Miners are selling you space in the block they're building, which you pay for with a transaction fee. But that block is getting built either way as long as it is profitable.
Putting this in terms of distributed systems, we would say that the act of mining a block is the act of executing a view change (or a leader election). We would say that transactions are the messages sent to all nodes during the epoch. Obviously a system that only executes view-changes is not an interesting system, and nor is a system that just transmits messages but cannot order them or decide which are accepted by other nodes. But my original point is, these are separate operations, and only one of them is energy-intensive.
Trying to say that both view-changes and message transmission are energy intensive because just one of them is not only belies a fundamental lack of understanding of how the system operates, but also hinders an effective regulatory response to stopping the real problem: pollution from fossil fuels.
What use-case does mining solve? There is no point in it without transactions (ok, there is in terms of an immutable ledger that can store some information, like this event have to have happened before that), so how can you measure a bitcoin transaction’s energy cost otherwise?
It's a view-change operation. If Bitcoin were a conventional replicated database, mining would be Paxos.
> There is no point in it without transactions
As I said above, a distributed system with only view-changes and no messages is not an interesting system. So, yes? But, that doesn't change the fact that only view-changes are energy-intensive in Bitcoin.
> ok, there is in terms of an immutable ledger that can store some information, like this event have to have happened before that
I think this is perhaps by far the most underappreciated value proposition of Bitcoin. I don't really care for the whole "hurr durr let's replace money and visa because freedom" song and dance routine that gets trotted out whenever Bitcoin comes up. But I do have use for a highly-available, highly-resilient, always-online append-only log. Even if that log is slow (messages take ~10 minutes to propagate in expectation) and expensive (I have to pay market rate for block space), the high uptime and resiliency make it worthwhile if you're willing to be clever in how you make use of it. My $DAYJOB makes liberal use of storing hashes and metadata for valuable data in Bitcoin OP_RETURNs, for example.
> so how can you measure a bitcoin transaction’s energy cost otherwise?
Consider the PoW difficulty and the power usage of the state-of-the-art PoW-solving hardware, and work backwards from there. The number of transactions has nothing to do with this calculation.
It's funny -- articles like this one are sooooo close to getting it, but so far away. They do the electricity usage estimation from the cumulative PoW difficulty and efficiencies of mining hardware over time just as I describe above, but then they commit a category error by saying that PoW energy usage tells us anything at all about how much energy a Bitcoin transaction takes.
Just because PoW have to happen regardless of transactions doesn’t make transactions decoupled from it - and thus they have a ‘price’.