An argument made the rounds in mid-September: bitcoin mining's electricity consumption may have already peaked, and the network may never again draw as much power as it did in 2024 and 2025. It came from Saifedean Ammous, author of The Bitcoin Standard, and it is worth taking seriously. It is also worth reading carefully, because the useful part for anyone running a machine is not the prediction. It is the math underneath it.
What the argument actually says
Ammous's case is not that efficient machines are quietly replacing old ones. His mechanism is money, not hardware. The block subsidy is now 3.125 BTC, cut in half at the last halving, and mining is a zero-sum race for that shrinking reward. He argues revenue is no longer keeping pace with the cost of competing, so miners at the margin stop adding capacity and switch off, which caps total power draw. He puts the network's consumption near 190 TWh annualized at the December 2025 peak, falling into the high 130s TWh by mid-2026, and adds that AI data centers are now outbidding miners for the cheap power that used to feed expansion.
That is a forecast, and the headline word is "may." It is not a measured fact, and it deserves the same scrutiny as any prediction.
What the data actually shows
The Cambridge index that tracks this puts bitcoin's annualized electricity use around 138 TWh in mid-2026, in line with the lower end of Ammous's range. But that same index has always been cyclical rather than a straight line. It climbs in bull markets when miners switch everything on, and it dips when prices fall and unprofitable machines go dark. The 2026 reading is a drawdown from a late-2025 spike, which is exactly what the normal cycle looks like. Network hashrate tells the same story: around 1,000 to 1,020 EH/s in 2026, off the roughly 1,160 EH/s peak from October 2025, but still near record levels.
So the honest read is that consumption has come off a high, not that it has permanently peaked. A peak has been called after every prior halving, and every prior time, hashrate eventually climbed to a new record. It may be different this time. It has not been proven yet.
The number that actually matters: 18.9% a year
Buried in the argument is a piece of arithmetic that is exactly true and worth understanding, because it applies to your machine no matter who is right about the peak. The halving cuts the block subsidy in half every four years. For the bitcoin you earn to hold its dollar value across that span, the price has to double over four years just to stand still. Doubling over four years works out to about 18.9% a year, because 1.189 multiplied by itself four times equals two.
Turn it around and it is the same fact: in bitcoin terms, the subsidy shrinks about 15.9% a year, every year, forever. That is the treadmill every miner runs on. It is not a price prediction, and it is not advice to expect any particular return. It is just the structural headwind built into the reward schedule. Transaction fees offset some of it in busy periods, and rising difficulty can make it steeper, but the core pull is always there.
Where efficiency comes in, and where the argument could be wrong
There is a separate, real trend that often gets folded into this debate: machines keep getting more efficient. A decade ago rigs ran near 90 J/TH. Today's fleets average roughly 15 to 24 J/TH, and the newest hardware is under 10. That genuinely decouples hashrate from power, because the network can compute more while drawing proportionally less. It is a strong point, but it is a different argument from Ammous's, and it cuts both ways. Cheaper power per hash can invite more total hashrate, not less, which is how efficiency gains can leave total consumption flat or even higher. Add stranded and flared power still coming online around the world, and old rigs that stay plugged in as long as prices cover their electricity, and you have several reasons the network's power draw could keep rising regardless of the halving.
What it means for your machine
Here is the part that survives whoever wins the argument. You cannot control the halving, the price, or the network's total energy use. You can control two things: how efficient your machine is, and what you pay per kilowatt-hour. Those are the only two levers that decide whether you stay ahead of the 18.9% treadmill. A more efficient rig earns more hash per watt. A low power rate widens the gap between your revenue and your cost, and keeps a machine profitable deeper into a down cycle when the marginal, high-cost miners are the ones switching off.
That is the whole reason low-cost hosting matters more as the subsidy shrinks, not less. If you want to see how the halving math plays against your own rate and hardware, model it with our profitability calculator, which uses live price and difficulty. Fathom Labs hosts at $0.07/kWh all-in in Southeast Texas, with no performance fees, so more of every block stays on your side of the treadmill.
Frequently asked questions
Has bitcoin mining's electricity use actually peaked?
Not confirmed. In mid-September 2026, The Bitcoin Standard author Saifedean Ammous argued it may have peaked in late 2025. Cambridge's index shows annualized use around 138 TWh in mid-2026, down from a late-2025 spike, but that index is cyclical and has dipped before without marking a permanent top. A peak has been predicted after every past halving and been wrong each time, so this is a forecast rather than an established fact.
Why must bitcoin's price rise about 18.9% a year?
Because the halving cuts the block subsidy in half every four years. For the dollar value of that subsidy to stay flat, the price has to double over four years, which is about 18.9% compounded annually (1.189 to the fourth power equals 2). Stated the other way, the subsidy shrinks roughly 15.9% a year in bitcoin terms. It is arithmetic from the reward schedule, not a prediction that the price will rise.
Does more efficient mining hardware lower total energy use?
Not necessarily. Efficiency, measured in joules per terahash, has improved from around 90 J/TH a decade ago to under 10 on the newest machines, which decouples hashrate from power. But cheaper power per hash can attract more total hashrate, so the network's overall consumption can stay flat or rise even as each machine gets more efficient. Efficiency clearly helps an individual miner's economics; its effect on the whole network is less direct.
What can a miner control as the subsidy shrinks?
Two things: machine efficiency and the price paid per kilowatt-hour. Neither the halving nor bitcoin's price is in your hands, but both of your costs are. An efficient machine on a low, fixed power rate stays profitable further into a downturn, when higher-cost miners are the first to shut off. That is why the power rate matters more as the block reward falls, not less. Fathom Labs offers $0.07/kWh all-in hosting with no performance fees.
Want to see where your setup sits against the halving math? Talk to us and we will run your numbers.