The cost and market cap estimates are probably both not good measures of how things would work out in practice on that scale, but even so- it looks like someone with a couple million dollars to invest could plausibly extract several billion from BTC via a 51% attack. It would cost substantially more than the estimate on here (since even non-NiceHash cloud GPU services probably wouldn't fully cover the required hashpower), and require substantial technical expertise, and decently-connected people would probably see it coming at least days in advance due to the movements of GPUs and GPU-power involved, and be pretty definitely illegal, and BTC's market value would be crumbling underneath you as you tried to sell out... but, the sheer profitability of such a scheme should seriously worry people with a stake (finanical or emotional) in BTC. A dollar bill sitting on the sidewalk doesn't stay there long.
I go into more detail in this [0] comment, but it would be substantially more expensive to pull off an attack if the hashing power was not available to rent from NiceHash. The 'nicehash-able' column is meant to represent how much hashing power is available - if it is > ~300%, an attack would be fairly do-able from just renting hash power.