JPMorgan analysts put the cost of producing one bitcoin at about $86,000, as reported by The Block on Friday 20 November from a research note the bank has not made public. Bitcoin traded near $55,760 at 05:45 UTC on Saturday 21 November. Set side by side, those two dated figures leave the market price about $30,200, or 35%, below the estimated bitcoin blockchain and earns a reward, which makes rewriting history very expensive.">mining cost. The gap is arithmetic on the two readings, not a number from the bank.
Revenue per machine is about to fall for a second reason. The network’s next difficulty adjustment is due at block 977,760, around Thursday 26 November, and mempool.space estimated it at +9.9% when read at 05:54 UTC on Saturday. A rise of that size means each machine earns less bitcoin for the same work, at a moment when each bitcoin buys fewer dollars of electricity.
One small filing shows what that pressure looks like on a balance sheet. PowerCompute, a Nasdaq-listed miner that runs less than 0.1% of the network’s computing power, told the SEC on 16 November that its lender had sold 267.4 pledged bitcoin to clear a loan of about $21.89 million, and that doubt about its survival remains. It is one company, not a trend.
What JPMorgan’s $86,000 bitcoin mining cost estimate says
According to The Block, the note was published on Wednesday 18 November by a team led by Nikolaos Panigirtzoglou. Bitcoin was then near $65,000, about $21,000 under the bank’s central estimate. By Friday the price had slipped below $60,000 for the first time since July, and The Block put the gap at more than $26,000.
The analysts wrote that miners had probably added to the selling as the gap grew, The Block reported. Operators with higher costs “may have sold reserves”, in the outlet’s account of the note, to raise cash, keep their machines running or pay for a move into artificial intelligence hosting.
JPMorgan also described how the gap could close without the price rising. In the words quoted by The Block: “we believe the gap could narrow as the hashrate declines and mining difficulty adjusts”. If loss-making miners switch off, the network lowers its difficulty and the cost per coin falls for those still running.
A model figure, with no published method
The note is not public, and the other outlets that carried it on Friday give the same figures as The Block. The Block’s article does not explain how the $86,000 is calculated: no electricity price, no machine efficiency, no word on whether hosting income from AI is counted. And it attributes no hashrate, hashprice or difficulty level to the bank.
A production cost is an average across very different operators. A miner with new machines and cheap power produces well below it; one with old hardware produces well above. When The Block reported JPMorgan’s previous estimate on 24 September, of around $85,000, it wrote that the analysts said production cost had historically acted as a “soft floor” for the price, and that bitcoin had spent 280 days below the estimate before moving back above it. Those are the bank’s words of September, as The Block gave them.
Hashprice: what a miner earns for a day of computing power
Miners watch a simpler number than production cost. Hashprice is the revenue expected from one petahash per second (PH/s) of computing power over a day. It moves with three things: the bitcoin price, the difficulty and the transaction fees paid in each block.
Luxor’s hashprice index stood at $28.58 per PH/s per day on 20 November, down from $30.45 on 16 November, according to readings compiled by CryptoSlate. That is a fall of about 6% in four days. The home page of Luxor’s Hashrate Index displayed $26.44 when read at about 05:55 UTC on Saturday. That is a live ticker with no time stamp of its own, so it is not a like-for-like reading with the daily figures above.
Fees are no help. Luxor counted a weekly average of 0.0229 BTC in fees per block on 16 November, next to a fixed subsidy of 3.125 BTC, CryptoSlate reported. Fees were under 1% of what a block paid.
CryptoSlate turned that into an electricity bill with a model that counts power alone, leaving out smart contract holding reserves of two or more tokens, supplied by users, that a decentralized exchange uses to execute trades automatically at a price set by a formula.">pool fees, hosting and the cost of the hardware. At $28.58, a machine that burns 30 joules per terahash covers its electricity up to 3.97 cents per kilowatt-hour. If difficulty rises 8.47% and nothing else moves, that ceiling drops to 3.66 cents. A newer machine at 13.5 joules per terahash goes from 8.82 cents to 8.13 cents. In the outlet’s example, at a power price of 3.8 cents the older machine pays its way today and stops doing so after the adjustment.
| Measure | Figure | Date and source |
|---|---|---|
| Estimated production cost of one bitcoin | About $86,000 | Note of 18 November, JPMorgan as reported by The Block |
| Bitcoin price | About $55,760 | 21 November, 05:45 UTC, market data snapshot |
| Gap between the two | About $30,200 (35%) | Arithmetic on the two lines above |
| Hashprice | $28.58 per PH/s per day | 20 November, Luxor via CryptoSlate |
| Previous difficulty adjustment | -9.13% | 14 November, 05:02 UTC, mempool.space |
| Next adjustment, estimate | +9.9% | Read 21 November, 05:54 UTC, mempool.space |
| Network hashrate | About 1,065 EH/s | Read 21 November, 05:55 UTC, mempool.space |

Difficulty is set to rise about 10% around 26 November
Difficulty is the setting that keeps bitcoin blocks about ten minutes apart. Every 2,016 blocks, the software compares the time those blocks took with the two weeks they should have taken and corrects the setting. Our guide to hashrate and the difficulty adjustment walks through the formula.
The last correction went down. On 14 November at 05:02 UTC, difficulty fell 9.13%, from 148.36 trillion to 134.82 trillion, the cut we had previewed in our piece on the 14 November difficulty drop. It made every machine still running about 10% more productive in bitcoin terms.
The next one points up. At 05:54 UTC on Saturday, with the chain at block 976,856, mempool.space showed 904 blocks left before the adjustment at block 977,760, an expected date of 26 November at about 23:10 UTC and an estimated change of +9.86%. Applied to the 134.82 trillion in force, that gives roughly 148.1 trillion, close to the level before the cut.
Other trackers are lower. MineOrNot, cited by CryptoSlate on 20 November, estimated +8.47%, to about 146.24 trillion, around 27 November. All of these figures move with every block until the adjustment itself.
Why difficulty can rise while the price falls
The estimate is a look in the rear-view mirror. It measures only how fast blocks have arrived since 14 November, and they have come every 9 minutes 7 seconds on average, according to mempool.space, against a target of 10 minutes. Faster blocks mean more computing power is working than the current setting assumes. The formula does not know the bitcoin price.
So the price drop of the past two days has had little time to show up. If miners switch machines off during the 904 blocks that remain, blocks will slow and the estimate will shrink. If they do not, difficulty rises by about a tenth and the weakest machines face lower revenue on both counts. Mempool.space put the network’s hashrate at about 1,065 exahashes per second (EH/s) on Saturday morning.
Friday’s fall in mining shares: CoinDesk points to the AI sell-off
Shares of listed miners fell hard on Friday 20 November. CoinDesk, which tracked the session, attributed the fall to the sell-off in artificial intelligence stocks, not to mining revenue. Nvidia closed nearly 14% lower, a semiconductor index lost 10% and the Nasdaq 100 ended the day down 5.5%, according to its live coverage.
Around midday in New York, CoinDesk reported Hut 8 down nearly 15%, Core Scientific and Bitdeer down more than 13%, MARA about 11% and Riot 8.5%. After the close it listed MARA down 18.6%, IREN 18% and Hut 8 17.7%, and said miners were among the worst hit because many have expanded into AI computing.
Many of these companies now sell power and data-centre space to AI customers, and their shares move with that business, a shift visible in the miner stocks held by the AI fund Situational Awareness.
PowerCompute: the lender sold 267.4 pledged bitcoin
PowerCompute is the former LM Funding America. It took its new name and the ticker PWCM on 22 July 2026, nine days after a 1-for-25 reverse stock split. Its quarterly report for the period to 30 September, filed with the SEC on 16 November, lists about 8,300 mining machines with a capacity of about 0.88 EH/s, hosted at sites that include Calumet, Oklahoma, and Columbus, Mississippi. Divided by the network’s 1,065 EH/s, that is roughly 0.08%.
The filing describes two loans from the crypto lender Arch during the summer. On 25 August, PowerCompute’s mining subsidiary borrowed about $21.89 million for 30 days at 6.5% a year, secured by 307 bitcoin, and used about $18.1 million of it to repay a first Arch loan dated 3 August. The loan carried a collar, a set of agreed prices that limited both sides’ exposure to bitcoin: a floor of $71,112 for the first 30-day period, a ceiling of $75,000 and a barrier of $93,500.
It lasted one period. “On September 24, 2026, the Company fully repaid and terminated the Second Arch Loan Facility,” the report says. Arch sold approximately 267.4 of the 307 pledged bitcoin to cover about $21.89 million of principal, $118.6 thousand of accrued interest and a $440.1 thousand loss on disposal, then returned the remaining 39.6 bitcoin or so to the company.
Who decided to sell matters for how the episode is read. CoinDesk’s headline of 20 November says the miner sold the coins; the filing says Arch did. Arch co-founder Himanshu Sahay told CoinDesk: “The sale was directed by PowerCompute. It reflected a change in their corporate strategy, not a market event.” The report gives no price per coin for the sale.
The debt is gone, the going-concern warning is not
Clearing the loan did not fix the accounts. At 30 September, PowerCompute had about $653,000 in cash and 63.8 bitcoin, valued in the report at about $5.3 million, or $83,781 per coin. At $55,760, the price at 05:45 UTC on Saturday, 63.8 coins come to about $3.6 million; the report does not give the holding after 30 September.
The company reports cumulative net losses of about $48.9 million from 1 January 2024 to 30 September 2026, and $9.98 million of cash used in operations over nine months. The third quarter shows net income of about $388,000, but only after a $6.68 million fair-value gain on bitcoin it was owed; the operating result was a loss of about $0.9 million. One loan remains, $1.25 million owed to Brown Family Enterprises, extended to 31 December 2027.
Management’s conclusion is written in the standard accounting language: “substantial doubt about the Company’s ability to continue as a going concern has not been alleviated.” Its plan rests on selling new shares through an at-the-market programme of up to $75 million signed with Maxim Group in March, and the report says that programme depends on the company staying within Nasdaq’s listing rules.
CoinDesk reported that the shares closed near 40 cents on Thursday 19 November, a market value of roughly $1 million, and that shareholders approved authority for another reverse split of between 1-for-2 and 1-for-10 on Tuesday 17 November.
The company is still mining, and still buying. On 26 October it ordered 600 Antminer S21++ machines for about $1.0 million, paid in November and due in service in December, according to the report. Its AI business so far is a single graphics processor bought in July, which brought in about $1,000.

Larger miners are selling coins or changing business
PowerCompute is tiny, and nothing in its filing ties its troubles to the JPMorgan note. The bank’s point about miners selling reserves concerns much larger holders, and those have their own disclosures. Hyperscale Data sold its last bitcoin and still warned about funding its AI build, as we reported in our piece on its cash warning. MARA is raising debt to buy a power plant, a deal described in our report on its $185 million Long Ridge notes.
Neither the JPMorgan note, as reported, nor the hashprice readings give a count of machines switched off since the price fell. The difficulty estimate, for now, says the opposite: more computing power than two weeks ago.
The dates that will show whether miners are switching off
The adjustment at block 977,760 is the first test. Mempool.space expected it on 26 November at about 23:10 UTC when read on Saturday morning; other trackers say 27 November. A final figure well under the current +9.9% would mean hashrate left the network in the days before it. A figure close to it would mean miners kept running through the drop.
Thursday 26 November is also Thanksgiving in the United States, and stock markets there are closed that day. MARA has said it expects to close its Long Ridge purchase on 27 November.
JPMorgan’s figure moves as well: The Block reported an estimate of around $85,000 in September and $86,000 in November.
