A huge delay.

Keeping the average block time close to 10 minutes when the hashrate is constantly changing is done so by changing the target. The miners must find a solution which is below the current target number. The target number is recalculated every 2016 blocks, and so the target number we are working to at the moment is based on the current hashrate of 200 EH/s.

So, if we suddenly lose 99% of the hashrate, for example, then the 1% which remains will be stuck with the current target until the next recalculation, which will be at most 2016 blocks away. Since the target will be based on 200 EH/s, and we would only be mining with 2 EH/s, then blocks will take significantly longer. Once we do recalculate, then the target would drop by a factor of four (which is the limit of how much the target can change at once), meaning the next 2016 blocks would be faster, but still nowhere near a 10 minute average. This would repeat until things balanced out again.

In your example of leaving only a single CPU miner running, then the discrepancy between hashrate and target would be so ridiculously large that either nothing would happen until other miners came back online, or bitcoin would need to fork to readjust the target or mining algorithm.