Repair Guide · Antminer S19 · Whatsminer M30 · Hashboard repair · Cost per TH
Used vs new ASIC miner: real cost per TH over 18 months
Most used-vs-new comparisons stop at the sticker price per terahash. That number is real, but it is not the cost of ownership. Over 18 months the deciding variables are failure rate, repair cost, downtime, and the electricity price at which the machine stops earning.
- 18
- Month ownership horizon
- ~10x
- Typical used vs new $/TH gap in lot quantities
- 3
- Hashboards per unit to budget for
The used-vs-new argument usually gets settled with one number: dollars per terahash on the invoice. That number matters, and the gap is wider than most guides admit — used S19-class hardware in pallet lots has been quoted under $2 per TH/s against roughly $17 per TH/s for a new S21 Pro. But invoice price is the entry ticket, not the cost of ownership. Over an 18-month horizon the machine that looked cheap can cost more per TH than the one that looked expensive, and the reason is almost never the purchase price.
The two numbers that actually decide it
There are two figures that set the outcome, and they pull in opposite directions.
- Capital cost per TH/s — what you pay up front, divided by nameplate hashrate. Used hardware wins this decisively, especially in lot quantities.
- The power price at which the machine stops earning — the rate above which revenue no longer covers electricity. Used silicon loses this, because older chips run fewer joules per terahash.
Everything else — payback period, depreciation exposure, pilot-site economics — is a consequence of those two. A farm that only looks at the first number is buying a machine that may lose money every day it runs. A farm that only looks at the second is overpaying for efficiency it may not need.
Run both numbers before you commit
A used S19-class unit has been quoted a shutdown threshold near 3.9¢/kWh at current network conditions. Buy fifty of them at any price and plug them in at 6¢/kWh, and you own fifty units that bleed cash daily. The capital saving does not rescue that math.
Building the 18-month model
The honest comparison is cost per TH/s delivered over 18 months, not cost per TH/s purchased. That means adding four things to the invoice price: repairs, downtime, spare inventory, and the residual value at month 18.
Capital $/TH/s
High — reference ~$17 for S21 Pro class
Low — reference under $2 for S19 class
Warranty coverage
Yes, typically covers early failures
None — you absorb parts and labor
Expected board-level repairs over 18 mo
Low, mostly warranty claims
At least one per unit; budget per hashboard
Downtime per failure
Swap-and-RMA, days of shipping
Swap-and-repair, hours if spares on site
Residual value at month 18
Depreciates fast as new silicon ships
Already near floor; limited further loss
Shutdown power price
Higher threshold, more headroom
Lower threshold — the binding constraint
The two rows that flip the decision are the repair row and the shutdown row. A used unit with no spare boards and no on-site repair capability turns a $2/TH capital advantage into a week of lost revenue the first time a hashboard drops. A new unit at a site with 7¢ power has headroom the used unit does not.
Where used hardware actually fails
Used batches fail in predictable places, and knowing them lets you price the risk instead of guessing at it.
Hashboard-level failures
The dominant failure is at the ASIC die and its power delivery. Look for shorted dies, unstable or missing voltage domains on the buck stages, and damaged harness connectors — the 6-pin and 18-pin links between control board and hashboard are a common weak point after repeated hot swaps. Cracked solder on the heatsink-side power path shows up as boards that pass a cold test and drop out under load.
Control board and PSU
Control boards fail less often but fail expensively when they do — corrupted firmware, dead Ethernet PHYs, failed SD-card boot paths. PSUs on used units are frequently the original units, running near end of capacitor life. A PSU that sags under load produces exactly the symptom operators misdiagnose as a bad hashboard.
Die and domain faults
Shorted ASICs, missing buck voltage domains, boards that fail only under thermal load.
Harness and connector wear
6-pin and 18-pin links degraded by repeated swaps and heat cycling.
PSU end-of-life
Sagging rails under load mimic hashboard faults and waste diagnostic time.
Firmware and control board
Corrupted images, dead PHYs, failed boot media on used control boards.
The downtime multiplier
Repair cost is only half of a failure's cost. The other half is the hours the board sits dead. This is where the used-vs-new comparison is won or lost operationally, not financially.
Step 1
Keep spare hashboards on site
One spare per 10–15 running units lets you swap a failed board in minutes and repair the pulled board on a bench, off the critical path.
Step 2
Stage a known-good PSU
A single spare PSU removes the most common misdiagnosis and lets you isolate power faults in one swap.
Step 3
Bench-test pulled boards under load
Cold-passing boards that drop under thermal load need a load test, not a visual inspection. Log which voltage domains fail.
Step 4
Track failures per batch
A batch failing faster than one board per unit per 18 months is a signal to renegotiate or return, not to keep repairing.
A farm with spares and a bench absorbs a hashboard failure in hours. A farm without them absorbs it in days of shipping and RMA queue. Over 18 months that difference in uptime routinely exceeds the capital gap between used and new.
The 18-month verdict, by site
- Cheap power, in-house repair capability, spare boards on site: used hardware wins on total cost per TH/s over 18 months. The capital gap is large and the downtime penalty is small.
- Cheap power, no repair capability: used hardware is a coin flip. Price in a board-level repair per unit plus shipping time, and it may still win — but only if the batch is clean.
- Mid-to-high power: the shutdown threshold decides before the invoice does. Run the power-price math first; if the used unit is near its threshold, the capital saving is irrelevant.
- Fleet expansion under time pressure: new hardware's warranty and uptime are worth the premium when every day of delay costs more than the efficiency gap.
The practical rule: compute cost per TH/s delivered over 18 months, not purchased. Add one board-level repair per used unit, add the downtime your site actually experiences, and check the shutdown power price against your real rate. If the used unit still wins after all three, buy it — and buy spare boards with it.
Frequently asked questions
Is a used ASIC always cheaper per TH over 18 months?
No. It is cheaper on capital almost every time, but the gap narrows or inverts once you add hashboard repairs, downtime, and the higher electricity price at which older silicon stops earning. The crossover depends on your power rate and how much of the repair work you can do in-house.
What electricity price makes a used S19-class machine unprofitable?
Reference figures put the shutdown threshold for a used S19-class unit near 3.9 cents/kWh at current network conditions. Above that rate the machine can lose money every day regardless of how little you paid for it. Recalculate with your own rate and today's network difficulty before committing.
Which failures should I budget for on used hashboards?
The common chip-level failures are ASIC die shorts, missing or unstable voltage domains on the buck stages, damaged 6-pin or 18-pin harness connectors, and cracked solder on the heatsink-side power path. Budget for at least one board-level repair per unit across 18 months, more if the batch ran hot or dusty.
Does buying new remove the repair risk?
It reduces but does not remove it. New units still fail, usually inside the warranty window, but you pay a large capital premium per terahash for that coverage. The right answer depends on whether you have on-site repair capability and spare boards to swap.
Repair & maintenance
Get Your Miners Back to Full Hashrate
Chip-level diagnostics, component-level repair, and burn-in validation — on-site across the USA, no shipping required.