I have excellent news for you. Lux @ ORNL and Equinox @ Argonne are to be completed by EOY, with Solstice (100k NVIDIA chips, currently spec'd to be Vera Rubins) in the next five years.
It's certainly large enough for trillion-param frontier-tier trainings, which will likely result in capable open-weight models, the thing you just wished for.
Is this supposed to be impressive? Five years for the equivalent of, what, Colossus 1? What a joke