Etheriums PoW algorithm isn't ASIC resistant, just largely unimplemented. Indeed a large reason NIST made Keccak SHA3 is that ASIC/FPGA acceleration is on par or even faster than SHA256 acceleration.

It may take a bit more ram and need a decent bus as a result which is why people mine on GPUs that have both. Rather than cheap ASICS that are cheap because they have neither. "Resistant" ain't the word to use.

But the ASICs are used because they are more efficient. Forcing miners to use less efficient hardware is going to drive up the average energy per hash calculated.

You're completely ignoring the difficulty mechanism in Bitcoin. When everyone has a 1TH/s miner that consumes 1kWh and suddenly a 10TH/s miner that consumes 1kWh comes out everyone is going to switch to the new miner. Because Bitcoin guarantees that a block is mined every 10 minutes it has to make the calculation 10x harder than before. The end result? The energy consumption didn't change.

What makes Bitcoin more energy intensive then? The price of Bitcoin. When the price of Bitcoin doubles that also means the mining profit doubles. More miners (= more energy consumption) join the blockchain until the profit margin is back to the previous level. The price of a Bitcoin is 70x higher than the price of Ethereum so that suggests that Bitcoin should use 70x more energy.

I don't think you can directly compare the prices of the coins, but you need to look at the reward that's at stake.

12.5 bitcoins every 10 minutes at 7277 USD/bitcoin = 9096 USD/minute

3 eth every 14 seconds at 132 USD/eth = 1697 USD/minute

then you'd expect the bitcoin miners to use 5.3x the energy of eth

EDIT: oh, someone already did the same calculations earlier a couple of comments below :)

Basically correct, except that it isn't the price of Bitcoin or Ethereum that matters, per se, but rather the value of the reward for each block relative to the amount of time required to mine it. For Bitcoin, ignoring transaction fees, the rate is 12.5 bitcoins per 10 minutes; for Ethereum it's 3 ETH per 15 seconds. If the price of 1 BTC is 70x the price of 1 ETH then you would break even expending about 7.3x over a given interval to mine 12.5 BTC vs. 1x to mine 120 ETH. If you spent 70x, however, you'd soon be bankrupt despite the higher BTC price, because you don't earn 70x the value.

So energy usage should drop quite a bit after some difficulty increases?

The difficulty increases or decreases to maintain the target block interval. The only way that impacts mining revenue is if the actual block interval deviates significantly from the target—for example, if the hash rate suddenly increased such that blocks were being solved in five minutes instead of ten, that would double the revenue rate (and thus the economical power consumption) temporarily until a difficulty increase brought the interval back to ten minutes per block.

What does cause significant changes in revenue is the periodic halving of the block reward. Absent an opposing increase in price and/or transaction fees, when the reward drops from 12.5 BTC/block to 6.25 BTC/block in May of this year the energy budget for profitable mining should be cut roughly in half.

Sorry yeah I was mixing those up. But is it going to use not-obscene amounts of electricity eventually?

The system is meant to balance itself out one way or the other. You throw more efficient resources at it, and it becomes less efficient to maintain its 10-20 second block times. It's actively anti-efficient.

The idea behind ASIC resistance is not efficiency, but rather to limit centralization. If you can use generic hardware that anyone can access it is intended to reduce capex required to become a cost-effective miner. Now of course, I don't think that makes sense, but that's the idea.

Ah, you're right. After having enough hashes to secure against attacks, they don't really do anything useful. People will just mine as many as they can with a given budget, which buys the same number of kWh no matter how efficient the hash calculations are.