There is no market price for a used ASIC miner. There’s a band, and it’s enormous. I pulled every sale of one byte-identical listing title out of our sold-listings database — “Bitmain Antminer S19 XP 141Th ASIC Bitcoin Miner ASIC 3030W - Used - BOX” — and found 22 completed sales between $175.50 and $455.00. Same title. Same model. Same stated condition. Same photos, presumably. The top sale cleared the bottom one by 159%.
So “what’s my S19 worth?” has no answer. The useful question when you sell bitcoin miners on eBay is which end of that band you land on — and that’s decided by a few things you control: what your title claims, whether you disclose the firmware, and whether you’re selling a machine or a box of parts. Written from doing it, not reading about it.
What used bitcoin miners actually sell for on eBay in 2026
Start with the numbers, because most guides skip them. These come from our own sold-comp database — 864,000 completed eBay listings, collected through 17 July 2026 — queried on 27 July 2026. “Whole machines” means the parts, boards, PSUs and multi-unit lots have been stripped out of the set:
| Model | Rows matched | Average across all matches | Whole machines found | Median, whole machines | p10 – p90 |
|---|---|---|---|---|---|
| Antminer S19 XP (141TH) | 1,743 | $685.43 | 45 of 75 on-model rows | $340.00 | $222.50 – $579.99 |
| Antminer S19j / S19j Pro | 1,712 | $694.75 | 19 of 74 on-model rows | $122.50 | $1.00 – $249.00 |
Those medians are low, and they should be. This is several-generations-old hardware in 2026 — an S19j Pro is not a machine you buy to earn on residential power any more, and the price reflects it. If you’re holding a shed of them waiting for the number to come back, that’s a position, not a plan.
Those two medians are the headline, not the whole picture. Split the S19 XP row further — full-power units against the derated 120-volt home conversions hiding inside it — and the price per terahash swings by more than 6×. That split, and why pricing an ASIC by hashrate doesn’t work, is worked through model by model in the companion piece.
Look at the gap between column three and column five, though. For the S19j, the average across everything the keyword matched is $694.75 and the median actual machine is $122.50 — or $139.99 once the rows are read by hand. Either way the headline number is five to six times the real one. For the S19 XP it’s about 2×. Price your miner off the first number and you’ll list it at four times what anyone will pay, watch it sit for six months, and conclude the market is dead.
Most of what “matches” your model isn’t a miner
Here’s why the averages blow up. Of the 74 sold rows that genuinely mentioned an S19j, only 19 were whole machines. The other 55 were:
- 27 control boards — mostly the Amlogic A113D C76/C87, which fits the S19j Pro, S19 XP, S21 and L9, so it matches everybody’s search. Seven of those sold at $38–$42.
- 11 fans and cooling parts
- 7 hashboards
- 4 power supplies
- 4 multi-unit lots
- 2 signal ribbon cables, at $8.99 and $10.99
And three of the 19 survivors aren’t machines either. Re-running this query on 30 July and reading all 19 rows by hand: two are “Bitmain Antminer Diagnosis/Quote: S19j pro and S19k pro models only” at $1.00 — a repair service, missed because the title says “Diagnosis” and not “service” — and one is an “ALTAIR s19j Board, Untested, Sold As-Is” at $9.99. Those three are the whole reason the 10th percentile above reads $1.00. Drop them, add back one real machine the filter wrongly binned (below), and you get 17 actual S19j Pros, median $139.99, $64.55 to $350.00 — 14% above what the automated cut reports.
Three out of four “sold S19j” rows are not a machine. They’re genuinely S19j parts — which is exactly why the average is worthless, because nothing filters them out for you. This is the same failure I’ve written up before in more detail — why keyword-matched sold averages are usually measuring the wrong product — and miners are the worst category for it I’ve found, because a single machine has four separately-saleable components that all carry the model number in their titles.
Searching harder makes it worse, not better
The instinct is to add words. Don’t. Our database has 1,743 rows for antminer s19. Add one word — antminer s19 pro — and it returns 64,095 matches, of which zero in a 500-row sample contained the string “s19” at all. The sample was a Fieldline Pro hunting backpack, a Nike Elite Pro basketball backpack, an Osprey Tempest Pro daypack and a Titan PRO can cooler. The word “pro” is so common it swamped the model number entirely.
Same trap on cooling. Search s19 hyd and you get Garvee SAC18-2-S19 mini-split air conditioners, because “S19” is a Garvee model number too. It is genuinely easier to get a clean comp set with a shorter query and your own eyes on the titles.
Your title is doing your pricing, whether you meant it to or not
This is the part I didn’t expect to find. Take that one model — the 141TH S19 XP — and group the sold rows by the exact title the seller used:
| Title ends with… | Sales | Range | Median |
|---|---|---|---|
| “…141TH 141T SHA-256 BTC Miner — WORKING“ | 3 | $469 – $499 | $469.00 |
| “…141Th ASIC Bitcoin Miner ASIC 3030W - Used - BOX“ | 22 | $175.50 – $455.00 | $325.50 |
| “…141Th ASIC Bitcoin Miner READ DESCRIPTION“ | 3 | $250 – $350 | $250.00 |
Two of those groups have three sales in them, so treat the exact figures as a hint rather than a law. But the direction is not subtle and the mechanism is obvious: “READ DESCRIPTION” in a title is a hedge, and buyers price hedges down. It tells them there’s a problem before they’ve clicked. “WORKING” is a claim, it’s checkable, and it sat at the top of the range.
The practical rule I’d take from this: put the strongest claim you can actually defend into the title, and put the caveats in the description where they belong. If the machine runs, say it runs. If it doesn’t, say what’s wrong — specifically. One of the sales in that same pull was titled “Bitmain Antminer S19 XP 141TH 141T SHA-256 BTC Miner — BAD CONTROL BOARD” and it cleared $349, more than the median working-condition sale of the “Used - BOX” group. A named, specific fault priced better than a vague one, because the buyer could work out exactly what they were buying and what it’d cost to fix.
Firmware: name it, and say what it’s tuned to
Custom firmware is standard on the secondhand S19 market and most listings handle it badly. In a 500-row pull on antminer vnish, only 16 rows mention any tune or custom firmware in the title at all. Five of those survive the parts filter: median $139.49, range $99.99 to $150.00.
The sixth is instructive, because the filter is wrong about it the other way. “1 x Bitmain Antminer S19j Pro 104TH Vnish/Braiins OS BTC ASIC Miner” sold for $185.99 and got binned as a multi-unit lot: the rule that catches “10x” and “4 pcs” also catches “1 x”. It’s one machine, and it’s the dearest tuned unit in the pull, so the automated cut drops the top of its own range.
Five or six sales is not enough to claim a firmware premium and I’m not going to. The bigger sample now exists — the firmware and resale value study pulls every sale on record naming VNish, Braiins or LuxOS and settles what the token is actually worth. What I will say here is that firmware changes what the buyer physically receives, so it belongs in the title regardless of what it does to price:
A stock S19 and a VNiSH-tuned S19 running a low-power 2800W profile are different products. So are a 110TH unit and the same unit boosted to 120TH, which draws more, runs hotter and stresses the hashboards harder. A buyer planning around a power budget needs to know which one is arriving. Three things to state:
- Which firmware — VNish, Braiins OS, LuxOS, or stock. Name it.
- What profile it’s on right now, in TH/s and watts. “Boosted to 120TH” and “low power 2800W tune” both tell a buyer something actionable.
- Whether it’s flashable back to stock, and whether any developer fee or licence travels with it.
If you genuinely don’t know what’s on the machine — bought it in a lot, never powered it — say that. “Firmware unknown, sold as-is” costs you less than a buyer discovering a locked bootloader after delivery.
Hydro units: eBay can’t price them, and probably shouldn’t sell them
I went looking for water-cooled comps and came back with almost nothing, which is itself the finding. A 500-row pull on antminer hydro returned 2,197 matches. Only 320 of the sampled rows contained the word “antminer”. Exactly one row contained both “antminer” and a hydro token — at $35, and it wasn’t a whole machine. The rest of the result set was Hydro Flask coolers, Gregory 3D Hydro reservoirs and USWE Hydro backpacks.
So: I cannot give you a hydro price from this data, and anyone who gives you one off eBay comps is guessing. That’s not a data problem so much as a market fact — a hydro S19 needs a facility with a cooling loop, and the people who own those buy in quantity through mining-specific channels and direct deals, not through consumer marketplace search. List one on eBay by all means, but price it from a direct quote and treat the listing as advertising.
Part it out, or sell it whole?
The component prices from the same pull, all real completed sales:
- S19 XP hashboard, tested working — $179 (“BHB56801 V1 Tested Working Stable Hashrate”)
- S19 XP hashboard, parts only — $45
- Control board (A113D, fits S19j Pro / S19 XP / S21 / L9) — $38 to $42, seven sales
- PSU for S21 / S19j XP — three sales, $235 to $279
Do the arithmetic on a whole S19 XP that medians $340: three working hashboards at $179 is $537 before you’ve sold the control board or the PSU. Parting out wins on paper by a wide margin.
It loses on everything else. You’ve turned one transaction into five, each with its own shipping, its own packaging, its own buyer, its own return risk — and the $45 “parts only” hashboard row shows what one board that doesn’t test clean does to the maths. My position, from doing both: a working machine sells whole. Part out the dead ones, where the hashboards are the only value left and the alternative is a “for parts, untested” listing that buyers correctly assume is worthless.
Condition: eBay offers six grades and your miner qualifies for two
Pick the condition last and you will pick it wrong. On a used ASIC only two entries in that dropdown do anything at all: Used and For parts or not working. The graded ones you have seen elsewhere on eBay — Very Good, Good, Acceptable — do not exist in Business & Industrial, and a listing that tries to use one is refused outright.
That is not read off a help page. It is what eBay’s API handed back to our own listing software: “The provided condition id is invalid for the selected primary category id.” The item was a Fanuc servo amplifier, the model grading it had settled on “used — good”, the app sent condition 5000, and the publish failed. Not the condition field — the whole listing, after every other field had been filled in correctly.
| eBay condition ID | What eBay calls it | Where you can actually use it |
|---|---|---|
1000 |
New | Every category. |
1500 |
New other (see details) | Every category. |
2000 |
Certified Refurbished | Eligibility-gated. A private seller will not have it. |
2750 |
Like New | Eligibility-gated and category-limited. |
3000 |
Used | Every category. Search results print it as Pre-Owned. |
4000 |
Very Good | Books, Movies & TV, Music and Video Games only. |
5000 |
Good | Books, Movies & TV, Music and Video Games only. |
6000 |
Acceptable | Books, Movies & TV, Music and Video Games only. |
7000 |
For parts or not working | Every category. |
EbayService.cs on 15 August 2026. The three media-only grades are the ones that bounce a Business & Industrial listing.2000 and 2750 look like the honest answer for a machine that is barely used, and neither is available to you. Both are eligibility-gated: eBay decides who may list a refurbished item, and a private seller shifting one miner is not on that list. So a dropdown that appears to offer six used grades offers you four IDs in total, and if the machine has ever been powered on, two.
The market never uses the other words either
Our sold-comp database now holds 927,193 completed eBay listings across 2,370 keywords, with sold dates running through 14 August 2026. Queried on 15 August 2026, the conditions “Very Good”, “Good”, “Acceptable” and “Like New” appear on exactly zero rows. So does “Used”, for a different reason: eBay’s API calls 3000 Used, and eBay’s search results print it as Pre-Owned.
One caveat, because it changes what the number means. That harvest is deliberately hardware-heavy — miners, servers, laptops, industrial spares — so a zero there is strong evidence about the categories we sell in and no evidence at all about Books. What it does rule out is the possibility that industrial sellers quietly use graded conditions and we simply have not noticed.
The graded-sounding values that do turn up are a different mechanism entirely. 26,505 rows carry “Excellent – Refurbished”, “Very Good – Refurbished”, “Good – Refurbished” or “Certified – Refurbished”. That is eBay Refurbished, a separate programme with its own application, its own warranty obligation and its own approved-seller list — not a grade you can simply select. Of the 3,634 rows whose titles name an Antminer, 102 carry a plain “Refurbished” and one reads “Refurbished in excellent condition”. Not one carries a programme tier.
What miner sellers actually pick, and what each one fetches
| Condition | Sales | Share | Median | 25th – 75th percentile |
|---|---|---|---|---|
| Pre-Owned | 509 | 73.1% | $300.00 | $158.00 – $825.00 |
| Brand New | 73 | 10.5% | $1,299.00 | $497.37 – $4,100.00 |
| Refurbished | 48 | 6.9% | $357.50 | $148.50 – $1,241.75 |
| New (Other) | 44 | 6.3% | $1,890.72 | $467.98 – $3,048.19 |
| Parts Only | 21 | 3.0% | $230.00 | $32.21 – $390.00 |
Six values across the whole market, and three quarters of every sale sits in one of them. Note what “Brand New” costs, because it is the opposite of what a bargain hunter expects: a genuinely new, sealed machine medians $1,299 against $300 for a used one. Nobody is dumping new stock. The cheap end of a new-condition search is not machines at all, which is a trap worth its own paragraph in the method section below.
“For parts” does not cut your price. It removes your ceiling.
This is the finding I did not expect. Declaring a machine For parts or not working moves the median from $300 to $230 — it keeps 77% of the price. Sellers treat that box as though it halves what they get, and on the middle of the range it does not come close.
The damage is at the top. The 75th percentile falls from $825 to $390, which is 47% — and 40% of used sales cleared $390, the entire upper quartile of the parts range. A parts listing is not a discount. It is a cap. You keep the median and you give away every outcome above it, because the buyers who pay the top of the range are buying a machine that runs and they filter for one.
Twenty-one sales. That is the whole population of whole-machine parts listings in the database, and it is too few to treat the $230 as a price you can plan around. Read the shape, not the decimal: the gap at the median is small and the gap at the top is large. That much holds however you cut it.
Which gives a rule that agrees with the part-it-out section above. If the machine hashes, do not declare it For parts to pre-empt an argument — you are paying the ceiling for a fault you do not have. If it does not hash, declare it, and be specific about what is dead. The $349 sale headlined BAD CONTROL BOARD earlier on this page is what a named fault is worth against a vague one.
The grade still reaches the buyer, through a different field
Collapsing everything to Used loses nothing, because the dropdown was never where the detail belonged. eBay’s ConditionDescription is a free-text field that sits directly under the condition on the listing page, and it is the one buyers of used hardware actually read:
Tested 72 hours on stock firmware, hashing 94.6 TH/s at 3,250 W measured at the wall. Two fans replaced March 2026. Chassis scuffed. PSU not included — this is the miner only.
No condition grade carries any of that. “Very Good” would have told a buyer nothing they could plan a power budget around, which is the same argument as the firmware section: the specific claim is the one that sells.
What the software does about it
Since 14 August the app picks the condition ID per category rather than per grade. Outside the media categories every used grade collapses to 3000, which every category accepts, and the grade the model actually assigned is written into ConditionDescription so the buyer still gets it. Media categories keep 4000, 5000 and 6000, decided from the category name. “Like new” now maps to 3000 too, for the same reason 2750 is unusable.
It is not in the download yet, and you should know that before you go looking for it. The installer on this site is the 13 August build, version 2.5.0.0. The condition fix was committed on 14 August, so it lands in the next one. What is on the site today will still send a graded ID in a non-media category and still get the listing bounced. If that happens to you, set the condition to Used by hand and put the grade in the description — which is what the fix does automatically.
Shipping, briefly, because it’s where the money leaks
An ASIC is a dense metal brick with fragile plastic fan assemblies bolted to both ends, and it will be handled by people who do not know that. Ship the fans protected — foam or a separate compartment — and double-box anything going more than a couple of states. State clearly whether the PSU is included; on the S19 line it’s a separate unit and “miner” means different things to different buyers. And price the shipping into your floor before you list, not after: on a $126 median machine, a $40 shipping mistake is a third of the sale.
For multiples, quote freight rather than eating parcel rates per unit. Both of the listings in the screenshot above are multi-quantity for exactly that reason. Whether multiples should go out as one lot listing or as separate ones is a narrower question than it looks: on four sold S19 lots the per-unit discount was about 8%, and eBay’s fee table pays you $2.80 to consolidate eight machines. The lot-versus-singles arithmetic is worked in full, including the one threshold where it stops being trivial.
The rest of it — Bitmain’s own packed weights, what 171 sold single machines were actually charged to ship, and the point where parcel stops and freight starts — is in how to ship heavy equipment.
What you owe when the miner sells for less than you paid
Every whole S19j Pro in the sample above closed between $64.55 and $350.00. If you bought yours to mine with rather than to flip, you paid a good deal more than that, which makes the sale a loss — and a loss is deductible. The catch is that a loss you can’t evidence isn’t a loss. With no record of what the machine cost, the IRS starts from the sale price and taxes all of it, as though the hardware arrived free. Our own earnings screen does the opposite with that same sale — no recorded cost, no claimed profit, because overstating profit is the one direction that screen must never be wrong in. The cautious number upstairs is the expensive one on the return.
What not knowing costs
Self-employment tax is 15.3% on 92.35% of net profit — 14.13 cents on the dollar — and half of it comes off your income before the bracket applies. At a 12% federal bracket the two stack to 14.13 + 12 × (1 − 14.13 ÷ 200) = 25.3¢ per dollar. Run it backwards: one miner that sold for $840 with nothing on record about what you paid carries about $212 of tax that need not exist. A bank line, or an invoice from whoever sold you the pallet, takes it straight off. Nothing else in this article pays that well per hour.
Miners make this harder than most categories, in four ways. The purchase is old — machines being liquidated now were bought years ago, often by wire, often from a vendor who stopped answering email. They arrived in lots: one payment for thirty machines, no per-unit figure anywhere, so you split the basis yourself and have to say how. They may already be written off — depreciate a machine when you buy it and your basis drops by what you claimed, possibly to nothing; any tool that answers that one for you is guessing. And parting one out splits the basis. The section above says part out the dead ones; do that and one $285 machine becomes three hashboard sales and a control-board sale, with $285 to allocate across four of them. Put it all against the first board and the other three look like pure profit.
The 1099-K is not your profit
eBay reports gross payments. Take the $340 median S19 XP with $45 of buyer-paid shipping on it: eBay collected $385, and $385 is what lands on the 1099-K. Off that comes the final value fee — miners sit under Coins & Paper Money at 13.25%, charged on the shipping too, so $51.01, plus the $0.40 per-order fee — then the label, then whatever the machine cost you. Two things catch sellers out. The sales tax eBay collected was never yours and never reached you, so it isn’t part of your receipts. And a fully refunded sale still appears in gross payments; it comes back out on the returns line rather than vanishing.
That refund rule is slipperier than it reads, and it produced a real bug when we built this. A full refund voids the sale but not the stock — the miner came back and is on the shelf, so you can’t also deduct what you paid for it. Our first pass did exactly that: a $285 miner refunded in full claimed $180 of cost of goods on a unit we still owned. A partial refund is the opposite case — the buyer kept the machine, it’s a price cut, and the cost still counts.
Pay in the IRS’s quarters, not calendar quarters
The estimated-tax windows are not three months each. They’re 3, 2, 3 and 4:
| Sales made | Payment due | Window |
|---|---|---|
| 1 January – 31 March | 15 April | 3 months |
| 1 April – 31 May | 15 June | 2 months |
| 1 June – 31 August | 15 September | 3 months |
| 1 September – 31 December | 15 January, next year | 4 months |
Four equal quarters put June’s sales in the wrong payment, which is the ordinary way a seller earns an underpayment penalty on a year they had the cash for the whole time.
What the software does, and what it won’t
A Tax Pack screen shipped on 30 July 2026, sitting directly under the earnings screen because the two are halves of one sentence: a Schedule C by line number — 1, 2, 3, 4, 5, 7, 10, 22, 27a, 28, 31 — with the arithmetic behind each line printed next to it, the cost-basis gap at the top because it’s the one figure that makes the total go down, a set-aside rather than a bill, the four real payment dates, a 1099-K reconciliation, and two CSVs: the summary an accountant reads and the ledger they check it against. Read-only. It files nothing.
What it won’t do matters more. Federal only, and most states charge income tax on top. One flat bracket, ignoring the standard deduction, the 20% QBI deduction most resellers can take, other income and filing status — a set-aside that errs large, not a return. And it prints no 1099-K threshold: the federal figure has moved three times since 2022 and several states set their own, so telling a seller they’re under one is worse than sending them to go and look. Vendoo, List Perfectly, ZIK and Crosslist don’t produce a Schedule C either, and none of them pretend to — it isn’t their job. Nor is any of this a substitute for an accountant. It hands the accountant a traceable number instead of a shoebox.
Everything in this section is US federal, describes how the software computes a set-aside, and is not tax advice. I sell miners for a living; I don’t prepare returns.
Where the software fits
Every number on this page came out of the same comp database that ING Listing Engine prices against, and the parts/lot filter described here is applied before it estimates anything — that filter exists because of the S19j problem above, not as a general nicety. If you already have miners listed and they’ve been sitting, the Inventory Health scan reprices live listings against the cleaned comp set and tells you which ones the market has walked away from. If you’re buying rather than selling, the auction ceiling calculator works out the most you can pay for a unit and still clear a margin — and on miners it very often answers “nothing here”.
Download ING Listing Engine free and run a comp lookup on a machine you already own. It’s in open beta — no paywall, no card, no trial clock. The setup guide covers what the two required connections actually cost you in time and money.
Related reading: Terapeak alternatives and how to clean a comp set goes deeper on the filtering method, and building an eBay listing from one photo covers the identification step — which matters here, because telling an S19 from an S19j Pro from a photo is exactly the sort of thing sellers get wrong before any pricing happens.
Common questions
What is a used Antminer S19 worth in 2026? In our sold data, whole S19 XP machines median $340 with a 10th-to-90th-percentile range of $222.50 to $579.99, and S19j/S19j Pro machines median $122.50 as the filter cuts them, $139.99 once the three non-machines that slipped through are read out by hand. Ignore any average that hasn’t had control boards, hashboards, PSUs and lots stripped out of it — for the S19j that average is $694.75, five times the real median.
Should I disclose custom firmware when selling a miner? Yes, in the title. Name the firmware, state the profile it’s currently running in TH/s and watts, and say whether it can be flashed back to stock. A boosted unit is a different product from a stock one and a buyer planning a power budget needs to know which is arriving.
Is it better to part out a bitcoin miner or sell it whole? Whole, if it runs. Three working S19 XP hashboards at $179 each beat a $340 whole-unit median on paper, but you’re then running five transactions with five lots of shipping and five return risks. Part out machines that are already dead — a working hashboard sells for $179 where a “parts only” one sells for $45.
How do I price a hydro-cooled miner? Not from eBay comps. In a 500-row sample of antminer hydro exactly one row contained both “antminer” and a hydro term, and it wasn’t a whole machine — the rest were Hydro Flask coolers and hydration backpacks. Get a direct quote from a facility buyer instead.
Does saying “READ DESCRIPTION” in the title hurt? The evidence is thin — three sales — but it grouped at the bottom of the range while “WORKING” grouped at the top. A specific named fault did better than a vague warning: a listing headlined “BAD CONTROL BOARD” still cleared $349. Put the defensible claim in the title and the detail in the description.
Do I owe tax on a miner I sold at a loss? You owe tax on profit, and if the machine cost more than it sold for there isn’t any — but only if you can prove what you paid. With no cost on record the IRS taxes the full sale price as though the hardware were free. At a 12% federal bracket that is about 25.3 cents on the dollar, so a single $840 sale with no recorded cost carries roughly $212 of tax that a bank line or an invoice would remove.
What condition should I list a used bitcoin miner as? Used, unless it does not hash. eBay’s graded conditions — Very Good, Good, Acceptable — exist only in the media categories, and sending one in Business & Industrial gets the whole listing rejected with “the provided condition id is invalid for the selected primary category id”. That leaves Used and For parts or not working. Across 696 whole-machine sales, declaring For parts kept 77% of the median price ($300 to $230) but only 47% of the 75th percentile ($825 to $390) — so it costs you the top of the range, not the middle. Put the detail that a grade would have carried into the condition description instead.
Why do miner sold-price averages come out so high? Because a single machine has four separately-saleable parts that all carry the model number. Of 74 sold rows genuinely mentioning an S19j, 54 were control boards, fans, hashboards, PSUs or lots. The keyword average is measuring the parts bin.
Method, so you can check it
Figures were produced on 27 July 2026, and every one of them re-run and re-checked on 30 July 2026, against our hosted sold-comps API (864,000 completed eBay listings, collected through 17 July 2026). For each model we requested the aggregate over all matches, then pulled a sample capped at 500 rows, kept the rows whose titles contained the model token, and split off anything matching a parts/lot rule covering hashboards, control boards, power supplies, fans, cooling parts, cables, brackets, repair and hosting services, and lot/pallet/bundle/unit-count language. Medians and percentiles are over the surviving set. Title-group figures are exact-string matches on the full listing title.
Added 15 August 2026, and it is a correction to the table at the top of this page. Re-running the cleaning on the current harvest of 927,193 listings showed the parts filter has a third hole, and this one is large. It removes “board only”, “motherboard”, “mainboard” and “logic board” — but not “control board”, which is the single most commonly sold Antminer part there is. Of 895 rows the filter passed as whole machines, 199 were control boards or hashboards, median $56.61. They drag every condition median down, and they are why a naive cut makes “Brand New” look like a $76 machine when a genuinely new sealed miner medians $1,299. The condition figures in the section above are computed with control board, hashboard and a list of accessories — silencers, chassis, power cords, cooling plates, SD cards — removed on top of the published filter. The pricing table at the top of this page has not been recomputed; it is the July study, stated as of when it ran. Treat any keyword average that has not had boards stripped out of it as measuring the parts bin.
A gap in the data, stated rather than papered over. eBay stopped exposing the condition on the sold-listing results our collector reads at the start of August 2026. Rows gathered through July carry a condition 98.0% of the time; rows gathered in August carry one 22.2% of the time. On the Antminer set specifically it is 92.0% against 13.5%. Every condition figure in the section above is therefore computed only on rows that actually carry a condition — 696 whole machines out of 3,634 Antminer rows — and skews toward sales collected before August. It is not a sample of the recent market and it is not presented as one. Prices in that section should be read as the shape of the condition gap, not as this week’s going rate.
The filter is a word rule and it’s wrong in both directions. Re-running every query on 30 July 2026 turned up a clean example of each. It lets parts through when the title doesn’t use the word it looks for: it matches “hashboard” but not “hashboards”, so “Bitmain Antminer S19XP 141Th ASIC Bitcoin Miner TWO HASHBOARDS” at $152.50 counted as a whole machine in the S19 XP set — take it out and that median moves from $340.00 to $345.00. The same gap put two $1.00 “Diagnosis/Quote” repair services and a bare “Board” into the S19j machines. And it bins real machines: the quantity rule that catches “10x” and “4 pcs” also catches “1 x”, which threw out a $185.99 single Vnish-tuned S19j Pro. Two earlier cases still stand — a $18.97 “Replacement Mop Holders for H40/M40/M30S Series Robot Vacuum” survived it in a Whatsminer M30S pull, and a row whose eBay condition field said “Parts Only” survived because its title didn’t say so.
The other limit is the source: a scrape of completed listings, not eBay’s ledger, so no accepted Best Offer prices, and Terapeak’s three years of first-party data is deeper. Every cut here was run for this page — none of it is recycled from the S19 analysis on our listing-software page or the S21 analysis on the Terapeak page. The tables report what the documented filter produces, so you can reproduce them — and where reading the rows by hand gives a different answer, this page prints both rather than quietly publishing the nicer one.
Filed under Reselling Hardware · Tagged ASIC miners, Sold comps, Dirty data · Browse all guides by topic