In my last column, I parked one objection to bitcoin for another day: the energy it consumes. That was mostly because it was an unfair argument to begin with. But let’s try to put it to bed.
The honest answer, five years after the outrage peaked, is that the world has moved the goalposts, thanks to AI. The electricity used to secure the bitcoin network is no longer the most interesting number in the room, next to the appetite of the machines now munching their way through gigawatt after gigawatt in the world’s data centres.
Cast your mind back. For a stretch around 2021, you could not read a tech publication without a headline scolding bitcoin for boiling the oceans, using more power than entire countries to run a speculative token. It made for arresting copy. It was also, even then, a strange place to plant the flag. When someone bothered to do the comparison, bitcoin did not come off worse than the banking sector it is accused of threatening.
A 2021 study by Galaxy Digital put the network at about 113TWh/year, against roughly 263TWh for the banking system and 240TWh for gold mining. Yes, really. Galaxy is a crypto firm with an obvious interest in that answer, so apply the same suspicion I am asking you to apply everywhere else. Nevertheless, the point survives: the branches, card networks, bullion vaults and armoured trucks of legacy finance are not free. They simply do not arrive with a running meter attached. Bitcoin does, which is precisely why it became a target. TechCentral columnist Rob Price argued as much in 2022.
The bitcoin network today draws an estimated 138TWh/year, on the Cambridge Centre for Alternative Finance’s 2025 numbers, or roughly 0.5% of the world’s electricity, and that figure has barely moved in years.
Here is where this argument usually cheats, so let me not. The world’s data centres used about 415TWh in 2024, around 1.5% of global electricity – based on International Energy Agency numbers. But that 415TWh is every data centre on Earth: cloud storage, streaming, e-mail, corporate workloads, the lot. AI is only a slice of it, and until recently a modest one.
Bounded
The IEA puts AI at roughly 5-15% of data centre power in recent years, which on 2024’s total works out at something like 20TWh to 60TWh. Our World in Data reckons AI had reached about 155TWh globally by 2025.
Measured like for like, AI has only just drawn level with bitcoin, and in 2024 bitcoin was comfortably the bigger user of the two. AI has not dwarfed bitcoin’s energy use – yet.
What it has done is more consequential. Bitcoin’s consumption is bounded, capped by what miners can profitably spend chasing a fixed reward, which is why it has barely moved in years. AI’s is not constrained. The IEA expects AI to take 35-50% of data centre electricity by 2030, somewhere between 330TWh and 470TWh on its own projection – two to three times bitcoin, with bitcoin’s line flat.
The sermon from the anti-bitcoin brigade did not stop because bitcoin got smaller. It stopped because attention shifted to AI data centres.
There is a second thing the comparison reveals, which the headlines of 2021 would have found unthinkable. Bitcoin is the cleaner of the two. On the IEA’s own accounting, data centres draw only about 27% of their power from renewables and 15% from nuclear, leaving close to 60% from fossil fuels, coal most of all. But bitcoin mining now runs at 43% on renewables, 52% if you include nuclear power. It is a little like a delivery van driver lecturing a cyclist about their carbon footprint.
Two things keep that from being a cheap shot. A data centre must sit near cities and fibre, so it takes whatever the local grid burns, while a miner can set up wherever power is cheapest, which is often exactly where clean or wasted energy is stranded. And data centres are greening, with the IEA expecting their mix to tip towards 60% clean by 2035. The point is not that bitcoin is saintly and AI is filthy. It is that the pollution sermon was preached to the wrong pew.
Bitcoin got cleaner for unsentimental reasons. A miner’s only real enemy is the electricity bill, so it chases the cheapest power on the planet, which is very often the power nobody else wants: stranded hydro, curtailed wind and solar, and gas flared at oil wells for want of a pipeline. Cambridge puts network emissions at 39.8 megatonnes of CO2 equivalent, the profile of a mid-sized industrial estate rather than a civilisational threat.
Steven Boykey Sidley, the South African bitcoin author and partner at Bridge Capital, puts it more bluntly. “No one talks about bitcoin mining and energy anymore because it became extremely clean, moving to renewables in search of lower costs,” he told TechCentral, pegging the network at 65% green. His figure sits at the optimistic end of a contested range, since Cambridge’s own accounting is a more conservative 52.4%. But take the cautious number or the bullish one, the direction of travel is the same.
The twist the energy debate didn’t see coming is about money, not megawatts. The April 2024, halving cut the block reward to 3.125 bitcoin, halving miners’ revenue per block at a stroke, while total computing power climbed to a record near 1 160 exahashes a second in October 2025, pushing up difficulty and thinning each machine’s reward.
Then came a second shock, courtesy of AI: DRAM contract prices surged by 90%-95% in the first quarter of 2026 alone, roughly doubling the cost of memory, with AI demand largely to blame. The two do not use the same chips, but they compete for the same fabs, capital and, above all, power.
The miners became landlords
The rational response wrote itself. If you own the scarcest asset in the AI buildout – a powered, permitted, grid-connected building – why fill it with machines earning a collapsing return when an AI company will pay far more for the same megawatts?
“The advantage of bitcoin mining is that you actually mine coins with a value attached, and some of it remains with the miner,” says Sidley. “Not so with AI. You just get paid for your electricity.”
A utility that mines keeps the upside if the coin climbs. One that rents its power out collects a fixed cheque and nothing more, though the coin can of course fall as easily as rise.
If miners are drifting to AI, is bitcoin mined more slowly, stretching the four-year halving cycle? No, thanks to the protocol’s design. Issuance is tied to block count, not miner numbers: the reward halves every 210 000 blocks, and the difficulty adjustment automatically retunes the puzzle every fortnight to keep blocks landing about 10 minutes apart.
And this is no longer hypothetical. Computing power has fallen about 20% from its October 2025 peak to somewhere around 900 to 930 exahashes, and difficulty is down roughly 15% across 2026, yet not a single coin has been issued behind schedule. Lower difficulty simply makes mining more profitable for whoever stays, which puts a floor under it.
The genuine worry, as Sidley warns, is not coin supply but network security, since a smaller hashrate is in principle cheaper to attack. Even so, at 900 exahashes the network still dwarfs anything an attacker could muster.
Sidley calls the wider shift the great defection. The capital that flooded into mining was never sentimental, and the moment the same electrons, land and transformers were worth more to a chatbot than to a coin, it repriced the industry and moved on.
If you still want to worry about a technology’s appetite for electricity, that is perfectly reasonable. Just point the worry at the right number. Make a fair comparison. – © 2026 NewsCentral Media
