It's technically possible to solve 160 by tomorrow if we start today, as long as there are enough GPUs in existence.

But yes, one of the issues with RC's claim is he used 500 to 1000 GPUs in just 2 months (500 was the overly optimistic case of him being extremely lucky (much more luckier than just his method working as stated) AND without considering any DP overhead AND with using his super-double-fast-than-possible CUDA kernel no one thinks can exist AND assuming his entire distributed system worked perfectly without any bugs, issues, etc. which is always unlikely in the real world). There's better chances winning the lottery than achieving so many advancements in different problem areas.

So, the issue here is that the smoke could be easily lifted on him, because it's not like an every day customer (or employee?) that seeks out renting/"managing" hundreds of GPU instances for months in a row.

Someone who wants to lose his tracks would do it over a very extended period of time, with fewer GPUs. This would also make the attack cheaper, because of spot prices & price per teraflop decreasing over time.

If someone calls this jealousy or being a cry-baby, maybe they should put on their glasses. Creating a curtain of smoke is part of the game, you should think about what's under the curtain, and I think it is about extracting information, under the sense that it might no longer be useful. Don't be fooled that psychology is not part of a very interested organization's toolbox.