Europe is starting to study Chinese AI models. In October 2026.

Politico reports that ENISA and the Commission's Joint Research Centre are now testing Chinese open-weight models, on their own secure computers. Good news. But these models have been free to download since January 2025.

In the meantime, our public institutions have had one real option: American AI in the cloud. And they can't put citizens' confidential data into it. So most of the useful work simply doesn't happen. That isn't caution. It's a lack of vision.

The irony is that the answer was already on the shelf. Take a free open model, run it on your own hardware, in an isolated network. No data leaves the building. Nobody in Beijing or California sees it.

Mistral has just announced Large 4, and I'm glad. But the weights aren't out yet, and nobody has tested it independently. Hope is not a strategy.

In the end, it doesn't matter who owns the model. What matters is who uses it, and on whose data. The US and China are competing so hard that they give their models away. That's an opportunity. Let's invest in our own compute and take it, before we order a study on why we didn't.

First published on LinkedIn.