You were right on both counts! I have fixed the mistakes in the article.

I did not enable the GPU when I was using TPOT because it does not support optimization on a GPU. (TPOT also does not test neural networks). Reading through some of the issues on GitHub, it looks like GPU support may have planned for the future, but was never completed.

As far as I can tell, the only automated machine learning project that can use a GPU is Google’s AutoML which has been used to optimize neural network architectures. Cloud AutoML is currently in Alpha and is not publicly accessible (although you can apply to be considered to test it out).