Bitcoin Mining Margins: The Power Strategy Playbook

Ask a Bitcoin miner what decides whether they turn a profit this quarter, and most will start with the same two numbers: the price of Bitcoin and the efficiency of their machines. Both matter. Neither is the full story. By August 2026, network hashrate sits at roughly 940 EH/s, hashprice has spent most of the year pinned near five-year lows, and the newest Bitmain and MicroBT hardware is only marginally more efficient than the generation it replaced. Under those conditions, the operators still posting positive cash margins are rarely winning on hardware alone. They are winning on power.

This piece works through the mechanics of a Bitcoin mining margin, from the basic hashprice-minus-cost equation to the harder question underneath it: where does a miner’s electricity actually come from, and what does that sourcing decision do to the number that matters most. That means grid demand response programs that pay miners to go dark during peak hours, flared gas that would otherwise be burned off for free, sovereign power deals in the Gulf, and stranded hydro in places where state utilities are already trying to claw back cheap terms. It also means the parts of the margin story that get less attention: hardware efficiency as a multiplier rather than a starting point, hedging, regulatory risk, and what happens to all of this once the 2028 halving cuts the block subsidy again.

What a Mining Margin Actually Measures

A Bitcoin mining margin is simple in concept and slippery in practice. At its core it is revenue per unit of hashrate minus the cost of running that hashrate, expressed either as a percentage or, more usefully for comparing operators, as a dollar figure per bitcoin produced. The complication is that “cost” can mean at least three different things depending on which line items get counted.

Gross or cash margin counts only the electricity bill: what a miner pays the utility or grid operator to keep machines running, measured against the market value of the bitcoin those machines produce. All-in or fully loaded margin adds hardware depreciation, hosting fees, labor, insurance, interest on debt used to buy rigs, and general overhead. GAAP margin, the number that shows up in a public miner’s income statement, adds mark-to-market accounting for bitcoin holdings and any fair-value swings on inventory, which can swing a quarter from profitable to deeply unprofitable on paper even when the underlying mining operation is cash flow positive.

None of these is “wrong.” They answer different questions. Cash margin tells you whether a facility should keep running today. All-in margin tells you whether the business as a whole is sustainable. GAAP margin tells shareholders what happened to reported earnings, which is not the same thing as what happened to the miner’s actual bitcoin production economics. Confusing the three is one of the most common errors in casual mining-profitability commentary, and it is why two people can look at the same company’s quarter and reach opposite conclusions about whether mining still works.

The Revenue Side: Hashprice in Mid-2026

The revenue half of the margin equation collapses into a single number: hashprice, the expected dollar revenue a miner earns per petahash of hashrate per day. It is a function of four inputs, the block subsidy, transaction fees, Bitcoin’s dollar price, and total network hashrate, and it is the number every other mining economics question sits downstream of.

Hashprice has had a rough 2026. After peaking near $63 per PH per day in mid-2025, it slid through the back half of that year and spent early 2026 carving out new post-halving lows. By mid-July it was trading in a $29 to $32 per PH per day range according to Hashrate Index’s weekly roundup, still down by roughly half from year-earlier levels even after a bounce that followed the July 11 difficulty adjustment. Bitcoin itself was trading around $63,000 as of August 1, 2026, with a market capitalization near $1.27 trillion on circulating supply of just over 20.06 million coins.

Two forces are squeezing hashprice at once. Network hashrate, while off its 2025 highs, remains historically large, meaning the same block reward gets split across more machines. And transaction fees, which have occasionally spiked during periods of network congestion, have mostly settled back under 5% of total miner revenue, leaving the industry overwhelmingly dependent on the 3.125 BTC block subsidy rather than fee income to cover costs. CoinShares’ James Butterfill has argued the picture is not permanently bleak. In comments reported by The Block, he said it would not be “an unrealistic assumption to see bitcoin prices recover to $100,000,” a move that on its own would lift hashprice back toward the high $30s even with hashrate unchanged, a reminder that the revenue side of the margin equation is still primarily a bitcoin price story layered on top of a hashrate story.

The Cost Side: Cash Cost Versus All-In Cost

If hashprice is the revenue side, cost per bitcoin is everything else, and this is where the cash-versus-all-in distinction stops being an accounting technicality and starts deciding who survives. CoinShares’ Q1 2026 Bitcoin Mining Report put the weighted average cash cost across public miners at roughly $79,995 per BTC in the fourth quarter of 2025, a figure that excludes depreciation, and estimated that 15% to 20% of the global fleet was running unprofitably at prevailing hashprice levels. Separately, JPMorgan analyst Nikolaos Panigirtzoglou pegged the network’s all-in production cost near $78,000 and calculated that bitcoin traded roughly 19% below that level for about five consecutive months into mid-2026, a stretch reported by TFTC in which he warned that “when bitcoin trades below its production cost, higher-cost miners power down, the hashrate declines, and difficulty adjusts lower.” The two estimates use different methodologies and cover different populations of miners, so they should not be read as measuring the exact same thing, but both point the same direction: a meaningful slice of the industry mined at a loss on paper for most of 2026.

The clearest illustration of what “cost” actually contains sits inside Riot Platforms’ own first-quarter 2026 10-Q filing. Riot reported a cost to mine one bitcoin, excluding depreciation, of $44,629, against a production value of $75,964 per coin, a healthy cash margin. Add depreciation back in and the cost to mine jumps to $96,283, or 126.7% of production value: the same quarter, the same machines, the same power contracts, but a business that looks solidly profitable on a cash basis and underwater on a fully loaded one. Depreciation is not a cash expense in the period it is recognized, so neither number is “fake.” They simply answer different questions: can the lights stay on today, and is the capital that bought the machines actually being recovered.

The Blind Spot in the Naive Margin Equation

Here is what both the hashprice trend and the cash-versus-all-in framework leave out: they both assume a miner faces one electricity price. In practice, the operators posting the best margins in 2026 are not necessarily running the newest chips or sitting on the cheapest flat-rate power contract. They are treating electricity as a market to be actively worked, not a fixed input to be minimized once and left alone. That distinction, more than any single hardware refresh, is what separates the public miners still expanding capacity from the ones quietly shutting down older sites.

Four power-sourcing strategies dominate the current landscape, and each turns the basic margin equation into something more complicated than hashprice minus the power bill.

Grid Services as a Margin Line Item: ERCOT Demand Response

Texas’ deregulated grid operator, ERCOT, has become the clearest example of power strategy as margin because it pays large flexible loads, including Bitcoin miners, to curtail consumption when the grid is stressed. Mining fleets are unusually well suited to this: ASICs can drop from full power to near zero in seconds, with no minimum runtime, no warm-up period, and no equipment damage from stopping and restarting, an operational profile Hashrate Index has described as faster and cleaner than almost any other large industrial load on the grid. That flexibility qualifies miners for a range of ERCOT programs:

• Non-Spinning Reserve, which requires load reduction within thirty minutes of a dispatch instruction
• Responsive Reserve Service, the fastest and best-paying tier, requiring near-instant response
• Four Coincident Peak (4CP), where curtailing during the single highest-demand fifteen-minute interval across each of June, July, August and September can meaningfully cut a facility’s transmission charges for the following year

The dollar impact is not theoretical. Riot disclosed $21.0 million in power curtailment credits for the first quarter of 2026 alone, up roughly 169% from $7.8 million in the same quarter of 2025, against a self-mining power cost of $72.3 million for the quarter, meaning grid-services income offset close to a third of the company’s entire power bill. CEO Jason Les has been making this case since well before the current cycle: in August 2023 he touted a monthly record of $31.7 million in combined power and demand response credits, more than the company had received across all of 2022, evidence that this is a repeatable, multi-year revenue stream rather than a one-off. Automation matters too: tools that monitor real-time ERCOT prices and dispatch conditions and curtail the least efficient machines first are what make consistent participation in the faster, higher-paying reserve tiers practical at scale rather than something achievable only through manual intervention.

The mechanism works because of a quirk specific to Bitcoin mining: the only cost of curtailing is the hashrate not produced during the event window. There is no spoiled batch, no restarted furnace, no idle staff standing around a paused production line. That makes miners some of the cheapest flexible capacity ERCOT can call on, and it is why grid operators increasingly treat large mining fleets as a resource to be managed rather than simply a load to be served.

Flared and Stranded Gas: Mining Where the Pipeline Isn’t

A different power-arbitrage model skips the grid entirely. When an oil well also produces natural gas and there is no pipeline nearby to move that gas to market, drillers have historically burned it off, a practice called flaring, because burning unsellable gas is cheaper and safer than venting it directly. Bitcoin miners realized that a generator plus a container of ASICs could be trucked to the wellhead instead, turning gas that generated zero revenue into bitcoin.

Crusoe Energy built the best-known version of this model, deploying modular data centers that co-founder and CEO Chase Lochmiller has described as solving a problem the energy industry had struggled with for years. “People have tried to solve flaring for a long time,” he has said, “but the standard chemical engineering approaches like compression or liquefaction don’t make sense economically,” in ways that mining’s simpler economics could sidestep. Crusoe’s own account puts its Digital Flare Mitigation units at work capturing close to 22 billion cubic feet of gas across hundreds of sites that would otherwise have been flared. But by 2026 Crusoe itself is no longer the operator: it sold its bitcoin mining business to NYDIG in a deal announced in March 2025, folding roughly 270 megawatts of flare-gas generation capacity into NYDIG affiliate Stone Ridge Energy, which separately controls access to more than 10 gigawatts of US natural gas production. Crusoe itself pivoted fully to AI infrastructure, building large-scale data center campuses for hyperscale AI customers instead, a clean illustration of how thoroughly the AI buildout has redirected capital that used to flow toward Bitcoin mining specifically.

Crusoe was never the only player chasing this model. JAI Energy and Giga Energy Solutions both built businesses on the same premise, striking deals directly with oil and gas producers, concentrated in the Permian Basin of West Texas, where a large share of all US gas flaring happens. The economics are structurally different from grid-connected mining: cost basis sits close to zero since the alternative is burning the gas for no revenue at all, but the supply is not permanent. A well’s associated gas output declines over its productive life, and flaring itself faces tightening state and federal restrictions independent of Bitcoin’s economics, so a flare-gas fleet behaves more like a rolling series of short-lived deployments than a fixed power plant.

The Global Power Map: Sovereign Subsidies and Stranded Hydro

Zoom out from individual companies and the same arbitrage logic plays out at the level of entire countries. Hashrate Index’s global hashrate heatmap, last refreshed in April 2026, put the United States at roughly 37% of global hashrate, with Russia around 17% and China, officially banned since 2021 but still tolerated in practice in parts of the country, around 12%. Together the top three jurisdictions account for close to two-thirds of the entire network, a concentration that has held remarkably stable across consecutive quarterly snapshots.

Below the top three, the interesting movement is in smaller jurisdictions building hashrate specifically around stranded or subsidized power. The UAE and Oman together account for roughly 6% of global hashrate, anchored by state-linked operators such as Abu Dhabi’s Phoenix Group, which runs mining as deliberate industrial policy rather than opportunistic private arbitrage, a model that trades the entrepreneurial upside of a Riot-style demand-response strategy for the durability of a government-backed power contract. Paraguay, driven by a hydro-powered build-out, and Ethiopia, where more than twenty mostly Chinese-backed firms signed power purchase agreements near Addis Ababa after the Grand Ethiopian Renaissance Dam came online, both cracked the global top ten on the strength of stranded hydro that had no other buyer.

That last point comes with a warning built in. Ethiopia’s state utility has already proposed rate hikes projected to push roughly half of local mining operations into unprofitability by mid-2026 and the large majority by 2027, as domestic electricity demand starts competing for power miners had been using essentially for free. Kazakhstan lived through a harsher version of the same story: after absorbing a large share of the hashrate that fled China’s 2021 mining ban, briefly approaching a fifth of global hashrate, grid strain forced the government into a mandatory state-run electricity marketplace and hard caps on miner consumption, and its share has since fallen to under 2%. Cheap power that is cheap because nobody else wants it is not the same as cheap power that will stay cheap once somebody else does.

Figures are from Hashrate Index’s April 2026 snapshot and drift from quarter to quarter as new capacity comes online and old capacity relocates, but the ranking of the top three jurisdictions has proven unusually sticky across consecutive updates.

Hardware Efficiency: The Multiplier on Every Power Strategy

None of this makes chip efficiency irrelevant, it just changes the question it answers. Efficiency does not create margin on its own; it determines how much margin a given power strategy can extract. A miner paying three cents per kilowatt-hour thanks to flared gas or demand-response credits can run older, less efficient machines profitably, while a miner on standard grid power at eight or ten cents needs the newest silicon just to break even.

The math is straightforward. A machine’s power draw equals its hashrate multiplied by its efficiency rating in joules per terahash, so the breakeven electricity price at a given hashprice works out to the hashprice, in dollars per petahash per day and divided by 1,000, divided by the machine’s efficiency multiplied by 0.024. At a hashprice of roughly $30 per PH per day, the numbers separate hardware generations clearly. Bitmain’s current flagship, the Antminer S21 XP, rated at 13.5 joules per terahash, can tolerate power priced above roughly 9 cents per kilowatt-hour and stay cash-profitable. The outgoing Antminer S19 XP, at 21.5 joules per terahash, needs power under roughly 6 cents to clear the same bar. Any five-year-old hardware still running at all in mid-2026 is itself a signal: it almost certainly sits behind one of the power-arbitrage strategies described above, because it could not survive on a standard grid contract at current hashprice.

This is the multiplier effect in practice: pair the most efficient hardware with the cheapest power strategy and the resulting margin compounds, which is exactly why the operators expanding capacity in 2026, rather than shutting it down, tend to be running newer fleets on curtailment-eligible grid connections or flared gas rather than newer fleets on ordinary retail power.

Public Miner Scorecard: Same Playbook, Different Bets

The two largest US-listed miners make a useful contrast because they are running the same basic playbook, power plus scale plus a pivot toward AI infrastructure, with different emphasis. Riot Platforms posted first-quarter 2026 revenue of $167 million, split between $111.9 million in Bitcoin mining and a new $33.2 million data-center segment, its first material non-mining revenue line, while producing 1,473 bitcoin at a deployed hashrate of 42.5 EH/s. MARA Holdings, still the larger miner by hashrate at 72.2 EH/s energized, reported revenue of $174.6 million, down 18% year over year, and a net loss of roughly $1.26 billion driven mostly by unrealized bitcoin fair-value changes rather than operating losses. MARA CEO Fred Thiel has been blunt about where this leaves the industry, calling mining “a zero-sum game” in comments reported by CoinGeek: “as more people add capacity, it gets harder for everybody else. Margins compress, and the floor is your energy cost.” He has gone further, predicting that by 2028 miners will “either be a power generator, be owned by one, or be partnered with one,” framing pure hashrate exposure without an attached power strategy as an increasingly untenable position.

A third operator, CleanSpark, offers a smaller-scale comparison: a leaner fleet run closer to a pure-play mining model than either Riot or MARA’s increasingly diversified balance sheets, without the same scale of AI-related capital commitments. None of the three has escaped the fundamental problem this article opened with: at a roughly $30 hashprice, cash margins are survivable but thin, and the companies posting the strongest numbers are consistently the ones layering a power strategy, not just a hardware refresh, on top of their fleets.

The AI and HPC Pivot: Power Strategy Taken to Its Logical End

If demand response and flared gas are ways to extract more margin from a power contract built for Bitcoin mining, the AI and HPC pivot is the industry deciding the power contract itself is worth more than the mining. CoinShares tallies more than $70 billion in cumulative announced AI and HPC contracts across public miners, led by deals like Core Scientific’s roughly $10.2 billion, twelve-year colocation agreement and TeraWulf’s $12.8 billion in total contracted HPC revenue. Butterfill has estimated AI could supply “as much as 70% of their revenues by the end of this year, up from roughly 30% today” for the miners pursuing the pivot most aggressively.

The logic holds up once framed as a power-strategy question rather than a bitcoin-price bet: a megawatt of power under a long-term AI colocation lease can generate more predictable revenue than the same megawatt run through ASICs at a $30 hashprice, and Needham & Co analyst John Todaro has made exactly that comparison in the same CoinGeek piece, noting that “the revenue per megawatt and EBITDA margins are far higher for HPC and AI colocation than for mining.” Riot’s own new data-center segment reflects the same shift: Jason Les has said the first quarter of 2026 “marks a definitive inflection point for Riot, as we officially transitioned into an active, revenue-generating data center operator,” language that would have been unthinkable from a Bitcoin miner’s chief executive a few years earlier.

This is not a costless pivot. Not every AI workload is equivalent, and part of what is drawing capital into this exact corner of infrastructure is demand for compute that can prove its own outputs are trustworthy, a theme also playing out in Ethereum’s restaking sector as it bets on verifiable AI infrastructure, whether that means training runs, inference serving, or newer approaches like the optimistic machine learning schemes now being explored for onchain AI. Miners without a signed anchor tenant are making multi-hundred-million-dollar bets on build-to-suit data centers before the revenue is locked in, and Thiel himself has warned that the fee-revenue transition long promised as Bitcoin’s subsidy shrinks “hasn’t happened” and the math “gets very tough after 2028” absent sustained double-digit annual price appreciation, which is exactly why so much of the industry is hedging that bet with a second, non-Bitcoin revenue line.

When the Subsidy Disappears: Grid and Policy Risk

Every power-arbitrage strategy in this piece carries a version of the same risk: the favorable terms exist because someone with authority over the power supply has decided, for now, to allow them, and that decision can be revisited. ERCOT’s demand-response programs are administered by a grid operator answering to Texas regulators and, increasingly, to a public conversation about whether Bitcoin miners and AI data centers are straining a grid that residential customers also depend on. Ethiopia’s rate hikes and Kazakhstan’s consumption caps, described above, are the clearest historical examples of a government simply changing the terms once stranded power stops being stranded.

Grid risk is not the only variable. Flared-gas mining depends on wells that deplete and on flaring rules regulators can tighten independent of anything Bitcoin-related. Sovereign power deals depend on the political durability of whichever state entity signed them. And even grid-connected demand response has a subtler risk: it works best when a miner has the balance sheet to absorb long stretches of curtailment without missing debt payments, meaning the strategy that helps the largest, best-capitalized public miners most is often least available to smaller, debt-constrained operators, reinforcing consolidation rather than spreading the benefit evenly across the industry.

None of this is unique to Bitcoin mining; businesses built around a temporarily mispriced input eventually have to contend with that price correcting. But it does mean a margin analysis built purely on hashprice and hardware efficiency, without accounting for how durable a miner’s specific power deal actually is, is missing the variable most likely to determine which companies are still operating by the next halving.

Hedging the Margin: Derivatives and Forward Curves

A newer piece of the margin toolkit lets miners lock in revenue rather than simply chase cheaper costs. Luxor operates a hashrate derivatives market offering forward contracts of up to twelve months alongside shorter daily-settlement products, letting a miner sell future hashprice exposure the way an airline hedges jet fuel. Recent quotes have priced a twelve-month forward around $28.94 per PH per day through the back half of 2026, implying a miner’s cost of capital in the 6% to 13% annualized range once the forward curve is compared against spot.

Hedging does not eliminate the underlying margin squeeze; a miner who locks in $29 per PH per day is still locking in a thin margin. But it converts an unpredictable number into a predictable one, which matters enormously for a business trying to service debt raised against future mining revenue. It is telling that the miners most active in this market tend to be the same ones running sophisticated power strategies elsewhere: hedging and grid-services participation are both symptoms of an industry that has professionalized well beyond plugging in machines and hoping the price goes up, even if the underlying commodity, hashrate itself, remains as volatile as ever.

The Regulatory Backdrop: Why Mining Isn’t Treated as a Security

One piece of good news for the industry’s margin outlook has nothing to do with power or hardware. In March 2025, the SEC’s Division of Corporation Finance issued a statement concluding that proof-of-work mining, whether solo or through a pool, does not constitute a securities transaction under the Howey test, reasoning that block rewards flow from a miner’s own computational contribution rather than the entrepreneurial effort of a third party, a position confirmed by The Block. A broader interpretive release in March 2026 extended similar no-securities framing to staking, wrapping, and no-consideration airdrops, part of a wider effort the agency has described as an attempt to draw “clear lines in clear terms.”

That regulatory clarity matters for margin in an indirect but real way: it removes one category of tail risk, namely an SEC enforcement action against the act of mining itself, from a business already carrying grid risk, commodity-price risk, and hardware-obsolescence risk. It is a small mercy relative to the agency’s broader, still-evolving posture toward the rest of the crypto industry, detailed in HOGE Wire’s report card on the SEC’s own crypto enforcement record. Mining, for now, sits outside that machinery entirely.

Outlook: Margins on the Road to the 2028 Halving

The block subsidy halves again in the spring of 2028, cutting the primary revenue source for every strategy in this article from 3.125 BTC to 1.5625 BTC per block overnight, a mechanical certainty explored in more detail in HOGE Wire’s ongoing halving cycle math series. Absent a large bitcoin price move to offset it, that is a direct hashprice cut of roughly half, layered on top of whatever the network’s efficiency and hashrate trends do independently between now and then.

The operators best positioned to absorb that are, unsurprisingly, the ones already treating power as a strategy rather than a fixed cost: fleets with demand-response income that scales with grid stress rather than bitcoin’s price, flare-gas contracts priced near zero regardless of what hashprice does, sovereign power deals insulated from spot electricity markets entirely, and, increasingly, a second revenue line from AI and HPC hosting that does not depend on bitcoin’s price at all. For everyone else, the arithmetic gets unforgiving fast: a miner on standard grid power running anything less than the newest hardware generation was already struggling at a $30 hashprice, and a mechanical halving toward something near $15, before accounting for any further hashrate growth or price recovery, would leave a large share of the current fleet needing power under three or four cents per kilowatt-hour just to cover cash costs. That is not a hypothetical scenario so much as a floor the entire industry is now visibly organizing its capital allocation around, whether that means buying power plants outright, signing hyperscaler leases, or building modular units next to a gas flare in the Permian.

Frequently Asked Questions

What is a good profit margin for Bitcoin mining?

There is no single healthy number, since it depends on whether you are measuring cash margin or all-in margin. In mid-2026, with hashprice near $30 per PH per day, well-run operators on cheap or subsidized power were still clearing cash margins in a range of roughly 30% to 50%, while the same fleets often showed thin or negative margins once depreciation, debt service, and overhead were included. A public miner reporting a positive all-in margin at current hashprice levels is running a genuinely efficient operation, not an average one.

Why are Bitcoin mining margins so low in 2026?

Two things happened at once. Network hashrate stayed historically high even as bitcoin’s price pulled back sharply from its October 2025 all-time high, and neither transaction fees nor a price recovery arrived to offset the resulting drop in hashprice, which spent much of the year near five-year lows around $28 to $32 per PH per day. CoinShares has estimated that 15% to 20% of the global mining fleet was unprofitable at those levels.

What is the difference between cash cost and all-in cost in Bitcoin mining?

Cash cost counts only the expenses a miner pays out in the current period, mainly electricity, and measures whether a facility should keep running today. All-in cost adds non-cash items like hardware depreciation plus hosting, labor, and overhead, and measures whether the business as a whole is economically sustainable. Riot Platforms’ first-quarter 2026 filings showed this gap clearly: a cash cost of $44,629 per bitcoin against an all-in cost of $96,283 for the same quarter.

How do Bitcoin miners make money from demand response programs?

Grid operators such as ERCOT pay large, flexible electricity users to reduce consumption during periods of peak demand or grid stress. Because ASICs can power down instantly with no equipment damage, mining fleets are exceptionally cheap sources of that flexibility compared with other industrial loads, and can earn both payments for being available and payments for actually curtailing during an event. Riot Platforms reported roughly $21 million in such credits in the first quarter of 2026 alone.

Will Bitcoin mining still be profitable after the 2028 halving?

It depends heavily on which power strategy a given operator is running. The block subsidy will fall from 3.125 to 1.5625 BTC overnight, roughly halving revenue at a fixed bitcoin price and hashrate. Operators with near-zero power costs from flared gas, sovereign subsidies, or grid-services income that does not depend on bitcoin’s price have real room to absorb that; operators paying standard grid rates on older hardware are the ones analysts expect to be forced offline first, similar to prior halving cycles.

Reporting by the HOGE Wire Mining and Staking desk.