This depends on the specific regulation. The datacentre moratoria probably give open-weight models time to catch up by tempering the extent to which the leading companies can turn their capital advantage into market share.
What infrastructure will these open weight models be trained on?
One, the infrastructure is being built for inference. Not training. If all we were doing was training on datacentres, I think America probably has enough already for near-term commercial needs.