Along those lines (synthetic data) I would keep an eye on the chinese labs given that they are probably quite data and compute constrained, in English at least.