You’ll need to give me a source for those numbers. My 1,300 square foot house has a 200 ampere main breaker, so I can’t draw more than about 0.024 megawatts. It would therefore take more than two thousand similar-sized houses to draw 53 megawatts—and that’s only at full power draw, which I never come remotely close to.

Also it’s worth mentioning that timespans don’t actually matter here: monthly usage and annual usage are the same number when dealing with watts. When a source says “average annual power consumption”, they likely mean that the power measurement is averaged throughout the year and that hot/cold seasons are accounted for.

If you wanted to do math involving time, you would need to use watt-hours (or kWh, MWh, etc.), for which time is a factor. For example, if you drew a constant rate of 1 kilowatt for a full day, you would have 24 hours * 1 kilowatt = 24 kilowatt-hours.

If only genetic sequences were so finite. And what happens when researchers are developing a new protein that never existed before? Protein folding is NP-hard—even if P=NP is solved, protein folding may still be a challenge. It is likely that there will always be a need for software like this, and that the software will need periodic updates as new sequence behaviors come to light.

That said, as another comment points out, Google is merely stepping back from the AlphaFold project, which will continue without their funding or oversight.