We send an AI the exact same question twice. The first time it has to do all the work, and you can watch the answer arrive word by word as it thinks. The second time, the identical request comes back complete in an instant, because the work was already done.
Most systems don't work this way. Ask the same thing a thousand times and they compute it a thousand times, and you pay for every one. On TAHO, work that has been computed once is referenced from then on.