Banks also create money out of thin air while consuming copious amount of power.
Is there a source for the numbers or estimates for power consumption you are referencing for comparison?
There have been some recent articles that tried to gauge the per-transaction power cost for Bitcoin versus Visa, though there's a lot of assumption built into the figures.
For example the following tries to compare, but I am not familiar enough with the field to know what assumptions might have be in error.
https://digiconomist.net/bitcoin-energy-consumption
It does reference Visa's internal estimate of 675000 GJ for the year 2016, although its data centers use slightly over half and Visa performs more functions than just the strict handling of electronic transactions versus the Bitcoin mining base.
Half of Visa's total consumption in GJ translates into ~.1 TWh in 2016 versus tens of TWh per year for just Bitcoin.
From that source, Bitcoin is projected to have ~56 TWh/year of consumption and Ethereum has a separate listing with ~16 TWh/year. In another crypto mining thread, there was an off-the-cuff estimate of 1/3 of GPU sales going into Ethereum versus other GPU-friendly coins. If that were to hold now, the overall consumption of GPU-mined coins would put Bitcoin+GPU coins at ~100TWh/year or 1000 times more consumption than one of a handful of world-spanning credit networks.
Even if accurate consumption numbers are found for the transactional processes, that's a separate consideration from money creation since Visa isn't a bank.
Actual money creation in the USA involves a smaller subset of banks with accounts that are adjusted by the Federal Reserve. That's a handful of increments/decrements at a slower/occasional cadence, with fractional lending amplifying the effect down the chain of credit and loan transactions. The power cost of the creation of money might be contained in a relatively low number of mainframes worldwide, while the individual fractional actions of the financial system would be occurring regardless.
There's inherent value in a framework that allows anyone anywhere in the world to transact value with anyone else anywhere else in the world without having to employ (and trust) a third party to act as an intermediary.
There is utility in the capability, but that's an ideal divorced from having to weigh that value versus the full accounting of the pros and cons of an implementation.
There is similarly an inherent value to being able to transact value with anyone in the world via a third party capable of navigating a wide variety of complexities, risks, and eventualities an individual may not be able to account for.
Past that, something like cryptocoin is more than just a transaction method. The distributed ledger functionality covers transferring specific tokens, but the question of their "value" and Bitcoin's (and all coins relevant to this discussion) choice of validation, incentives, and monetary policy go beyond just that initial source of utility.
The current implementations are far from perfect, but then again so are the existing financial systems and they have been in development for hundreds of years. So, maybe it's worth giving it a few more years to see where it goes.
I'm pretty convinced at this point that much of the cryptocoin market is an example of ignoring the lessons of the existing systems, since so many of the disasters that struck them over the centuries are happening in overdrive, and it's a possible mark of the immaturity or possibly unsoundness of the approach that many cryptocoin proponents have that one common reaction to attempts to learn from these repeated mistakes is to take one's ball and go to another cryptocoin that makes a point not to have learned from them.
I think the concept of a decentralized, proof-of-work, zero-sum reward, and deflation-directed trustless transactional network is likely to be considered in hindsight to be a lesson in what not to do.
The mining and reward system needs to keep the cost of overriding the trustless system too high for a hostile actor, and it does so with a function that pretty generally devolves to cost of electrical power for the coin type under discussion.
That's separate from the delivered utility for the transaction throughput of the system.
The incentive system does not see power consumption as a limiting factor, and in fact rewards greater absolute consumption without limit. Existing financial entities see power costs as an expense to be minimized, whereas any power not consumed is competitiveness lost for mining.
The unbounded profit function is coupled with an arms race where the motivation is to make it as hard as possible on everyone else to succeed at successfully adding a (valid or invalid) block to the consensus.
The system has relatively few methods for stabilizing value, which hurts the transaction utility of the system.
Being decentralized, attempts at corrections or the application of lessons learned run into the problem that those currently enjoying the benefits of the reward system tend to not agree if corrections impact their unbounded profit function.
The cryptocoins being gauged by their price in fiat also makes it in the interest of those with the most coins or mining throughput in a system whose value stability and transaction utility is marginal to force the fiat to crypto ratio as high as possible, which tends involve minimizing the few remaining rational levers for moderating the system.
I think the current implementation is half-built on something that sets an actor's rational self-interest towards abuse outside of the transactions themselves, and the structure or lack thereof is opposed to imposing collective action or individual restraint necessary to change that.
