AI is a kind of average result on a question, and here we will need an absolute answer, so I doubt it would work but I would love to have a expert in AI to answer if it's possible or not
FPGA does, they are excellent for Integer and parallel processing, that's why I get a high speed due to the parallel operations.
When I have the Sbox in the BRAM (as ROM) it set a low fmax due to the address decoder in the nature of the way a RAM does work (the more address lines the slower.
Creating the SBOX with logic are faster the BRAM, but the you use way to much logic and have empty BRAM's so it's better to have more cores even they are slower
Also the fmax are limited due to the big internal routing in the FPGA chip and here ASIC would be way better faster and more cores on less space ... but incredible expensive unless you order millions of chips.