Bitcoin Mining Margins: The Profit the Network Takes Back

For most of 2026, running a Bitcoin mine has been a grind. Hashprice, the industry shorthand for what a unit of computing power earns in a day, spent the summer near five-year lows; Bitcoin traded far below its October 2025 record for months; and by CoinShares’ count, something like 15% to 20% of the global fleet was mining at a loss. The August rally changed the mood in a hurry. As Bitcoin pushed back toward $80,000, hashprice jumped about 22% over the month to $39.63 per petahash per second per day, according to Bitcoin.com News. Margins that had been scraping breakeven finally had a little air beneath them.

Then the network did what it always does. On September 6, at block height 965,664, mining difficulty adjusted upward by 1.31% to 127.45 trillion, the eighth increase of the year, the same report noted. That reads like a footnote. It is not. Difficulty is the one cost in mining that automatically rises when mining becomes more profitable, and it had just begun clawing back a slice of the raise the rally handed out.

That tug of war is the whole story of mining margins, and it is why the topic confuses so many newcomers. A miner’s profit is the gap between what hashrate earns and what it costs to produce, and both sides of that gap move. Revenue this year tracks the dollar price of Bitcoin almost one for one, because fees are a rounding error (just 0.43% of block rewards in the 24 hours before the September retarget). Costs are anchored by electricity and, above all, by difficulty, a protocol-level governor that keeps dragging the whole fleet back toward the break-even point of its least efficient machine. Bitcoin mining, in other words, is a commodity business with a thermostat wired into its cost base. This guide walks through how those margins are built, why the August windfall is already being competed away, and why the days around September 15 could decide whether the recovery holds.

Here is the state of play as September’s decision week approaches:

What a mining margin actually is

Miners, like any producers, talk about more than one margin, and the differences matter enormously when you are trying to judge whether an operation is healthy. Strip away the jargon and there are three.

The gross margin, or power margin, is the crudest: mining revenue minus the direct cost of the electricity that produced it. It tells you whether the machines are worth switching on at all. A rig that cannot cover its own power bill is a rig that should be idle.

The cash margin (often reported as cost to mine one bitcoin, excluding depreciation) is the one operators live by day to day. It adds the other running costs, pool fees, hosting fees, salaries, and site overhead, to power, but leaves out the cost of the hardware itself. A miner can be cash-positive, generating real money each month, while still losing money on paper.

The all-in margin adds depreciation, the accounting charge that spreads the up-front cost of the ASICs across their useful life, plus interest and corporate overhead. This is the number that determines whether a mine is truly profitable in the sense a shareholder cares about. It is entirely normal, and was the reality for much of 2026, for a miner to be comfortably cash-positive and deeply all-in-negative at the same time.

Keep those three straight and most of the confusion around mining profitability disappears. When a company says its cost to mine a bitcoin is $50,000 and the coin sells for $79,000, that is almost always a cash figure, and the same company may still be reporting a net loss once the machines and the debt are counted.

Hashprice: the revenue side in a single number

Every revenue conversation in mining runs through one metric: hashprice. Coined by Luxor’s Hashrate Index desk, it is the expected daily revenue per unit of hashrate, quoted in US dollars per petahash per second per day (divide by 1,000 for terahash). It bundles four moving parts into a single price.

Hashprice is approximately equal to 144 blocks per day, multiplied by the block subsidy plus fees, multiplied by the Bitcoin price, all divided by the network hashrate. Read it left to right and you can see exactly what pushes a miner’s revenue around. The network mines roughly 144 blocks a day. Each pays the current subsidy of 3.125 BTC (halved from 6.25 in April 2024) plus whatever fees the block carried. Multiply by the dollar price of Bitcoin to get total daily revenue, then divide by the total hashrate competing for it. If you own 1% of the hashrate, you earn about 1% of that pie.

Two features of that formula explain 2026. First, hashprice rises with Bitcoin’s price and falls as hashrate grows, which is why a rig can earn less this year than last even as the machines get faster: there is simply more competition for the same 144 blocks. Second, because fees are tiny right now, hashprice is effectively a leveraged bet on the Bitcoin price. At $39.63 per PH/day it is up about 22% on the month, tracking the rally almost exactly, per Hashrate Index. It is worth quoting hashprice in Bitcoin terms too, around 0.0005 BTC per PH/day, because in coin terms it has barely moved. The recovery is a dollar-price event, not a network event, a distinction we will come back to.

The cost stack: power, pools, machines, and time

If hashprice is the revenue side, the cost stack is everything trying to eat it. For a typical industrial miner it breaks down roughly like this.

Electricity is the giant. It is the single largest variable cost and the number every serious operator obsesses over, usually quoted in dollars per kilowatt-hour or per megawatt-hour. The difference between power at 3 cents and power at 6 cents is, quite literally, the difference between a thriving mine and a shuttered one. It is why miners chase stranded hydro, flared gas, and interruptible grid contracts, and why the location of the plug matters as much as the speed of the chip.

Hardware is the second giant, but it shows up differently. You pay for ASICs up front, then recognize the cost slowly as depreciation. Because a modern rig is effectively obsolete in a few years, that depreciation charge is heavy, and it is the main reason all-in costs sit so far above cash costs.

The rest is a supporting cast: pool fees (commonly 0% to 4% of revenue, depending on the payout scheme), hosting fees for miners who rent space rather than own it, staff, maintenance, cooling, insurance, and the interest on whatever debt funded the build. None of these is trivial, but power and depreciation dominate the picture.

One nuance trips up readers of mining financials constantly: the self-mining versus hosting split. A company that owns its sites books power directly and reports a low cash cost; a company that rents space at someone else’s facility pays an all-in hosting rate that already bundles power and overhead, so its cost line is not comparable. Two miners can report wildly different cost-to-mine numbers and both be telling the truth.

Cash margin versus all-in margin: the line between survival and solvency

The cleanest way to see the cash-versus-all-in gap is to read a real disclosure. Take Riot Platforms’ second-quarter 2026 results. Riot reported a cost to mine one bitcoin, excluding depreciation, of $49,912. Against a production value of roughly $71,667 per coin that quarter, that is a healthy cash margin: every bitcoin it produced was worth about 44% more than the cash it cost to make. On a cash basis, Riot was clearly profitable.

Now add depreciation. Including it, Riot’s cost to mine a single bitcoin rose to $90,631, or about 126% of the coin’s production value, according to the company’s quarterly results. On an all-in basis, the same bitcoins that looked profitable in cash terms were produced at a loss. That single quarter is the entire cash-versus-all-in lesson in one company: cash-positive and all-in-negative, simultaneously, by design.

Zoom out to the sector and the picture is similar. CoinShares’ mining research pegged the public miners’ weighted-average cash cost at roughly $79,995 per bitcoin for the fourth quarter of 2025, with all-in costs including depreciation closer to $138,000, per its 2026 mining report. With Bitcoin near $79,000, that puts a large share of the industry right at the cash-cost line and well underneath the all-in line, exactly the squeeze that pushed 15% to 20% of the fleet into unprofitable territory earlier in the year.

A word of caution when comparing these figures across companies: they are not standardized. Depreciation policies differ, some operators net power-curtailment credits against their cost line, and hosted miners report an all-in rate that self-miners do not. The table below lines up a few disclosed figures to show how far apart the definitions can sit, not to rank the companies. MARA Holdings, for instance, reported a cost per bitcoin of $38,690 in the same quarter, but that is a narrow purchased-energy figure that is not comparable to Riot’s broader cash cost.

Efficiency is destiny: the ASIC curve and the break-even power price

The reason power cost decides everything is that it interacts with a second number: machine efficiency, measured in joules per terahash (J/TH). The fewer joules a rig burns per unit of work, the more hashrate it squeezes from each kilowatt-hour, and the higher the power price it can survive.

The efficiency curve has bent dramatically. Bitmain’s 2016-era Antminer S9 ran at about 98 J/TH; the 2020 S19 series landed near 30 to 34 J/TH; the 2024 S21 XP hit 13.5 J/TH, per Bitmain; and the 2026 flagship, the hydro-cooled Antminer S23, is the first series miner below 10 J/TH at about 9.5 J/TH, as CoinDesk reported. Roughly a tenfold efficiency gain in a decade. That progress is exactly why hashrate can keep climbing even when miners are not spending more on power: each new generation converts the same megawatt into far more hashes.

For margins, efficiency translates directly into a break-even power price. A useful rule of thumb: the highest electricity price a rig can pay and still cover its power bill is hashprice divided by (24 multiplied by J/TH). Run that at today’s $39.63 hashprice and the spread across the fleet is stark.

Two things jump out. First, the newest machines can pay roughly triple the power price of the oldest and still break even, which is why the fleet is bifurcating into efficient survivors and obsolete rigs headed for the scrap heap or a cheaper country. Second, notice how the whole column shifted right as hashprice climbed: a month ago at $32.42, an S19j Pro needed sub-5-cent power to break even; at $39.63 it can survive closer to 5.6 cents. That is the rally widening everyone’s headroom at once. The difficulty increase on September 6 is the first tug in the other direction.

Hashrate Index frames the same idea as revenue per megawatt-hour by efficiency tier. At roughly $39 to $40 per PH/day, the most efficient fleets (under 14 J/TH) earn around $136 per MWh, while the 25-to-38 J/TH tail earns closer to $51. Same energy in, wildly different revenue out, decided entirely by the vintage of the silicon.

The marginal miner sets the floor

Here is where mining stops behaving like a technology business and starts behaving like a commodity. In any commodity market, the long-run price gravitates toward the marginal cost of production: the cost borne by the least efficient producer whose output the market still needs. Oil, corn, aluminum, all obey some version of this. Bitcoin mining obeys it too, with a twist.

The twist is that miners cannot set the price of what they sell. The dollar price of Bitcoin is exogenous, handed to them by the market. What adjusts instead is difficulty, and through it, the revenue each unit of hashrate earns. When mining is profitable, capacity that had been idled switches back on and new machines plug in; hashrate rises; difficulty follows; and the revenue per petahash falls back toward the cost of the marginal miner, the last rig still worth running. When mining is unprofitable, the reverse happens: the highest-cost rigs power down, hashrate falls, difficulty eases, and the survivors’ margins recover.

JPMorgan’s Nikolaos Panigirtzoglou put the mechanism plainly: “When bitcoin trades below its production cost, higher-cost miners power down, the hashrate declines, and difficulty adjusts lower,” he wrote in a client note reported by TFTC. His team pegged the all-in production cost around $78,000 and estimated a difficulty-to-price sensitivity of about 0.62, meaning difficulty moves a little over six-tenths as far as price does over time. That is the servo in action.

MARA chief executive Fred Thiel says the same thing from the operator’s chair. Mining, he told CoinGeek, “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.” The floor is the marginal producer’s energy cost, and difficulty is the mechanism that keeps pressing the industry down onto it. Understand that, and the August-to-September sequence stops looking like noise and starts looking like the system working exactly as designed.

The difficulty thermostat: the governor no other business has

Difficulty is Bitcoin’s automatic thermostat. Every 2,016 blocks, roughly every two weeks, the protocol recalculates how hard it is to find a block so that blocks keep arriving about every ten minutes regardless of how much hashrate is competing. Add hashrate and difficulty rises to slow you back down; remove hashrate and it falls. No committee votes and no central authority intervenes; the adjustment is baked into consensus.

For margins, that thermostat is a cost line no other industry carries. Imagine a bakery where, every time bread got more profitable, an invisible hand quietly raised the price of flour for every baker in the world until the extra profit was gone. That is difficulty. It is why mining margins are structurally mean-reverting: profitability invites hashrate, hashrate raises difficulty, and difficulty competes the profitability away.

2026 has been a vivid demonstration. Difficulty started the year around 146 trillion, fell through a long run of downward adjustments as weak prices and the pull of AI computing idled machines, and bottomed near 125 trillion in the summer; across the year the ten decreases have outweighed the eight increases for a net decline of roughly 13%, Bitcoin.com News reported. Then the rally arrived, idled rigs came back online, and on September 6 the thermostat clicked up 1.31% to 127.45 trillion. It is the clearest sign yet that the network has started to reclaim the rally’s gift. With hashrate hovering near 934 EH/s and the next retarget due in mid-September, the direction of travel has flipped from cooling to warming.

The lesson for anyone modeling a mine’s economics: never extend a good month straight out to the horizon. A rally that lifts hashprice also, with a lag of days to weeks, lifts difficulty, which pulls the hashprice per unit of your hashrate back down. The windfall is real, but it is partly a loan the network calls back.

A price event, not a network event

It is worth being precise about what actually recovered in August. In dollar terms, hashprice surged. In Bitcoin terms, it barely budged: around 0.0005 BTC per petahash per day, roughly where it has sat all summer. Nothing structural improved for miners. The subsidy is unchanged, fees are still negligible, and hashrate is flat. The only thing that moved was the dollar price of the coin.

That is why this year’s margins remain, as CoinShares head of research James Butterfill described the whole period, “one of the most challenging” in memory even after a 22% hashprice bounce: the recovery is entirely borrowed from price. Bitcoin.com News noted that fees made up just 0.43% of miner rewards in the day before the September retarget, which means, in its words, that “miner profits are very much connected to BTC’s price this year.” Take away the price move and the margin recovery evaporates.

This matters because it tells you where the risk lives. A miner whose entire margin cushion rests on a dollar-price rally is exposed to anything that can knock the dollar price down. It is a sharper version of a tension that runs through the whole mining and staking world: proof-of-stake validators face their own squeeze as more capital chases the same rewards, a dynamic we unpack in our look at SSV restaking and the institutional bid. And the calendar for the second half of September is stacked with exactly the events that can move price.

September’s decision week is the real margin risk

If margins are a leveraged bet on Bitcoin’s price, then the price is a leveraged bet on three days in the middle of September. The macro calendar has rarely been this concentrated.

First, on September 11, the Bureau of Labor Statistics releases August inflation data. A hot print, above roughly 3.2% year over year, strengthens the case for a Federal Reserve rate hike and pressures risk assets; a soft print buys the market room, per the Finance Calendar. Bitcoin already slipped back below $80,000 in early September after a stronger-than-expected jobs report pushed traders to price in a higher chance of a September hike, settling around $79,116 even as spot ETF inflows hit $3.8 billion over the strongest three-week stretch of the year, The Rio Times reported.

Second, on September 15, the US Senate holds a cloture vote on the CLARITY Act, the market-structure bill the industry has been waiting on for a year. It needs 60 votes to advance, and the odds have deteriorated sharply: Polymarket traders put the chance of the bill becoming law in 2026 at 16%, down from 82% in February, with fights over presidential crypto-income ethics, DeFi developer liability, and a stablecoin-yield provision blocking the path, according to crypto.news.

Third, the following day, September 16, the Federal Open Market Committee delivers its rate decision and updated projections, with the funds target sitting at 3.50% to 3.75% under Chair Kevin Warsh. Three market-moving events on consecutive days, each capable of swinging the Bitcoin price several percent, and therefore capable of swinging mining margins by far more thanks to the operating leverage described above. We laid out the full setup into this window in our Coiled Spring preview; for miners the takeaway is blunt: the margin recovery they are enjoying was underwritten by a price rally, and that price now faces its stiffest test of the quarter.

The AI escape hatch: when a megawatt is worth more elsewhere

There is a reason the phrase mining margin is starting to feel incomplete. For the largest operators, the real margin question in 2026 is no longer just how much a megawatt earns mining Bitcoin; it is whether that megawatt earns more doing something else entirely. Artificial-intelligence and high-performance-computing workloads pay far more per unit of power than hashing does, and they pay it on long, fixed contracts rather than a volatile hashprice.

The numbers are hard to ignore. Needham analyst John Todaro summed up the appeal: “The revenue per megawatt and EBITDA margins are far higher for HPC and AI colocation than for mining,” he told CoinGeek. CoinShares has tracked more than $70 billion in cumulative AI and HPC contracts announced across the public mining sector, and Butterfill has said some listed miners could draw up to 70% of their revenue from AI by the end of the year, up from roughly 30%, per Bitcoin.com News. The clearest single data point came in August, when Riot Platforms signed a 20-year lease with a leading frontier AI lab (reported to be Anthropic) for 191 megawatts, worth about $9.1 billion in base contract revenue and up to $16.1 billion with extensions; chief executive Jason Les called it “a defining moment in our evolution into a leading developer of large-scale data centers.”

For margins, the AI pivot is both a lifeline and a redefinition. A megawatt redirected to an AI tenant earns a stable, contracted return that is immune to hashprice and difficulty, exactly the volatility this article has been describing. That demand does not appear from nowhere; it is the same appetite for compute driving projects like decentralized inference networks and Bittensor’s push to train AI models without a data center, and it competes directly with hashing for every available megawatt and every high-end chip. Thiel’s forecast is that the divide will define the sector: “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.” The miners with cheap, abundant, well-located power increasingly have a choice about how to sell it, and that optionality is becoming the most valuable asset on their balance sheets.

Taking the swing out: hedging, forwards, and grid credits

Because margins swing so violently with price and difficulty, a maturing set of tools now lets miners lock in revenue and shave costs rather than simply riding the cycle.

On the revenue side, hashprice has become a tradeable instrument. Luxor offers hashrate forward contracts out to twelve months, and CFTC-regulated hashrate futures let a miner sell forward the revenue a unit of hashrate will earn, converting a volatile daily payout into a fixed one. Tellingly, the forward curve has spent much of the recovery in backwardation, with six-month contracts pricing below spot, the market’s way of saying it expects the rally-driven hashprice to fade as difficulty catches up. A miner who believes the same can lock in today’s higher number and sleep at night.

On the cost side, the biggest lever is the power contract itself. Operators in flexible markets earn real money by curtailing, powering down during demand spikes in exchange for grid payments. Riot, sitting in Texas, booked $10.1 million in power-curtailment credits in the second quarter of 2026 and $31.1 million across the first half, a direct offset to its cost of mining, its results show. Because a miner’s only cost of curtailing is the hashrate it briefly forgoes, with no spoiled product and no restart penalty, mining is unusually well suited to selling flexibility back to the grid. For the best-run operators, demand response is not a side hustle; it is a structural piece of the margin.

None of this changes the underlying commodity dynamics. It just lets a disciplined operator smooth the ride, trading some upside for a lot less downside, which in a business this volatile can be the difference between surviving a bad quarter and becoming someone else’s distressed acquisition.

The 2028 halving is the next margin cliff

Every four years, Bitcoin’s block subsidy halves, and every halving is a scheduled shock to the revenue side of the margin equation. The last one, in April 2024, cut the subsidy from 6.25 to 3.125 BTC. The next, expected in the spring of 2028, will cut it again to 1.5625 BTC. Because the subsidy is the overwhelming majority of miner revenue, a halving is, all else equal, an instant halving of the revenue that defines the margin.

The long-standing hope is that transaction fees eventually grow to replace the vanishing subsidy. So far, they have not. Fees ran at just 0.43% of block rewards heading into the September retarget, and even the brief 2024 spikes around Ordinals and Runes proved fleeting. That leaves miners depending on the other half of the equation, price. Thiel has warned that without sustained Bitcoin appreciation, on the order of 50% a year, “the math gets very tough after 2028,” as the security budget shrinks with the subsidy.

This is where the mining-margin story connects to Bitcoin’s longer-term security question, and to the scaling work happening at the edges of the network. If fees are ever to matter, block space has to become genuinely scarce and valuable, which is part of why so much attention is going to off-chain and second-layer designs, from Lightning to newer statechain approaches like Spark. For a miner planning capital expenditure today, the 2028 cliff is not abstract: a machine bought now will still be running when the subsidy halves, and its margin math has to survive that day.

How the SEC frames mining

For US-based miners and the investors who back them, one regulatory question sits underneath the margin math: is mining itself a regulated securities activity? The answer, as of 2026, is no. In March 2025 the SEC’s Division of Corporation Finance issued a staff statement concluding that proof-of-work mining, whether solo or through a pool, is not the offer or sale of a security, because a miner’s rewards come from its own computational effort rather than the entrepreneurial efforts of a third party. A broader 2026 interpretive release extended similar reasoning to staking, wrapping, and no-consideration airdrops.

That clarity matters for margins in two ways. First, it removes a layer of legal risk and cost that once hung over publicly traded miners, most of which are SEC registrants filing the detailed quarterly numbers this article has leaned on. Second, it draws a clean line between the assets and the instruments: spot mining is not a security, but hashrate futures and other derivatives are commodity products overseen by the CFTC, not the SEC. Miners hedging their margins in the forward market are operating in that CFTC-regulated lane.

The regulatory posture is not the same everywhere, and mining itself remains lightly touched compared with the exchanges and custodians that move the coins afterward. But for the question that governs mining margins, whether the core activity is legal and unencumbered in the United States, the answer is settled, and settled favorably.

Frequently Asked Questions

What is a good profit margin for Bitcoin mining in 2026?

There is no single figure, because it depends almost entirely on a miner’s power price and machine efficiency. A useful benchmark: with hashprice near $39.63 per PH/day, a modern rig at 13.5 J/TH breaks even on power at about 12 cents per kilowatt-hour, so a miner paying 4 cents enjoys a wide cash margin while one paying 10 cents is barely above water. On an all-in basis including depreciation, much of the public fleet was still unprofitable in 2026 even when cash-positive, with CoinShares estimating 15% to 20% of machines mining at a loss at the year’s lows.

Why do rising Bitcoin prices not guarantee higher mining margins?

Because difficulty adjusts. When price rises, mining becomes more profitable, so idle machines switch back on and hashrate climbs; roughly every two weeks the network raises difficulty to compensate, which lowers the revenue each unit of hashrate earns. The September 6 increase of 1.31% was exactly this: the network reclaiming part of the margin the August rally created. A price rally lifts margins immediately, but difficulty competes some of it back over the following weeks.

What is hashprice, and why does it matter for margins?

Hashprice is the expected daily mining revenue per unit of hashrate, quoted in US dollars per petahash per second per day. It is the revenue side of every mining margin in a single number, derived from the block subsidy, transaction fees, the Bitcoin price, and total network hashrate. In early September 2026 it sat around $39.63, up about 22% on the month. Comparing hashprice with a machine’s cost per unit of hashrate gives you the margin.

What is the difference between cash cost and all-in cost to mine a Bitcoin?

Cash cost, or cost excluding depreciation, covers the ongoing expenses of producing a bitcoin, mainly power, pool fees, and overhead, but not the price of the hardware. All-in cost adds depreciation of the ASICs plus interest and corporate overhead. The gap is large: Riot Platforms reported a cash cost of $49,912 and an all-in cost of $90,631 per bitcoin in the second quarter of 2026, meaning the same coins were cash-profitable and all-in-unprofitable at once.

How will the 2028 halving affect mining margins?

The halving will cut the block subsidy from 3.125 to 1.5625 BTC, roughly halving the revenue side of the margin unless price or fees rise to compensate. Because fees remain under 1% of block rewards, the burden falls on the Bitcoin price; analysts including MARA’s Fred Thiel argue the network needs strong continued price appreciation for miner economics to hold after 2028. Miners buying hardware today are already underwriting that risk.

By Yuki Tanaka, mining and markets correspondent, HOGE Wire.