I believe that would be up to each individual app? It seems that in addition to ARM cores and a Mali GPU (targeted without issues by OpenGL ES for example) the Google Tensor chips also have an “Edge TPU” and it seems to be usable via the NN API (developer.android.com/ndk/guides/neuralnetworks) and a quick search through the GrapheneOS Github also yields this: github.com/GrapheneOS/…/edgetpu, in particular:
[Bringup] Enable darwinn EdgeTPU implementation of NNAPI on PRO.
The NNAPI is now fully functional and passes all the tests. This CL enables the Darwinn NNAPI service by default on the master branch.