I've long envisioned a glue architecture around a scalable SmallTalk core. The ST bytecode can be run on various size compute cores from 8-bits to 64-bits. All the ST does is handle code flow. All compute is handed off to co-processors (Integer, floating-point, matrix, GPU, etc). The compute functions can be correspondingly scaled based on family design. The general goal is to allow thousands of cooperating compute processors with edge processors for communication.
https://www.tcl-lang.org/doc/scripting.html