(Even if you use resistive heating to heat your living space, it's not necessarily a win to make a computer convert that electricity into heat and computation results. They're 100% efficient, but heat pumps are MORE than 100% efficient. So unless you really want the compute results, it's a waste of electricity.)
Another angle that people always seem confused about is latency versus throughput. When you decide to look up cat pictures, your computer needs to perform billions of calculations in a very short period of time. That means that chips have to be able to do a lot of work in a short interval of time, or nobody will want them. A higher utilization could be achieved by building a machine that is less powerful, but spreads the cat picture rendering calculations out over time; if you want to see 8 cat pictures a day, and 100 compute units are required to display a cat picture, then you only need a CPU that can process 0.01 compute units per second. But, then the human is idle for 10,000 seconds in the time between wanting to see a cat picture and seeing the cat picture, which is annoying to the humans. So the chips have to be "overpowered" to be able to display cat pictures in a timeframe that is non-annoying to humans.
TL;DR: don't feel sad for engraved chunks of sand.