Rob is making a lot of good points here, but just to add my two cents:
Given Germans' aversion to most things AI and given the German political landscape, "Germany [...] deploying a meaningful share of its €1 trillion defence expansion to this kind of thing" seems very, very improbable.
A joint European effort with joint debt guarantees (a la NextGenerationEU, a borrowing mechanism introduced during COVID) or some sort of pooled funding, coordinated by the EU commission, seems more likely.
No chance this will persuade Germans:
1. They're more worried about US domination than French domination, as the US is both more powerful and in a more bullying mood.
2. In exchange for money, compute and staff Germany could require joint custody of weights and legal guarantees on model access. Not 100% solid but much better than the US, which is explicitly indicating they'll cut off access.
3. Even setting that aside, having 3 possible model suppliers (the US, France and probably China) puts you in a stronger negotiating position.
4. Germany has more leverage to bring to bear against France (in non-AI domains) than against the US.
5. In this situation middle powers will be motivated to stick together for pragmatic reasons - France would naturally prefer to be the leader of a non-US AI bloc rather than go it alone, as they lack the resources to compete individually.
This line of argument may sound good to Americans. It won't sound good to others.
The question is whether any grouping of middle powers have the will and capability to pull off a near-frontier fast-follower model, even if they try.
It's a heavy lift but not out of the question, especially if the US prompts an exodus of non-national staff who opt to return home and lead the effort themselves.
Or things reach the point that Germany is deploying a meaningful share of its €1 trillion defence expansion to this kind of thing.