Every big artificial-intelligence system today - from OpenAI, through Google, to Anthropic - belongs to a private company. One nonprofit wants to rewrite exactly that sentence. The idea is simple and ambitious: public AI infrastructure, open to everyone, free - the way the internet itself once was.

The organisation is called Current AI, and behind it stands not Silicon Valley but - Paris. The French government launched it with 100 million dollars, and then the Ford and MacArthur foundations, DeepMind and Salesforce joined in, pushing the total to 400 million. „They aren't investors, they're financiers“, says executive director Ayah Bdeir. The difference isn't cosmetic - an investor wants a return, a financier wants a result that stays public.

Why is this a story for the Balkans at all? Because half the world's languages are on the road to disappearing, and the big models feed almost only on English. Every people whose language isn't „profitable“ enough for the big companies ends up invisible to the technology that promises to change everything. Macedonian, Albanian, every smaller language - the same logic applies.

Bdeir, former head of AI strategy at Mozilla, draws a sharp line. „The big firms build multilingual models to expand their market“, she says, „regardless of consent or context.“ An example that stings: for indigenous languages, missionary translations of the Bible become training data before the communities even set any rules.

The first money is already distributed - 3.2 million dollars to four organisations in Kenya, Lebanon and the Brazilian part of the Amazon. In Kenya, AI data is being built for over 50 African languages; in Lebanon, Arabic cultural history is being digitised into databases controlled by communities, not the tech giants. Last month in Geneva, an open chatbot was released, assembled in seven weeks by a coalition that includes Hugging Face, Mozilla and MIT Media Lab.

„Scale isn't always the measure. That's the paradigm of the big firms“, Bdeir says. The question hanging over the whole venture is whether 400 million can stand up to companies that spend that much on a single training run. But the very framing of the question - whose data, whose language, who decides - is something the big players long ago stopped asking.