Indeed. Even a pretty mediocre modern microcontroller is capable of incredible feats of computation and speed, doubly so if you glue it to an FPGA, even a cheapo one. The fact that each is probably a few mm across and costs almost nothing just adds to it. Many analogue devices and DSP systems would be downright supernatural if you showed it to an engineer in the 70s.

99% of computing power is used for "make work"¹ (graphics, teetering stacks of abstraction and now AI) so things don't really feel different to humans on a desktop level other than "shinier, drop shadows and in 4k I guess?", but the actual capabilities of computers are virtually unlimited in the context of some tasks.

If the robots turn against us and they don't need to use all their cycles on the abstractions and other human frippery, then we're really in trouble. A true AGI will know how to wring everything out of a scrap of silicon and human engineers will be wondering how a program that looks like random noise and fits in a STM8 can possibly be the controller of a captured drone, right before they get headshotted with a ball bearing fired by a passing drone at 1000 feet that picked their heartbeats out of the ambient soundscape or something.

Humans' best defense then would be somehow hide behind something computationally intractable where the AI couldn't use it's raw computing power. I'm not really sure what that would be, though (if I were, I'd probably write a novel!).

¹: well technically all human endeavour is make work, so this isn't meant as a slight, though I have some opinions on the state of modern software, just that the vast majority of the cycles aren't doing the core thing you're trying to use the computer to do. For example a graphical calculator program may be running the thick end of a hundred million instructions to run a handful of actual ALU ops.