What Used ASIC Miners Sold For on eBay: 90 Days of Sold Comps

A used Antminer S19j Pro sold on eBay for a median of $145 over the last 90 days. The same machine, torn down and rebuilt into a quiet 120 V box that does about a quarter of the hashrate, sold for a median of $260. That is an 80% premium for a machine that has been made deliberately worse at mining, and it is the single most consistent thing in this dataset. On a plain S19 the gap is wider still: $150 stock against $375 converted, a 150% premium.

I sell used miners for a living, so I have a side in this. It does not change the arithmetic. If you want hashrate, the conversion market is the worst place on eBay to buy it. If what you actually want is a warm, quiet appliance that pays for some of itself, then you are not buying hashrate and the premium is not irrational — it is just not a mining purchase, and it should not be compared to one.

Below is what everything else sold for, counted the same way, with the sample size next to every figure.

Cite this page. ING Mining, “What Used ASIC Miners Sold For on eBay: 90 Days of Sold Comps”, 16 August 2026. https://ingmining.com/used-asic-miner-sold-prices — Edition 2. Sample: 4,829 mining-hardware sold listings scanned over the twelve months to 14 August 2026, of which 1,161 were single working machines priced in US dollars. The price tables are the last 90 days of that window, 15 May to 14 August 2026: 398 machines, 278 of them graded used. Medians only, sample size stated on every row. Corrections and re-use are welcome — there is a contact form, and I would rather be corrected than quoted wrongly.

Correction, 16 August 2026. Edition 1 of this page, published yesterday, overstated several figures. Our collector stores the sale price as a bare number and the currency only in the raw record, and from May 2026 the feed started returning Brazilian listings. 334 sales priced in reais were counted as dollars, at roughly 5.4× their real value, and every one of them fell inside the 90-day window the price tables use. The S19 XP median was $331 and is $298. The S21 was $1,200 and is $1,005. The S19 XP conversion premium was published as +731% off sixteen sales; fifteen of those sixteen were reais, so the row is gone rather than corrected. Edition 1 said that row looked like one seller rather than a market, which was the right suspicion for the wrong reason. Non-dollar sales are now dropped, not converted — an exchange rate on the sale date would be a number this page invented. The Edition 1 download is still on the server and still wrong; use the Edition 2 files below.

Method, sample and date

These are completed eBay sales, not asking prices. The figures come from our own sold-comps collector, which pulls eBay sold listings for mining hardware and has 927,193 rows in it going back to 2015. This page uses the mining-hardware slice of that: 4,829 sold listings whose title names Bitmain, Antminer, Whatsminer or Avalon and which closed between 15 August 2025 and 14 August 2026. The altcoin section at the bottom is a second, separate query against the same collector and states its own sample.

Every number is a median, never a mean. One $8,000 hydro unit in a bucket of twenty air machines moves a mean and does not move a median, and this data has exactly that kind of contamination in it. Nothing is modelled, smoothed or carried forward.

Of the 4,829 listings scanned, 1,161 survived to be counted. What went out, and why:

Two classification rules are worth stating because getting them wrong changes the answer. An S19 XP+ is not an S19 XP — 169 TH/s against 141 TH/s, and the obvious pattern match puts them in one bucket. Over the full twelve months the XP+ ran a median of $470 across 16 sales against $418 for the XP; merging them lifts the XP figure to $433. Likewise hydro and immersion units are pulled out of every air bucket before anything is counted. A wet machine is a different product from the air machine that shares its number.

“Used” here means the seller graded it pre-owned, open box, refurbished or remanufactured. Prices are the item price and exclude shipping; 776 of the 1,161 sales shipped free, and the median charge on the rest was $50.

Download the data. Every figure in the tables that follow is in one file — CSV, 22 KB or JSON, 72 KB — 253 rows, one per model per window, with the sample size, median, quartiles and full range on each. Both come out of the same query that renders the tables below, so the file and the page cannot disagree. It also keeps the buckets with fewer than five sales that the tables drop, so you can see what was left out instead of taking my word for it. Re-use it with attribution (CC BY 4.0). If you quote a figure, quote the edition: that filename is dated and I will not overwrite it when I re-run this next month. The altcoin section near the bottom of this page is a second query and is not in this file yet; it goes in next month’s. The Edition 1 file is still at its own URL and is still wrong; it was left in place rather than silently rewritten under anyone who had already cited it.

Median used sale price, last 90 days

Window is 15 May to 14 August 2026: 398 machines, 278 of them graded used. Those are the last 90 days of the 1,161 above, not a separate sample. Only models with ten or more used sales appear here. The p25–p75 band is the middle half of the sales — where a wide band means something, I say so underneath.

Edition 1 said the S19 XP and the S21 carried wide upper bands because those buckets mixed bare machines with machines sold with a PSU or an enclosure. That was a reasonable guess and it was wrong. The bands were reais. Corrected, the XP’s middle half is $264–$355 and the S21’s is $939–$1,495, and both buckets are about as tight as you would expect. The row that still has a genuinely wide band is the S19 Pro at $158–$479 on 13 sales, which is a small bucket and a real spread. Use the median.

The L9 top of $3,200 is one sale, on 11 in the window. That is what a full range is for — it shows you the outlier exists rather than burying it. In Edition 1 that same top read $12,966, which was R$12,966.

Too thin to publish

These models had between five and nine used sales in the window. I am listing them because a sample size of six is information, and because leaving them out silently would make the coverage look better than it is. Do not price a machine off these.

Twelve months of monthly medians

Monthly medians for the four models with enough monthly volume to be worth plotting. Months with fewer than five sales are dropped rather than printed thin.

August 2026 is a part month — it stops on the 14th, which is why its samples are small.

The S19 XP is the story here. It sold for a median of $1,112 in August 2025 and $290 in August 2026: a 74% fall in twelve months, on a machine that is still one of the most efficient air-cooled units you can buy used. The S19j Pro fell from $260 to $147 over ten months and has sat between $136 and $147 for the last two, which is roughly where a 96 TH/s machine stops being worth shipping. Below about $120 the freight and the bench time cost more than the machine.

The Avalon Q runs the other way, and it is the only model here that does. It rose from $827 in April to a plateau around $1,058–$1,199. That is a factory-built quiet home miner rather than a datacentre machine, and it is priced by people who want the appliance — the same demand that produces the conversion premium below, except Canaan builds it properly instead of someone rebuilding an S19 in a shed.

The altcoin machines nobody publishes comps for

Everything above is Bitmain, Whatsminer and Avalon, because that is what most people ask about. It is not what I get the most confused emails about. Those are the Scrypt and Ethash boxes — someone bought a VolcMiner or a Mini Doge, it has been sitting in a closet for a year, and there is no public sold price for it anywhere. So I pointed the same query at the other six brands: IceRiver, Goldshell, ElphaPex, Jasminer, VolcMiner and Fluminer.

Separate query, separate sample, same method. 1,079 sold listings whose title names one of those brands, closed in the same twelve months. 493 were a single working machine priced in dollars: 373 were parts, 150 were priced in reais, 27 were multi-unit lots, 17 named no model I track, and 18 were not miners at all. That last number earns a sentence. “IceRiver” also matches KastKing IceRiver fishing gloves, and “Nano 3” matches an iPod. If you build one of these yourself you need a not-a-miner filter, or your Kaspa comps will have gloves in them.

The window below is four months, not twelve, and that is a limit in our records rather than a choice. eBay’s condition grade is missing on 369 of the 403 rows we hold for these brands before April 2026 — the collector was keeping the price and the title for this slice but not the grade. Only 206 of the 493 machines carry a grade at all, and every one of those closed between 18 April and 13 August 2026. So these are four-month used medians. I would rather say that than print a twelve-month header over four months of data.

Two things in that table are worth more than the prices themselves.

The VolcMiner D1 Mini Pre and the ElphaPex DG Home 1 are nearly the same machine on paper — 2.2 GH/s at 500 W against 2.1 GH/s at 630 W — and the VolcMiner sells for $100 more. Buy the VolcMiner. That 130 W difference is about 1,140 kWh a year, which is $114 at ten cents, so the price gap pays for itself inside the first year and the efficiency edge is free after that. If your power is cheap enough that $114 a year is noise, take whichever one you find first.

The Goldshell Mini Doge III is the bad buy in this group, and it is the one I get asked about most, because it is small and it is cheap. It draws the same 500 W as the VolcMiner and does 810 MH/s against 2.2 GH/s: 2.7 times less work for an identical power bill. Per gigahash it is the most expensive of the three at roughly $351, against $318 for the VolcMiner and $286 for the DG Home. It is a fine thing to own if what you want is a quiet box that mines a little Doge on your desk. It is a bad way to buy Scrypt hashrate, and the $284 sticker makes it look like the opposite.

One caveat on that row. Sellers use “III”, “3” and “III+” interchangeably — several titles in the bucket say “MINI DOGE 3 III + PLUS” in one breath — so it mixes the 700 MH/s unit and the 810 MH/s Plus. I did not split them because the titles do not reliably tell them apart. Read that median as “a Mini Doge III of some flavour”.

The Jasminer X16-Q mines a different algorithm and does not compare to the other three. 1,950 MH/s of ETChash at 620 W, $600 median, and its middle half sits in a $459–$650 band — tighter than I expected for a machine mining a coin most people have written off.

Too thin to publish, altcoin side

Between five and nine graded used sales each. Same rule as the table further up: this is information about coverage, not a price to trade on.

What is missing here is the entire Kaspa side, and that surprised me. IceRiver is the most-listed brand in this slice — 42 sold listings for the AE1 Lite by itself — and not one KS-series bucket cleared ten graded used sales. The KS5L got to four. The sales are there; the condition grades are not, for the reason above. That should fix itself next month, because everything closing now carries a grade.

If you are shopping rather than valuing, these machines live on two shelves here: Litecoin and Dogecoin miners and Kaspa, Aleo, Dash and CKB miners.

The conversion premium

This is the finding I did not expect to be this large or this consistent. A “home miner” conversion is a datacentre ASIC pulled apart and rebuilt: underclocked to a fraction of its rated hashrate, rewired for a 120 V wall outlet, quieted down, usually given WiFi. It does less work and uses less power. It sells for more.

There was a fourth row here in Edition 1: a converted S19 XP at $2,750 against $331 stock, a +731% premium. I flagged it as looking like one seller rather than a market, because the titles repeated almost word for word. The real answer was worse. Fifteen of those sixteen sales were priced in reais, and once they go the bucket has one sale left in it, which is not a number. So the row is deleted rather than corrected. The three rows above are spread across many sellers and they survived the fix unchanged in shape — the conversion premium is real, it is just not 731%.

Here is what that means in hashrate. A converted S19j Pro at $260 doing roughly 28 TH/s is $9.29 per TH/s. A stock one at $145 doing 96 TH/s is $1.51 per TH/s. You are paying six times as much per unit of work for the privilege of the machine being quiet. That is a real thing to want. Just know that is what you are buying, because nobody selling one puts it that way.

If you want the quiet machine, buy the one that was designed quiet. If you want hashrate, buy the stock machine and deal with the noise properly — there is a whole page on ASIC miner cooling and another on power requirements, and both are cheaper than a 150% premium.

What the low end of “brand new” is

One more thing worth publishing, because it will save somebody money. Among Avalon Q sales graded brand new, the median was $489 across 17 sales, against $1,100 for used across 35. New selling at 44% of used is not a bargain and it is not a market inefficiency.

Seventeen of those sales cluster between $259 and $490, from listings with near-identical titles, and the used floor across every seller in the same window was $879. A current-generation machine does not sell new for a third of what a worn one fetches. I am not going to assert what each of those listings actually was, because I did not buy one. I will say the pattern is the reason this page reports used prices separately from new ones, and the reason I would not use a raw eBay “sold” average for anything.

That is the general lesson from the whole exercise. Of 4,829 mining-hardware sold listings, only 1,161 were a single working machine of an identifiable model priced in dollars. A naive average over eBay sold comps is averaging hashboards, fan simulators, pallets and bait into the same number.

How to use this, and what it will not tell you

These medians tell you what a machine changed hands for on eBay, and that is genuinely useful for two things: sanity-checking a price you are about to pay, and setting a floor when you sell. It is what I use before quoting on a fleet somebody wants us to buy.

What it does not tell you is whether the machine works. An eBay sold price bundles the machine, the seller’s testing, the warranty if there is one, the PSU if it is included and the risk that none of it is true. Our own used miners are priced above some of these medians on purpose, because they are tested on our bench and they ship with what the listing says they ship with. Whether that is worth the difference is your call, and the honest way to make it is to know the market number first. That is why this page exists.

It also will not tell you whether the machine is worth running. Price is one input; power cost and difficulty are the others, and the buying guide works through the running-cost side properly.

I plan to re-run this monthly and republish with the date and the new sample size. Every figure will be re-derived, not carried forward — when a number moves, that is the point of the page.

Nick Squires runs ING Mining, one of the larger used ASIC dealers in the US. Twenty years in IT before this, and he still builds the PSUs and runs part of the floor on solar. If a number here is wrong, tell him — the method is published so it can be checked.