"AI compute module"s exist, they are called GPUs. All the matrix calculations that go into neural networks are highly parallelizable, which means GPUs are optimal for them. A cheap used GPU will beat anything you can cook up yourself.