They would have been right, just as the person you're replying to will be. There was not the compute equivalent of ENIAC in every person's office by 1950. The real estate requirements would have been enormous.
Did every office wind up with more compute than an ENIAC? Yes. Did it take as much real estate as ENIACs would have? Of course not.
Did it happen in five years? Absolutely not.
Our AI models will get better and more efficient. Our compute will get more efficient. But how much has FLOPs per watt improved in the last five years? Not much. How much more power efficient have models gotten? Not much, we haven't been optimizing for that, since quality isn't even there yet for many tasks; we've instead been making models larger and require even more compute. It's not hard to extrapolate these trends and see that if you were to replace a billion jobs in five years, enormous amounts of power will be required.
It’s more that energy usage isn’t a useful metric - because trying to minimize it leads in entirely different direction (and pretty dark).
Ideally those 1b people would contribute value elsewhere in the world, leading to a big net gain rather than just displacement - and we’re hopefully getting bigger gains for the energy invested. Of course that’s ignoring likely concentration of value, etc.