while preserving the local language and culture.
AMALIA is named after Amália Rodrigues, a famous Fado singer and a symbol of Portuguese culture. Its success could pave the way for other countries to design their own strategies to achieve control over the artificial intelligence that their economies may ultimately depend upon. The Portuguese project was developed by a consortium of researchers coordinated by NOVA University Lisbon and was financed with public funds—€5.5 million, with a further €1.5 million committed through 2027.
Importantly, the model is open source, and its training conforms with EU legislation governing data usage and privacy. Notably, AMALIA is not intended for direct use by the public but rather to serve as a foundational technology that public institutions, universities, companies, and researchers can use to build tailored AI applications for their specific needs.
The obvious question is: Why spend any money on AI capabilities when other models already exist, already handle Portuguese in some capacity, and can be accessed at a fraction of the cost of building a foundation model from scratch? After all, AMALIA, at 9 billion parameters, is incredibly small compared to the systems behind ChatGPT of Open AI and Claude, the tool developed by Anthropic.
If the focus of the project was capability alone, one would be tempted to call it a failure. That, however, is the wrong way to look at AMALIA. What Portugal is doing is investing in its autonomy over an input that is rapidly becoming as essential as any public utility, such as water or electricity, which, incidentally, are both critical to LLMs themselves. In this respect, Portugal is treating AI as it should be treated in the United States: a public good.
The public good framing is not a metaphor. A public good is one whose use by one party does not diminish its availability to others and from which no one can easily be excluded. An open-source foundation model meets precisely these criteria. For instance, a hospital that fine-tunes AMALIA to meet its clinical and research needs takes nothing away from a university that uses it for educational purposes, a court that relies on it for legal proceedings, or a company that adopts it to automate specific tasks. Moreover, the open license means no gatekeeper can shut anyone out.
Economics has a well-known prediction about goods of this kind: Private markets will not supply enough of them, because whoever builds them cannot capture more than a fraction of the benefits. That is precisely the pattern visible in language. Portuguese is the fifth most spoken language in the world, with some 260 million speakers, yet it accounts for roughly 2 percent of the web data commonly used to train large models. The benefits of a model that renders a language and its culture faithfully are spread across every school, firm, and public agency that uses it, which is exactly why no private developer has found it worthwhile to provide one openly; where private developers have tried, as in Brazil, the result is a closed commercial product. Portugal's government stepped in where the market predictably did not.
The United States has organized things the other way around. The most capable models in the world are private goods in the strictest sense. Their inner workings are closed, access is sold on a subscription basis, and the terms can be revoked or repriced at will. There is no comprehensive federal legislation governing how these systems are trained or deployed, and governance is, for practical purposes, self-governance by the firms themselves. Even the open-weight models released by US companies are instruments of corporate strategy rather than public commitments, and what strategy gives, strategy can take away.
This represents a break with how the United States handled the general-purpose technologies of the last century. The internet began as a Pentagon research project; the Global Positioning System (GPS) was built and is still operated by the US government; the basic science behind modern computing flowed out of federally funded laboratories and universities. These technologies were seeded by government support, which kept the foundational layer open for the private sector to build on, the same architecture Portugal has now chosen for AMALIA at national scale. With artificial intelligence, however, the United States inverted the sequence, keeping the foundational layer privately owned, and making the public sector line up as any other customer.
The inversion is all the more striking because the public is already carrying the cost of the industry and its many externalities. Data centers across the United States receive tax abatements, subsidized land, and priority access to power and water, with grid upgrades financed in part by ordinary ratepayers. Meanwhile, those same households are bearing the brunt of noise pollution, relative water scarcity, and high electricity bills. The inputs of artificial intelligence are being socialized, while the outputs remain proprietary—the mirror image of the public utility model, in which public investment in the foundation is justified by open access to its outputs.
To be sure, public provision carries its own risks, and AMALIA could yet fall victim to the familiar ones. The model might age while the frontier advances. Its budget needs might not be met over time, particularly in a challenging political environment. Users might have an incentive to use other models, thereby depriving AMALIA of improvements they might otherwise make to it. The test of a foundational technology is whether anyone builds on the foundation, and that is what deserves watching over the next two years. But the terms of the experiment should be stated plainly. For roughly €7 million, less than the cost of a highway interchange, Portugal has bought itself a public stake in a technology its economy will increasingly run on.
Once upon a time, the United States understood this logic better than anyone. It has now chosen to rent from its own companies instead, and on their terms.
Data Disclosure
This publication does not include a replication package.
