Bitcoin Mining Margins in 2026: One Industry, Ten Margins
Miners just posted their best quarter since 2017, but an October difficulty jump is already clawing it back. And there is no single mining margin, only a fleet spanning an order of magnitude.
Bitcoin miners just finished their best stretch in years. The third quarter of 2026 handed Bitcoin a gain of roughly 43%, its strongest third quarter since 2017, pulled along by billions of dollars in spot exchange-traded-fund inflows, according to FinanceFeeds. October then delivered the sector’s richest revenue month since January, around $1.12 billion, per Shattered. Then, on October 16, the network did what it always does when money comes back: it got harder. Mining difficulty jumped 5.92% to about 140.58 trillion, the third-largest increase of the year, as hashrate climbed back above one zettahash per second, news.Bitcoin.com reported.
That is the paradox of mining margins in 2026. The recovery is real, but it does not reach every miner the same way, and the network competes a slice of it back almost as fast as the price delivers it. Dig into the quarterly filings and one fact jumps out: there is no single mining margin. In the second quarter, the cost to produce one Bitcoin ran from about $44,000 at the cheapest public operator to more than $350,000 at the most expensive, an order of magnitude apart, in the same business, mining the same coin on the same network. This is a guide to how that gap opens, why it is widening rather than closing, and how to read it.
There is no single mining margin
The phrase mining margin suggests one number, a profit per coin that rises and falls with the Bitcoin price. The reality is a distribution that spans an order of magnitude. In its Q2 2026 Bitcoin Mining Report, published September 15, CoinShares put the weighted-average cash cost to mine one Bitcoin across the listed miners at roughly $75,500. With Bitcoin ending June near $58,400, that average sat well above the coin’s price, which is why CoinShares described the listed sector as, in its words, below cash breakeven in aggregate.
But the average hides everything interesting. Within the same report, American Bitcoin produced a coin for about $43,851 in cash terms while Cipher Mining spent about $215,723; on an all-in basis that includes depreciation, the spread stretched from roughly $76,529 to $353,168. Only four miners (American Bitcoin, Bitdeer, HIVE and IREN) mined below the Bitcoin price and stayed cash-profitable through a brutal quarter. The rest did not.
Three structural variables drive that dispersion: the price a miner pays for electricity, the efficiency of its machines, and the business model it runs on top of the hardware. A fourth, the pivot into artificial-intelligence hosting, is now pulling the fastest movers onto a different track entirely. Understand those four axes and the scoreboard stops looking random.
Gross, cash, and all-in: the three margins
Before comparing miners, it helps to be precise about which margin is on the table, because the same company can look profitable on one measure and deep in the red on another in the very same quarter.
Gross margin is revenue minus the direct cost of the electricity that produced it. It is the fastest read on whether the machines should be switched on at all; if power costs more than the coins are worth, a rational miner curtails.
Cash cost adds the rest of the operating bill (site operations or hosting, pool fees, labor and overhead) but leaves out depreciation. It answers whether the operation funds itself today.
All-in cost folds in depreciation of the ASIC fleet and the data-center build, plus items like stock-based compensation. It is the closest thing to an accountant’s view of whether the business creates value after the hardware that wears out is paid for.
Riot Platforms’ second quarter is a clean worked example. The company reported a cost to mine one Bitcoin of $49,912 excluding depreciation against a production value of $71,667 per coin, so every coin was cash-profitable, according to its Q2 results. Fold in depreciation and the cost jumped to $90,631, or 126.5% of production value, turning the same coins into a loss on paper. Both numbers are true. Which one matters depends on the question you are asking.
| Margin measure | What it includes | Riot, Q2 2026 (per BTC) | Verdict |
|---|---|---|---|
| Production value (revenue) | Bitcoin mined at the price realized | $71,667 | Top line |
| Cash cost (excludes depreciation) | Power, site ops or hosting, pool fees, labor, overhead | $49,912 (69.6% of revenue) | Cash-profitable |
| All-in cost (includes depreciation) | Cash cost plus ASIC and build-out depreciation | $90,631 (126.5% of revenue) | Loss on paper |
Hashprice: the revenue side in one number
On the revenue side, the industry compresses everything into a single figure called hashprice, a term coined by Luxor’s Hashrate Index. It is the expected daily revenue per unit of hashrate, quoted in US dollars per petahash per second per day (PH/day). The recipe is simple: take the roughly 144 blocks mined per day, multiply by the block reward (the 3.125 BTC subsidy plus transaction fees), multiply by the Bitcoin price, and divide by total network hashrate.
Plug in mid-October numbers (a 3.125 BTC subsidy, fees running near 0.42% of the reward, Bitcoin around $82,500, and network hashrate above 1,000 EH/s) and you land near the $36.75 per PH/day that trackers reported on October 16. A miner multiplies that by its own hashrate to size its top line: a 10 EH/s fleet at $36.75 grosses about $367,500 a day before it pays for a single kilowatt-hour.
Two features make hashprice the master gauge of revenue. First, it collapses four moving parts (price, subsidy, fees, and difficulty-through-hashrate) into one comparable dollar figure. Second, because fees are a rounding error in 2026 and the subsidy is fixed until 2028, near-term hashprice is essentially the Bitcoin price divided by how much competing hardware is pointed at the network. That second term is where the trouble starts.
The recovery the network is already clawing back
Hashprice is a tug-of-war between price and difficulty, and in October 2026 both pulled hard. The Q3 rally, and a Bitcoin price that spent the autumn in the $80,000s, lifted the dollar value of every block. But the reward for a stronger market is more competition: idled rigs switched back on, new machines shipped, and hashrate climbed back over one zettahash per second. On October 16 the protocol repriced the contest, raising difficulty 5.92% to about 140.58 trillion, the third-largest upward adjustment of 2026 and the highest reading since March. Difficulty now sits only about 10% below its November 2025 record near 156 trillion, per CoinWarz.
The effect on margins is direct. Difficulty is the denominator in the hashprice formula, so a 5.92% increase mechanically shaves roughly that much off revenue per unit of hashrate before a miner sees a cent. Hashprice, which had touched about $40 early in the month, printed $36.75 on the day of the retarget. The rally had not failed; the network was simply taking its cut.
This is the self-correcting loop at the heart of mining economics. JPMorgan’s Nikolaos Panigirtzoglou pegged the network’s all-in production cost near $78,000 earlier in 2026 and described the mechanism to The Block: “When bitcoin trades below its production cost, higher-cost miners power down, the hashrate declines, and difficulty adjusts lower.” In October 2026 the loop ran in reverse. Price firmed, high-cost rigs came back, hashrate rose, and difficulty adjusted higher, dragging the whole fleet’s revenue back toward the cost of the marginal, least-efficient miner still willing to run. Whether the cycle’s price low is already behind us is the question we took up in our look at the halving cycle math.
The cost stack: where a miner’s dollar goes
If hashprice sets the top line, the cost stack decides what survives to the bottom one. For a self-mining operator it breaks down roughly as follows, in order of size for most fleets.
- Electricity. Almost always the largest single line, often 50% to 80% of cash cost, and it scales directly with how inefficient the machines are.
- Site operations or hosting. For miners that rent space rather than own it, a hosting fee (usually a power markup plus a management charge) replaces part of the electricity line and removes the operator’s control over when to curtail.
- Pool fees. The cut taken by the mining pool that smooths revenue, typically between 0% and 4% depending on the payout scheme.
- Overhead and labor. Salaries, security, maintenance, insurance, and corporate costs.
- Depreciation. Non-cash, but the single biggest reason a cash-profitable quarter can still be a loss on paper, because ASICs lose value quickly and depreciation cannot be switched off by curtailing.
The first two lines explain most of the dispersion between miners’ cash costs; the last explains most of the gap between a miner’s cash cost and its all-in cost. The next three sections take the structural drivers in turn.
Axis one: power and geography
Electricity is the first and largest axis of dispersion. Because power is the dominant variable cost, a few cents per kilowatt-hour separate a fleet that prints money from one that bleeds it. The break-even is easy to express: a machine’s break-even power price in dollars per kWh equals its hashprice divided by 24 times its efficiency in joules per terahash. At a hashprice of $36.75, a cutting-edge sub-10 J/TH machine can pay up to about $0.16/kWh and still cover its electricity, while a five-year-old 30 J/TH rig breaks even near $0.05/kWh and is underwater almost everywhere that power is not effectively free.
Geography is how miners chase the right side of that line. The cheapest industrial power in large US mining hubs runs well under $0.05/kWh, and operators tune their sites around demand-response programs, in Texas especially, where curtailing during peak grid stress earns credits that offset the power bill. Riot booked $10.054 million in power-curtailment credits in the second quarter alone and $31.077 million across the first half, credits that flatter its reported cash cost relative to a miner with no such program. That single mechanism helps explain why one quarter can produce very different cost figures for operators running identical hardware.
| Machine | Efficiency (J/TH) | Break-even power at ~$40 hashprice | Break-even power at $36.75 hashprice |
|---|---|---|---|
| Bitdeer SEALMINER A4 Ultra (hydro) | 9.45 | $0.177 | $0.162 |
| Antminer S23 Hydro | 9.5 | $0.176 | $0.161 |
| Antminer S23 (air) | 11 | $0.152 | $0.139 |
| Antminer S21 XP (air) | 13.5 | $0.124 | $0.113 |
| Antminer S21 Pro | 15 | $0.111 | $0.102 |
| WhatsMiner M60S | 18.5 | $0.090 | $0.083 |
| Antminer S19 XP | 21.5 | $0.078 | $0.071 |
| Antminer S19j Pro | 30.5 | $0.055 | $0.050 |
The table also shows what the October difficulty jump did to the margin of safety. As hashprice slid from roughly $40 to $36.75, every machine’s break-even power price fell with it, pushing the least efficient rigs closer to the line even though the Bitcoin price barely moved.
Axis two: chip efficiency and the sub-10 J/TH frontier
The second axis is the hardware itself, measured in joules per terahash: the lower the number, the less electricity a machine burns for the same work. The 2026 frontier has crossed a milestone that sounded far off only two years ago, with the best machines now running under 10 J/TH. Bitdeer’s SEALMINER A4 Ultra Hydro is rated at 9.45 J/TH and Bitmain’s Antminer S23 Hydro at about 9.5 J/TH, with air-cooled flagships like the S23 near 11 J/TH, according to Hashrate Index. Set that against the previous generation’s Antminer S21 Pro at 15 J/TH or the legacy S19j Pro near 30 J/TH, and the efficiency gap between a new and an old fleet is roughly three to one.
That gap is the difference between life and death in a high-difficulty market. At $36.75 hashprice, the efficiency column in the break-even table doubles as a survival ranking: a 9.5 J/TH fleet clears its power bill almost anywhere, a 21.5 J/TH fleet needs power below about seven cents, and a 30 J/TH fleet needs power that is practically free.
The catch is that efficiency has to be bought, repeatedly. ASICs do not become more efficient with age; the network does, as rivals deploy newer chips. A rig’s economic value therefore decays even if it never breaks, and that decay shows up as the depreciation line that turns cash profits into accounting losses. Chasing the sub-10 J/TH frontier means continuous reinvestment, which is why the capital structure of a miner (how it funds the next fleet) matters as much as the power contract. It is also why so many miners are now eyeing a different use for all that power.
Axis three: the business model and the AI exit
The third axis, and the one reshaping the industry fastest, is what a company does with its megawatts. Through 2026 the most valuable thing a miner owns has quietly shifted from its ASICs to its power interconnect, and artificial-intelligence hosting pays far more for a megawatt than mining does. CoinShares estimates AI and high-performance-computing work generates annualized profit of roughly $1.5 million per megawatt, against about $0.5 million for mining, as reported by CoinGeek. Needham and Company analyst John Todaro put the same point plainly to CoinGeek: “The revenue per megawatt and EBITDA margins are far higher for HPC and AI colocation than for mining.”
The signature deal is Riot’s. On August 10 the company announced a 20-year lease of 191 megawatts of data-center capacity at its Rockdale, Texas, site to a counterparty it described only as a leading frontier AI lab, worth about $9.1 billion over the initial term and up to $16.1 billion if extension options are exercised. Bloomberg, citing people familiar with the matter, reported the tenant is Anthropic, though neither company has confirmed it. CEO Jason Les called the lease “a defining moment in our evolution into a leading developer of large-scale data centers.” Combined with an earlier lease to AMD, Riot now has 241 megawatts and roughly $9.8 billion of contracted revenue pointed at AI rather than hashes.
MARA’s chief executive Fred Thiel frames the shift as existential. Mining, he has said, “is a zero-sum game. As more people add capacity, it gets harder for everybody else. Margins compress, and the floor is your energy cost.” His conclusion: “By 2028, you’ll either be a power generator, be owned by one, or be partnered with one. The days of being a miner plugged into the grid are numbered.” James Butterfill, CoinShares’ head of research, has projected that listed miners could draw as much as 70% of their revenue from AI by the end of 2026, up from roughly 30%, he told The Block.
A note of caution belongs here. The AI pivot is mostly a promise, not yet a cash flow. CoinShares counts more than $100 billion of disclosed AI and HPC backlog across the sector supporting only about $1.1 billion of annualized revenue so far, with roughly 550 megawatts actually billing against more than four gigawatts contracted. The demand for that compute is real enough, it is the same hunger for verifiable machine work that is spawning entirely new proving markets, but for most miners the AI margin is still a slide, not a bank statement.
The public-miner scoreboard, and why no two costs match
Put the public miners side by side and the dispersion is stark. CoinShares’ Q2 2026 figures, calculated on one consistent methodology, run from American Bitcoin’s sub-$44,000 cash cost to Cipher’s $215,723, with all-in costs stretching past $350,000 at the high end.
| Miner | Cash cost, ex-tax ($/BTC) | All-in cost ($/BTC) |
|---|---|---|
| American Bitcoin (ABTC) | $43,851 | $76,529 |
| HIVE | $60,192 | $116,739 |
| Bitdeer (BTDR) | $61,051 | $89,737 |
| IREN | $64,667 | $141,593 |
| CleanSpark (CLSK) | $71,995 | $135,691 |
| Hut 8 (HUT) | $73,197 | $130,113 |
| Riot (RIOT) | $74,955 | $113,499 |
| MARA (MARA) | $86,126 | $163,866 |
| TeraWulf (WULF) | $174,683 | $327,553 |
| Cipher Mining (CIFR) | $215,723 | $353,168 |
Two cautions make the table honest. First, the highest numbers (TeraWulf and Cipher) are distorted by the AI transition itself: these companies are pouring power and capital into HPC build-outs while mining relatively little Bitcoin, so their cost per coin balloons and is not comparable to a pure miner’s. Second, and more important, a company’s cost to mine is not a standardized figure. CoinShares puts Riot’s Q2 cash cost at $74,955; Riot’s own filing, which nets its large curtailment credits against the power bill, reports $49,912. Same company, same quarter, two numbers that differ by half.
The lesson for anyone reading a miner’s disclosures is to check the definition before trusting the number. MARA, for instance, headlines a cost per Bitcoin of $38,690 in its Q2 shareholder letter, but that is a narrow cost-of-energy figure for its owned sites at roughly four-cent power, not a cash cost comparable to Riot’s. On CoinShares’ uniform all-in basis, MARA’s cost per coin was $163,866. Neither is wrong; they measure different things. Reading a mining stock well means knowing which is which.
Fees, the security budget, and the 2028 cliff
Every margin discussed so far rests on a revenue line that is more than 99% block subsidy. Transaction fees, which some expect to carry Bitcoin’s security budget one day, are running near 0.42% of the block reward in late 2026, a rounding error. The 3.125 BTC subsidy does the rest of the work, and it is scheduled to halve to 1.5625 BTC at block 1,050,000, expected around April 2028, per the CoinWarz halving countdown.
That cliff is the structural shadow over every miner’s margin. All else equal, halving the subsidy halves hashprice, which means the break-even power prices in the tables above roughly halve too, and the gap between efficient and inefficient fleets becomes even more decisive. Thiel has warned that without sustained Bitcoin price growth of 50% or more a year, “the math gets very tough after 2028.” In other words, every structural axis in this piece, power, efficiency, and business model, gets squeezed harder the moment the subsidy steps down, and the miners already closest to break-even are the ones with the least room to absorb it.
Margins are leverage on the Bitcoin price
Mining is one of the most leveraged bets on the Bitcoin price available, because its costs are largely fixed in dollars while its revenue floats with the coin. A move in the price swings margins by a multiple, and the swing is largest for the least efficient fleets sitting closest to break-even. That cuts both ways, as 2026 has shown in miniature.
The autumn has been jumpy. A year after its October 6 record of about $126,080, Bitcoin spent much of October 2026 stalled in the $80,000s, trading around $82,500 in the middle of the month, roughly a third below that high, per CoinGecko. The backdrop is the one-year mark of the October 10, 2025 crash, when a threat of 100% US tariffs on Chinese goods triggered a record cascade of about $19 billion in liquidations and knocked Bitcoin from roughly $122,600 to $104,800 in a single session, as CoinDesk recounted on the anniversary. A year on, Bitcoin has been trading largely on its own signals, a story we followed as it traded alone into the anniversary.
For miners, the near-term swing factor is the Federal Reserve, which meets on October 27 and 28. Markets lean heavily toward no change in rates, but with a stalled price and difficulty climbing, any macro disappointment lands straight on the thinnest margins. Where the price goes from here, and how much rides on the Fed’s December decision, is the subject of our latest price-target roundup.
How the SEC frames mining
For US readers, one regulatory point clears up a common confusion: Bitcoin mining itself is not a securities activity. In a statement dated March 20, 2025, the SEC’s Division of Corporation Finance said that proof-of-work protocol mining, whether solo or through a pool, does not involve the offer or sale of securities and therefore does not require registration (sec.gov). Independent legal analyses, including one from Dechert, read the statement the same way.
What that means for margins is indirect but real. Mining is treated as an industrial activity, not a regulated financial service, so the margin disclosures that matter to investors arrive through ordinary securities filings, the quarterly reports, current reports, and shareholder letters that public miners file with the SEC, rather than through any mining-specific regime. The discipline those filings impose, and their inconsistent cost definitions, are exactly why the non-comparability problem above exists. The derivatives that let miners hedge hashprice, by contrast, are commodities overseen by the CFTC, a separate lane entirely.
What to watch into year-end
Three things will shape mining margins through the end of 2026. The first is the next difficulty retarget: estimates for late October ranged as high as a further double-digit increase, though such projections swing with block times and should be treated as soft. If difficulty keeps climbing toward its November 2025 record while price stays rangebound, hashprice compresses again and the squeeze returns through the back door.
The second is Q3 earnings season. Neither Riot nor MARA had reported third-quarter results as of mid-October; when they do, the figure to watch is not the average cost to mine but the dispersion, and especially how much of each company’s story has shifted from hashes to AI leases. CoinShares has argued that even a Bitcoin price recovery is unlikely to reverse the AI transition, because the economics of HPC are simply better.
The third is the macro tape, where a rangebound Bitcoin keeps the whole fleet on a knife-edge between the efficient and the marginal. For the broader market context into which these margins resolve, see our 2026 year-end outlook. The through-line is dispersion. The Q3 recovery and October’s revenue bounce did not lift all miners equally, and the October 16 difficulty jump is already competing the windfall back, hardest for the operators who can least afford it. There is no single mining margin in 2026, and the gap between the best and the worst is the real story.
Frequently Asked Questions
What is a good profit margin for Bitcoin mining in 2026?
There is no single benchmark, because costs vary enormously. In the second quarter of 2026 the cheapest public miner produced a Bitcoin for about $44,000 while the most expensive spent more than $200,000 in cash terms, according to CoinShares. A useful rule of thumb: a fleet is cash-profitable when its cost to mine excluding depreciation sits below the Bitcoin price, and genuinely profitable when its all-in cost, including depreciation, does too. In October 2026, with Bitcoin around $82,500, many efficient miners cleared the first bar but not the second.
What is hashprice and what is it now?
Hashprice is the expected daily revenue per unit of hashrate, quoted in dollars per petahash per second per day. It combines the Bitcoin price, the block subsidy, transaction fees, and network difficulty into one number. On October 16, 2026, trackers reported it at about $36.75 per PH/day, down from roughly $40 earlier in the month after a 5.92% jump in mining difficulty.
Why did mining difficulty jump in October 2026?
Difficulty rose 5.92% on October 16, 2026, to about 140.58 trillion, its third-largest increase of the year, because hashrate climbed back above one zettahash per second as a stronger Bitcoin price drew idled machines back online and new hardware shipped. Difficulty adjusts roughly every two weeks to keep blocks about ten minutes apart, so more competing hardware automatically makes mining harder and trims revenue per unit of hashrate.
Why is AI mentioned so much in Bitcoin mining coverage now?
Because hosting AI and high-performance computing pays far more per megawatt than mining. CoinShares estimates roughly $1.5 million of annualized profit per megawatt for AI work against about $0.5 million for mining. Miners with cheap power and grid connections, like Riot with its multibillion-dollar Rockdale lease, are leasing capacity to AI firms, which changes their margin profile entirely. For most miners, though, the AI revenue is still contracted backlog rather than realized cash.
Is Bitcoin mining regulated as a security in the United States?
No. The SEC’s Division of Corporation Finance stated on March 20, 2025 that proof-of-work mining, solo or pooled, is not the offer or sale of a security and does not require registration. Mining is treated as an industrial activity; the financial disclosures that matter to investors come through ordinary SEC filings, and hashprice derivatives are overseen separately by the CFTC as commodities.
Yuki Tanaka covers mining and staking for HOGE Wire.