I am in software (primarily in optimization) and I very much want that law to be true, but I am pretty sure it's not. We use multiple GPUs at 100% capacity (al

I am in software (primarily in optimization) and I very much want that law to be true, but I am pretty sure it's not. We use multiple GPUs at 100% capacity (all cores) for ML training. When need be, we use parallelization/concurrency to use every core in computer for our application. Almost no computer application today buffers while processing (thanks to parallelization / concurrency) Software can improve a lot faster than hardware can (due to low capital required)