InnovateUS 
There is nothing automatic about new technologies bringing widespread prosperity. Whether they do or not is an economic, social, and political choice. (Daron Acemoglu and Simon Johnson, Power and Progress)
The deepest purpose of technology is a simple one: to solve problems, and to concretely improve people’s quality of life. A man who understood technology well told us as much seventy years ago: Adriano Olivetti, the Italian industrialist who built one of Europe’s first computer companies around the belief that a factory should enrich its whole community, not just its shareholders.
For him, technology had to serve people and community and never become an end in itself.
Public AI is not simply AI purchased or hosted domestically. It is AI infrastructure and capacity designed to serve public needs and subject to meaningful public control, with systems developed under transparent governance, public accountability and equitable access to core components.
Public AI describes the capacity, and democratic AI describes who governs it and how. Meaningfully achieving these goals raises three questions.
Who pays? Public AI needs public investment: infrastructure, models, and data that public institutions and smaller organizations can use, not only large technology firms.
Who has control? A taxpayer-funded model can still be closed, unauditable, and shaped by whoever builds it rather than by the public that paid for it.
Who decides AI’s purpose? Even if the dataset is open, publicly governed capacity needs a purpose. Is it set by a vendor’s product roadmap, or by the needs of citizens and the institutions that serve them?
Our argument is that public money is where Public AI starts, not where it ends. Without public control and a publicly decided purpose, investment buys capacity that others will steer. Who pays, who controls, and who decides AI’s purpose are political choices, and Europe should make them democratically, before the firms selling the technology make them by default.
Our argument is that public money is where Public AI starts, not where it ends.
Three cases each illuminate one question: Spain shows what public money can build, the United States what happens when it does not buy public control, and Italy who should decide AI’s purpose.
The European Strategy
Facing growing dependence on U.S. and Chinese technology, Europe first answered with the 2025 AI Continent Action Plan, which called for a network of AI factories and “gigafactories,” backed by the InvestAI initiative to mobilize €200 billion, along with measures to expand data, skills, and adoption across the continent.
More recently, it adopted the June 2026 Tech Sovereignty Package, which includes a Chips Act 2.0, a Cloud and AI Development Act, and an EU Open Source Strategy. The package’s aim is to develop and control key technologies, data, and infrastructure while reducing reliance on non-EU providers.
But a further need is now emerging: sovereignty measures secure Europe’s capacity, but not the purposes that capacity serves. That calls for a durable public interest approach built on a public AI framework that treats European AI as public digital infrastructure, designed for public purpose.
Spain: What Public Money Can Build
Spain’s ALIA project, announced by Prime Minister Pedro Sánchez in January 2025, shows that public investment can develop AI models with an emphasis on openness and transparency.
Coordinated by the Barcelona Supercomputing Center and trained on MareNostrum 5, Spain’s public supercomputer, ALIA offers open language models in Spanish and the co-official languages of Catalan, Valencian, Basque, and Galician.
Spain is a clear answer to the first question: it shows what public investment can provide.
But the sequence matters. Public investment bought Spain capacity; it did not by itself buy democratic control over how that capacity is used. That is the second question, and the American debate shows what happens when it goes unanswered.
The United States: When Public Money Doesn’t Buy Public Control
If Spain shows what public money can build, Washington shows where accountability breaks down when AI is bought rather than built with public oversight.
In the U.S., the politics of Public AI have surfaced through a different door: a dispute between the U.S. Department of War and Anthropic over contract terms. The Pentagon wanted to use the company’s models for “any lawful use”; Anthropic refused to allow mass domestic surveillance or fully autonomous weapons, and was designated a supply chain risk in response.
It has become a broader argument that government procurement is not a back-office function but a form of democratic governance.
When the only actor capable of restraining a government’s use of AI is the firm selling it, democratic control has already been contracted out. A safeguard that depends on a vendor’s conscience is not a safeguard; it is a commercial preference that can be revised at the next board meeting.
When the only actor capable of restraining a government’s use of AI is the firm selling it, democratic control has already been contracted out.
Italy: Who Decides AI’s Purpose
The third question is the most political, and in Italy one party has begun to put it to voters. On 9 June 2026, Italy’s Partito Democratico hosted an event at its national headquarters titled “The European Third Way: An AI at the Service of People.” (Marco Bani spoke at the event.)
Its central message was that designing public services well, using the best available technology, is not a question of aesthetics; it is a question of democracy. From a private service that doesn’t work, you can always switch providers. From the state, you cannot. And every service that fails is a crack in public trust in institutions, a crack that, in times like these, we cannot afford to widen.
The vision offered by Italy’s main opposition party was for a genuinely democratic AI: not a laboratory toy, nor another startup built to be sold off, but essential infrastructure, on par with highways and parks.
One way to give that vision institutional form is a CERN for AI, modeled on Europe’s joint particle physics laboratory, an idea championed since 2018 by CAIRNE, a European network of AI research labs. The case for a treaty-based laboratory is that its work program can be commissioned by public demand rather than by a product roadmap, with priorities set together with the institutions that carry Europe’s hardest problems: health systems facing an aging population, energy and climate agencies, and schools and municipalities working with shrinking budgets. Whatever it produces would be open by statute. Build the capacity, but let public needs decide its purpose.
Objections: Why Not Just Buy?
Some may say, “commercial AI is better and available now.”
Perhaps, but today’s AI feels almost free because it is being subsidized, and public institutions cannot assume that will last. As these systems become infrastructure, pricing is likely to shift toward metered consumption, as it did with cloud computing.
The institutions least able to absorb that cost are the ones we claim to be equipping: schools, municipalities, public health services, and the non-profit sector. A new digital divide risks forming, defined not by who owns a device but by who can afford good AI.
Others may say, “governments are bad at building technology.”
Often true, which is why the answer is not for every ministry to build its own model. It is shared capacity, pooled across institutions and borders and built with research centers that already know how, as ALIA was, and governed so that public bodies remain the clients rather than the captives.
What a Public and Democratic AI Political Agenda Might Include
Build shared capacity to keep AI affordable. Spain shows that public investment can build real capacity on shared public infrastructure. Public authorities need to extend that now through pooled compute, joint purchasing, and small, efficient models for routine work. Otherwise, they risk spending the 2030s paying a handful of companies for the tools they need to serve their citizens. This answers who pays.
Keep safeguards in public hands. The Washington dispute shows that procurement is governance, but also that procurement alone is not enough when the purchaser is the party whose use needs limits. Public institutions must use their purchasing power deliberately, requiring portability, accountability, and interoperability. But legislatures and independent oversight bodies, not vendors, must set the limits on how governments use AI. This answers who controls.
Start with public needs. The Italian proposal reminds us that vision comes first, technology second. The right model is a community that first decides what it wants to become, and only then uses technology to get there faster and better. Choosing a technology is not a neutral decision to delegate to technologists, let alone to whoever happens to be selling it. The use case, not the vendor, should set the specification. This answers who decides AI’s purpose.
Choosing a technology is not a neutral decision to delegate to technologists, let alone to whoever happens to be selling it.
Make these choices part of the democratic debate. With national elections due in 2027 in France and Italy, it is wise to prepare now with a vision for Democratic and Public AI and a portfolio of initiatives that put it into practice. That means being ready to explain to voters who is paying for the data centers, how jobs are changing, how electricity bills are affected, and who decides.
The Larger Question
The real question is never “how do we use AI?” It is what kind of society we want, and whether we intend to point this technology at the problems that matter: the climate, cancer, the care of an aging continent.
Olivetti answered that question for his factory seventy years ago. For AI, it is currently unanswered, and politics does not tolerate a vacuum for long.
Leave these answers empty, and the solutions will be filled by the firms selling the technology, and by those who argue openly that democracy is in the way. The reason to answer it now is that someone else already is.
Building European computing infrastructure, supporting open models, reforming procurement, and investing in research and training all require significant resources. But, paraphrasing a well-known line: if you think shaping your own future is expensive, try irrelevance.
Democratic and Public AI Series: InnovateUS and its newly forming European sister entity, InnovateEU, are organizing a new edition of the series on Democratic and Public AI, focused on practical strategies for buying, building, and governing AI. The three-part series explores how governments are acquiring AI today, what those choices mean for innovation, cost, flexibility, transparency, and public oversight, and what alternatives are emerging.