ASIC Miner Lifespan: How Long Before It Stops Paying?

Two deaths, one machine

Lesson 1 of this module warned that an ASIC can become scrap overnight — here is the mechanism behind the teaser. Every miner runs against two clocks. The physical clock is boring: fans, capacitors and solder joints wearing out over years. The economic clock is the one that actually retires machines: the network grows around yours, newer chips do the same work on fewer watts, and the protocol schedules pay cuts in advance. The generational march that turned a 2016 flagship into a 2026 relic is M1’s story; all that matters here is that the march continues, and it never checks whether the last generation finished paying for itself.

The physical clock: what actually wears out

The chips barely age. Repair-shop literature (community-sourced, not manufacturer) is consistent about the order of arrival: fans first — the only fast-moving parts, eating dust all day. Parts listings rate the stock 120×38 mm ball-bearing replacements at 50,000–70,000 hours, and that is a manufacturer-at-rated-conditions figure: clean air, rated temperature. Community-measured failures come years earlier in hot, dusty rooms, and an hour rating is a statistic, not a service-life guarantee. Power supplies next — capacitor aging, again accelerated by heat and dust. Hashboards last — sustained heat, thermal cycling from restarts, humidity and corrosion. When a machine “loses hashrate,” it is almost never the silicon slowing down; it is a dead board or thermal throttling — a repair question, not aging.

The reframe that saves buyers money: a fan is a wear item — brake pads, not a death sentence. Cheap, standard, swappable. Boards are the expensive death. And the honest band for the whole box: hosting and repair firms converge on roughly 4–7 years of maintained physical life, shortening to 2–3 in hot, dusty or unstable-power rooms. Those are hosting-company and retailer claims — we publish them because they converge, and we label them because they are not lab data.

The economic clock: three forces against every machine

1. Difficulty creep. Your machine holds its rated TH/s for life — and that is exactly the problem. The network keeps adding drawers-pulled-per-second around it, so a fixed hashrate is a shrinking share of every block. Your machine doesn’t slow down; the library speeds up. Competitor content routinely calls this “performance degradation.” It isn’t — it’s share dilution, and the two tiles below are the live receipt. A 200 TH/s example machine currently holds 1 in 4.84 million of the network’s drawer-pulls (repainted live; the dated snapshot holds otherwise). The network’s growth history — and its pullbacks — live at /btc/hashrate.

2. The efficiency march. Each hardware generation lowers the energy cost of the same drawer pull. The dated series, manufacturer specs as archived by industry listings:

The wider band as of 2026-09: mainstream machines run 15–17.5 J/TH, flagship air sits around 13.5, and announced hydro flagships reach ~9.5 (all manufacturer specs). Why J/TH is the survival stat — and how to compute it from any spec sheet — is lesson 2’s whole subject; the worked math lives there. Here the point is one sentence: when the wage falls, the J/TH bar decides who keeps running, and the bar only moves down.

3. Halvings. A pay cut with a block number on it, scheduled years in advance. It gets its own section below, because it is the one force with a countdown attached.

The shutdown price — and why the network self-balances

Put the forces together and you get the concept that decides economic death: the shutdown price (glossary). The wage per drawer pulled falls until, for some machines, the watts cost more than the drawers earn. Where that line sits depends on exactly two numbers: the machine’s J/TH and its owner’s power rate — nothing else. The live wage itself — what one TH/s earns per day — is published and charted at /btc/hashprice; we won’t re-derive it here. And the shutdown math for your model lives in the downside checks on /rigs — per-model numbers stay there, where they update live.

Then comes the half of the story sell-side content skips. When enough machines drop below their line and power off, blocks arrive slower than the ~10-minute target — and difficulty retargets down at the next 2,016-block checkpoint — about two weeks at 10-minute blocks, longer while blocks run slow. The wage partially recovers for every machine still running. Difficulty is not a one-way ratchet: 2026 supplied its own worked example, with third-party estimates showing short-window readings near 1.4 ZH/s early in the year followed by a pullback into the mid-900s of EH/s by late September (third-party figures, dated 2026-09 — short-window hashrate estimates are noisy, which is why the tiles above trust our own node instead). The retailer version — difficulty only ever rises, so buy new hardware now — conveniently forgets the loop, and the loop is what makes the whole system self-balancing rather than a death spiral.

The ~2028 landmine: a pay cut with a block number on it

The block subsidy is 3.125 BTC (as of 2026); at block 1,050,000 it becomes 1.5625 BTC. The block number and the amount are protocol facts. The calendar date is not — it drifts with real block times — so this page counts in the honest unit and treats the date as the estimate it is. Multiplying the blocks remaining by the live block interval puts the cut around February–April 2028 (estimate — drifts with real block times; fallback 2026-09-27).

Honesty discipline, stated once: the subsidy halves on schedule — that part is certain. What the wage does afterward depends on fees, price, and how much hashrate exits, and we predict none of those. “Halving = doom” and “halving = guaranteed pump” are both predictions, and both are banned here. The practical consequence needs no prediction at all: any payback window that crosses block 1,050,000 must survive a halved subsidy — and most payback windows bought today do cross it. Competitor calculators quietly extrapolate today’s wage forward across the cut; now you know to check.

Old iron: the S9 question, as a framework

So is a physically-alive relic economically alive for you? Three inputs decide, and none of them is a verdict we can publish: its J/TH — an S9’s 98 against the ~13.5 flagship-air class is roughly 7× the energy per drawer, and that ratio is physics, not opinion; your electricity rate — the assumption every “still profitable in 2026” listicle buries, because the assumption is the verdict; and whether the heat has a second job — heat you were paying for anyway, or stranded and waste power that bills at nearly nothing. The S9’s own record reads as a case history, hedged accordingly: roughly a four-year mainstream run from 2016 into the 2020 halving era, with survivors a decade on persisting mainly in exactly those unusual-power niches (community-documented deployment history — anecdote-class, and never a claim about any particular unit).

Two levers genuinely stretch economic life, both concept-level: running a machine below rated power improves its J/TH (voltage and frequency scale down together — widely replicated community practice, and aftermarket tuning firmware exists as a category; we recommend no products), which lowers its shutdown price. And selling into the second-hand and heater tail is an exit, not a death — that afterlife is lesson 1’s territory. For the machine you’re actually eyeing: if it has a preset, its downside check on /rigs owns the shutdown math; for an S9 there is no preset — take the dated 98 J/TH figure and your own rate straight to the calculator.

FAQ

How long does an ASIC miner actually last?

Two answers, and the split is the point. Physically: hosting and repair firms converge on roughly 4–7 years with maintenance (their claim, not lab data — shorter in heat and dust). Economically: commonly 2–4 years, but it isn’t a fixed number — it’s decided by the machine’s J/TH, your electricity rate, and the live wage, which is why the real answer comes out of the calculator, not a blog.

Do ASIC miners slow down or lose hashrate over time?

The chips don’t, materially. A healthy S9 pulls the same 13.5 TH/s it did in 2016. What shrinks is its share: the network keeps adding drawers-pulled-per-second around it, so the same hashrate earns a smaller slice. Sudden hashrate loss usually means a dead hashboard or thermal throttling — a repair question, not aging.

What is a shutdown price?

The point where your watts cost more than your drawers earn. It’s different for every machine-owner pair: the machine contributes its J/TH, you contribute your power rate. When the wage falls below that line, rational operators power off — and because difficulty then retargets down within about two weeks, the wage partially recovers for the machines still running. Your model’s line lives in its downside check on /rigs.

Is an old miner like the Antminer S9 still worth running?

Framework, not verdict: an S9 spends about 7× the energy per hash of a 2026-flagship-class machine (98 vs ~13.5 J/TH, manufacturer specs), so at typical power rates the arithmetic is unforgiving. The exceptions are unusual-power situations — free or waste power, or heat you were paying for anyway. Run your actual rate through the calculator before believing anyone’s answer, including ours.

What happens to miners at the 2028 halving?

At block 1,050,000 — currently projected around spring 2028, and that date drifts — the subsidy drops from 3.125 to 1.5625 BTC per block. That’s a scheduled pay cut every machine must plan to survive. What the wage does afterward depends on fees, price, and how much hashrate exits — mechanisms we can teach, outcomes nobody can honestly predict.