To be able to hold a 51% attack for long, the attacker would spend more energy than many countries combined. This kind of attack is almost impossible to have any success with the current bitcoin network hashrate.
You don't have to hold the attack for very long. You can at most execute only one attack, if you are out for the profit. The network will react accordingly to the attack to minimize the impact after the attack happen.
Rather than electrical consumption, the main factor limiting something like this is actually the costs of executing such an attack. Purchasing the ASICs for something like this is ridiculously expensive, keeping in mind that it's only good for one attack and they are rendered useless afterwards. Even if you were able to double spend and get your Bitcoins back, what is the use when the price crashes right after? You're still left to deal with ASICs that probably won't be used after as the community explores another algorithm.