Ah, important point, although “modelling” might not be the right word, for a bunch of different answers, take a look at https://towardsdatascience.com/theres-more-than-one-kind-of-data-mesh-three-types-of-data-meshes-7cb346dc2819. The short version is:

• modelling in terms of “context-specific modelling” should happen in the data mesh node, the team emitting the data.
• more modelling might still happen in central teams. For instance, analytics teams might still ingest some raw + modelled data from the data mesh nodes, and create new stuff which combines things (like revenue per customer segment).