The "cores" are becoming more specialized and optimized for domain specific tasks. Compiler technology advancements are needed to take advantage of such heterogenous architectures in a transparent way. LLVM MLIR started that already.[1,2] The alternative is being stuck with each silicon vendor's proprietary solutions like CUDA.

[1] https://mlir.llvm.org/ [2] https://www.theregister.com/2022/04/04/compiling_the_future/