Most arguments about blockchain energy use run on a single statistic: energy per transaction. Take the network's total power draw, divide by transactions processed, publish the result.
The arithmetic is fine. The problem is what's in the numerator and what's in the denominator, because they aren't connected the way the division implies.
What actually sets the energy
Proof-of-work mining is an open competition. Anyone can add machines, and people do so as long as it pays. That pushes the industry toward the ordinary competitive equilibrium, where marginal revenue meets marginal cost:
aggregate miner spend ≈ block rewards + fees
Miners spend up to roughly what mining pays. A large share of that spend is electricity. So the network's energy consumption is anchored to what the protocol pays out, not to how busy it is.
Rewards are issued on a schedule: so many coins per block, halving at fixed intervals. That schedule doesn't consult the transaction count.
What doesn't set the energy
Here's the part the metric misses. Consider a block with n transactions in it:
hashing work to mine a block with 1 transaction = W
hashing work to mine a block with 3,000 transactions = W
Identical. Mining is a search for a hash below a target. The block's contents are an input to that hash, but the number of transactions inside doesn't change how many attempts the search takes. Difficulty sets that, and difficulty is adjusted to hold block time steady, not to reflect how full blocks are.
So the marginal energy cost of adding one more transaction to a block is approximately zero.
Why the division misleads
Put those together and the metric looks like this:
energy per transaction = (energy set by the reward schedule)
─────────────────────────────────────
(transaction count, set by demand)
A numerator driven by one thing, a denominator driven by something unrelated.
Which produces genuinely strange behavior. If transaction volume halves tomorrow while hashrate stays flat, "energy per transaction" doubles, and not one additional joule was spent. If the network's blocks fill up more, the figure falls, again with no change in consumption. The statistic moves when nothing physical changes.
It also means the number can't be used the way people use it. Choosing not to send a transaction doesn't save its "share" of the energy, because that share was never caused by the transaction. And batching many payments into one on-chain transaction improves the ratio dramatically while changing actual consumption not at all.
Better questions
If the point is to measure real consumption, measure it directly: energy per unit time, which is just power. That number is answerable and it moves when something real moves.
If the point is to judge whether the consumption is worth it, the honest framing is energy per unit of security. Proof-of-work spends energy deliberately, to make rewriting history expensive. Whether that's a good trade is a fair argument to have, and it's the argument actually on the table.
It's also why proof-of-stake is a genuine change rather than an optimization. It doesn't make the hashing cheaper; it removes the hashing, and sources security from bonded capital at risk instead of from burned energy. Different security model, different assumptions, energy demand in an entirely different regime.
Reasonable people land in different places on that trade. They should at least be arguing about the trade, and not about a ratio whose denominator was never driving the numerator.