When people talk about bitcoin mining they often bring up the energy costs. They are not however considering how bitcoin is set up over time. The reward for miners is reducing over time to near zero.
In the beginning the mining reward was a high number of bitcoin, say 250 bitcoins, but the price of bitcoin was very low, say 1c. This meant that you only got rewarded $2.50 worth for each block you mined, so not so many people felt it was worth mining. Bitcoin needed enough people to mine so that some other people who didn't like bitcoin, with more computer power, couldn't get in and mine and effectively kill it. It needed to be strong enough so that a government super computer couldn't hack it.
Now bitcoin computer power is way beyond what any one bad actor can harness. This security challenge is surpassed, likely by more than a 1000x at the moment and the mining reward has halved to 12.5 bitcoins. That reward comes to around $120,000 per ten minute block.
People are complaining that it has gone too far, and they might be right. It uses as much power as a decent size country, but I would argue that this is short sighted. Next year the reward will be 6.125 bitcoins and if the price does not go up we will see less bitcoin mining happening and do we really know what amount of mining is enough to keep a potentially world changing technology safe? We are high now, but the percentage of bitcoin resource being allocated to this reduces and is only counteracted by price.
This counteracting effect only works so much, eventually (in something like 100 years) the mining reward is programmed to go to zero which will make $0 per block. We can expect the mining to be much less power intensive as compared with the whole bitcoin economy at this point. At this stage mining will still happen though as voluntary transaction fees also go to miners. This type of payment needs to counteract the block reward completely at this stage to make sure that bitcoin has enough computer power.
When the block reward is zero, bitcoin mining power consumption will be directly proportional to how much value we place on security and transactions. I would argue at this point it is nearly 100% efficient.
