An AI researcher who builds open models himself makes an uncomfortable prediction. More companies will release small open models, but fewer and fewer will release open models near the frontier. Training them costs too much, and selling them pays too well. His conclusion, from someone who says he hates consortia: a group of companies jointly funding a set of open base models is the only stable path.
Europe should read that as a job offer. We talk about sovereignty and open source in every policy paper, then wait for an American or Chinese lab to keep giving away its best work. That is not a strategy. It is hoping for charity. If European governments and companies want open AI they can trust, they will have to pay for it together. The bill is real either way. Better to pay it as an owner than as a customer.