ASIC efficiency, a decade of progress in joules per terahash

If you want to understand the economics of Bitcoin mining in one number, ignore hashrate and look at joules per terahash. Hashrate tells you how much work a machine does. Efficiency tells you whether doing that work earns or loses money, and the history of SHA-256 ASICs is best read as a decade-long march down the J/TH curve, with every step down rendering a generation of machines above it obsolete.

The curve, generation by generation

Consider the flagship Bitmain units as representative mileposts, with competitors from MicroBT and Canaan tracking the same trajectory within a few percent.

• 2016, Antminer S9: roughly 100 J/TH. The machine that industrialized mining, produced in the millions, and profitable for so long that it distorted expectations of how long an ASIC should live.
• 2018, S15 and the 7nm transition: around 57 J/TH.
• 2019, S17 family: 40 to 45 J/TH.
• 2020, S19 and S19 Pro: 34 and 29.5 J/TH. The workhorse generation of the 2021 cycle.
• 2022, S19 XP: 21.5 J/TH.
• 2023, S21: 17.5 J/TH, with hydro variants pushing lower.
• 2024 to 2025, S21 Pro, S21+ and the S21 XP line: 15 down toward 12 J/TH, with the newest hydro units in the low teens and the marketing decks already promising single digits.

Call it a factor of eight improvement in ten years. Note the shape, though: the early jumps halved energy use in a generation, while recent steps grind out 15 or 20 percent. Silicon process gains below 5nm are slower and costlier, so the curve is flattening, which matters for the survival math below.

Why efficiency, not hashrate, decides survival

A mining machine dies on the day its electricity bill exceeds its revenue, and both sides of that ledger are set per terahash. Revenue per TH is the same for every machine on the network; it is determined by price, difficulty and fees, and it only trends one way over time as difficulty rises. Cost per TH is where machines differ, and it is exactly your electricity price multiplied by your J/TH.

Run the numbers for a common industrial rate of 6 cents per kWh. An S9 at 100 J/TH spends about 0.144 dollars per TH per day on power. An S21 at 17.5 J/TH spends about 0.025 dollars. Whenever network revenue per TH sits between those two figures, and for most of recent history it has, the same electricity contract feeds a profitable S21 and a money-losing S9 sitting on the same shelf.

This is why halvings function as extinction events. Overnight, revenue per TH drops by nearly half, the survival line jumps down the efficiency curve, and every machine now sitting above it becomes scrap or migrates to the only place it can survive: cheaper power.

Electricity price is the other axis

That migration explains the geography of mining. A machine that is hopeless at 8 cents per kWh can be comfortable at 3 cents, so old hardware flows toward stranded hydro, flare gas and other marginal-cost power, while operators paying grid rates must stay near the efficiency frontier. Roughly, every halving of your power price doubles the J/TH you can tolerate, which extends a machine's life by one to two generations.

For any specific machine and rate, this is arithmetic, not speculation. Put your electricity price and a machine's specs into the mining calculator and the breakeven falls out immediately. Comparing your own candidates side by side on /hardware with your real power cost is the single most useful exercise before buying anything, new or used.

What the flattening curve changes

The strategic picture for the next few years follows from the physics. When efficiency doubled every two years, buying the newest machine was almost always right, because it outlived two generations of difficulty growth. With gains per generation shrinking, machines stay competitive longer, payback periods stretch, and the purchase decision shifts toward price per TH, cooling and power contracts rather than chasing the last few J/TH.

It also means the brutal filter of the past decade, where each generation buried the one before it, is giving way to a slower grind where electricity price and uptime decide margins between operators running similar hardware. The efficiency race made mining industrial. The flattening curve is making it operational.

Either way, the rule that has held since the first ASIC powered on still holds: know your J/TH, know your power price, and know which side of the survival line their product puts you on. Machines do not die of old age. They die of inefficiency at a given electricity price, and both numbers are on the spec sheet.