Bitcoin Mining Margins 2026: The Third Shakeout and the AI Exit

A $9.1 billion escape hatch that 2018 and 2022 never had

On August 11, 2026, Riot Platforms told investors it had signed a 20-year lease for 191 megawatts of computing capacity at its Rockdale, Texas campus. The company described the counterparty only as “one of the world’s leading frontier AI labs”; Bloomberg reported it to be Anthropic. The base term is worth about $9.1 billion, rising to as much as $16.1 billion if two five-year extensions are exercised, and it runs through June 2048, according to Riot’s own disclosure and reporting from DataCenterDynamics. Riot’s stock jumped roughly 17 percent on the day.

Read that against the same company’s mining numbers, filed the same week, and the picture turns strange. In the second quarter of 2026, it cost Riot $49,912 in cash to produce a single Bitcoin and $90,631 once depreciation is included, or 126.5 percent of what that Bitcoin was worth when it was mined. The mining business, in other words, lost money on a full-cost basis in the very quarter the company signed one of the largest data-center leases in its history.

That contradiction is the story of Bitcoin mining margins in 2026. This is the third time in eight years that the economics of mining have collapsed hard enough to threaten the survival of the companies doing it. The first was 2018. The second was 2022. Both ended in bankruptcies, forced sales, and hardware sold, as one 2018 headline put it, “by the pound.” The third is happening now, with hashprice, the industry’s core revenue metric, sitting near a five-year low. What separates 2026 from the two shakeouts before it is not the severity of the squeeze. It is that this time there is somewhere else for the power to go.

This piece walks through what a mining margin actually is, why the 2026 squeeze looks the way it does, how it compares in depth and mechanics to 2018 and 2022, and why the pivot to artificial intelligence and high-performance computing is turning what used to be a clean bankruptcy cull into something messier: a bifurcation between miners who have an exit and miners who do not.

What a mining margin actually is: three numbers, not one

A common mistake is to talk about the mining margin as if it were a single figure. Public miners report at least three, and on the same quarter they can tell very different stories.

• Gross or hashprice margin: revenue per unit of computing minus almost nothing, useful only as a top-line reference.
• Cash cost margin: revenue minus the cash cost of production, dominated by electricity. This is the number that tells you whether a machine should stay switched on today.
• All-in cost margin: cash cost plus depreciation, plus corporate overhead, plus the non-cash charges that come from carrying Bitcoin and hardware on the balance sheet. This is closer to whether the business creates value over the life of its machines.

Riot’s second quarter is a clean worked example, because the company discloses both ends. Cash cost to mine one Bitcoin, excluding depreciation, was $49,912, or 69.6 percent of production value. Include depreciation and the cost jumps to $90,631, or 126.5 percent of production value, per the company’s results. The same coin was cash-profitable and GAAP-unprofitable at once. Neither number is wrong; they answer different questions. Keep the machines running (yes, cash-positive) versus are the machines, at today’s prices, worth what they cost to buy and power over their lives (no, not at these levels).

The gap between those two numbers is depreciation, and depreciation is where much of the confusion in mining-stock analysis lives. We return to it below, because it is also where miners can flatter or bury their real economics. For readers who want the full disclosure-by-disclosure walkthrough, our field guide to reading miner margins goes line by line. Here the point is simpler: a margin squeeze is not one event. It is the cash line and the all-in line moving toward, and then through, the value of the Bitcoin being produced.

The revenue side: hashprice at a five-year low

Every miner’s revenue reduces to one number the industry calls hashprice: the expected daily revenue from one unit of computing power, quoted in US dollars per petahash per second per day. It is set by four inputs, namely the block subsidy (3.125 BTC per block since the April 2024 halving), transaction fees, the Bitcoin price, and total network hashrate. The first is fixed until 2028; the second is usually tiny; the third and fourth do the work.

As of the second week of August 2026, hashprice sat at about $31.73 per PH per day, according to Luxor’s Hashrate Index. That is down by roughly half from the mid-2025 peak above $60, and it is in the same five-year-low zone the metric first reached in late 2025. The cause is a squeeze from both directions: Bitcoin traded near $63,000 in mid-August, about half its October 2025 record near $126,000, per CoinGecko, while network hashrate stayed historically high at a seven-day average near 911 EH/s. More machines are splitting a smaller dollar reward.

The mechanical result is that revenue per machine has fallen faster than most miners can cut costs. When a modern rig earns half what it did a year ago and its electricity bill has not halved, the margin gives. That is the revenue side of every mining-margin story, and it is why hashprice, not the Bitcoin price alone, is the number miners watch.

The cost side: energy is the floor

If hashprice is the revenue line, electricity is the cost line, and for a pure-play miner it is the cost line that matters. Fred Thiel, chief executive of MARA Holdings, the largest US-listed miner, put the economics plainly to CoinGeek: mining “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 efficiency of a machine, measured in joules per terahash (J/TH), decides how much electricity it burns for a given amount of computing. The newest air-cooled units, such as Bitmain’s Antminer S21 XP, run near 13.5 J/TH. Mid-generation rigs like the WhatsMiner M60S sit closer to 18.5 J/TH, and the previous generation of S19-class machines ranges from about 21.5 J/TH up to nearly 30 J/TH for the older S19j Pro.

That efficiency, combined with hashprice, sets a break-even electricity price: the cost per kilowatt-hour above which a machine loses money on a cash basis. The relationship is direct. Break-even dollars per kWh equals hashprice (in dollars per TH per day) divided by daily energy use per terahash. At today’s hashprice, a top-tier S21 XP breaks even near 9.8 cents per kWh, while an old S19j Pro needs power below roughly 4.5 cents to stay cash-positive.

These figures are illustrative: hashprice already nets out fees, and the table ignores pool fees, downtime, cooling overhead, and hosting markups. The takeaway holds anyway. Since a large share of the global fleet still runs mid- and older-generation hardware, and since few sites buy power below 4 to 5 cents, a meaningful slice of the network is underwater. CoinShares put that slice at 15 to 20 percent of machines in its 2026 Bitcoin Mining Report. This is why power sourcing, not chip shopping, is the real margin lever, a point we have made at length before. The floor Thiel describes is not a metaphor; it is the number on the utility bill.

The first great shakeout: 2018

To see why 2026 is being called the third shakeout, start with the first. Bitcoin peaked near $19,700 in December 2017 and spent all of 2018 grinding lower, ending the year around $3,200, a drawdown of roughly 84 percent. Mining economics did not just weaken; they inverted. Rigs that had been printing money at the top were worth less than the electricity they consumed at the bottom.

The emblem of that cycle was Giga Watt, a Washington-state mining host that had raised money through a 2017 token sale and offered power rates that turned out to be unsustainable. It filed for Chapter 11 on November 19, 2018, owing its twenty largest unsecured creditors nearly $7 million against assets it valued at under $50,000, as CoinDesk reported at the time. Around it, the broader industry contracted violently. One widely shared account described miners selling hardware by the pound. Bitmain, then the dominant machine maker, cut staff and shelved a planned Hong Kong public offering.

The network told the same story. In the first week of December 2018, Bitcoin’s mining difficulty fell 15.13 percent in a single adjustment, one of the largest downward moves in its history, as miners switched off en masse, per CryptoSlate. That is the self-correcting mechanism at the heart of Bitcoin mining: when price falls below the cost of production, the least efficient miners capitulate, hashrate drops, and difficulty follows, lowering the cost of production for whoever is left. In 2018 that capitulation lasted about 72 days and was followed by a long recovery. The survivors, the ones with cheap power and clean balance sheets, inherited the network.

The second great shakeout: 2022

The 2022 cycle rhymed, with a twist: leverage. Bitcoin fell from a November 2021 record near $69,000 to about $15,500 a year later, a decline of roughly 77 percent. This time the miners going under were not undercapitalized hosts but large, publicly traded or well-funded operators that had borrowed against their machines and their coins during the boom, then got caught between a collapsing Bitcoin price and a 2022 spike in energy prices.

Compute North, one of the largest hosting providers, filed for Chapter 11 on September 22, 2022, owing as much as $500 million to at least 200 creditors, according to CoinDesk. Three months later, Core Scientific, then one of the biggest listed miners in North America, filed for bankruptcy on December 21, 2022, though it kept its machines running and even took a fresh $17 million loan from BlackRock days later; it would not emerge from Chapter 11 until early 2024. Celsius’s mining arm went down with its parent. Others, such as Argo Blockchain, avoided bankruptcy only by selling their flagship facilities and taking rescue financing.

The lesson of 2022 was that the margin squeeze does not kill miners directly; leverage does. A miner running underwater on cash cost can idle machines and wait. A miner that has pledged those machines and its Bitcoin as collateral cannot, because the loans come due whether or not the rigs are switched on. That distinction, cash-underwater versus balance-sheet-underwater, is exactly the one that separates a survivable squeeze from a fatal one, and it is why the 2026 disclosures deserve a closer read than the headline cost-to-mine number.

The third shakeout: 2026, and why it looks different

The 2026 squeeze does not have the shape of the first two. There was no euphoric top to fall from: Bitcoin’s October 2025 record near $126,000 gave way to a long, grinding drift rather than a crash, and the metric under pressure is not primarily the coin price but hashprice, which the halving and relentless hashrate growth pushed to a five-year low even while Bitcoin held in the $60,000s.

The cost math is stark. CoinShares put the weighted-average cash cost of production across public miners near $79,995 per coin, above where Bitcoin has traded for much of the year, in what its head of research called the most challenging quarter for miners since the April 2024 halving. JPMorgan reached a similar conclusion from the other direction: analyst Nikolaos Panigirtzoglou estimated an all-in production cost around $78,000 and noted Bitcoin had traded roughly 19 percent below that level for about five straight months into mid-2026, per coverage of his client note. His description of what happens next is the same mechanism that governed 2018: “When bitcoin trades below its production cost, higher-cost miners power down, the hashrate declines, and difficulty adjusts lower.”

That thermostat is already visible. Network difficulty has fallen roughly 19 percent from its late-2025 peak near 156 trillion, and the next scheduled adjustment in late August 2026 was estimated at about minus 3 percent. The self-correction is doing its job: as the weakest machines go dark, the survivors’ economics quietly improve. What is different this time is what the weakest machines, and the companies that own them, are doing instead of simply going bankrupt. The macro backdrop matters here too, because a shifting Federal Reserve rate path feeds directly into miner cost of capital and refinancing, a link we unpack in our look at collapsing Fed hike odds and the crypto reaction.

The release valve: AI and high-performance computing

Here is the structural break from 2018 and 2022. In those cycles, a stranded megawatt of mining power had essentially one alternative use: mine less, or mine nothing. In 2026, that same megawatt, sitting behind an interconnection agreement and a substation that took years to build, is one of the scarcest things in the American economy, because it is exactly what an AI data center needs.

The numbers moving between the two businesses are not close. John Todaro, an analyst at Needham, told CoinGeek that “the revenue per megawatt and EBITDA margins are far higher for HPC and AI colocation than for mining. Capital markets are rewarding AI-focused data centers with much higher multiples than traditional miners.” A megawatt that earns a mining company a volatile few tens of dollars a day can, under a long-term AI lease, earn a fixed, contracted stream for twenty years. CoinShares counts more than $70 billion in cumulative AI and HPC contracts announced across the public mining sector.

Riot’s August lease is the clearest single proof of the shift. The 191-megawatt, 20-year agreement is worth about $9.1 billion in base contract revenue, with estimated net operating income of $7.3 to $8.2 billion over the term, and Riot secured $573 million from Morgan Stanley to fund initial construction. Combined with a 50-megawatt agreement with AMD, Riot now has roughly 241 megawatts under contract for about $9.8 billion of long-term revenue. Chief executive Jason Les called it “a defining moment in our evolution into a leading developer of large-scale data centers.”

What that compute actually does is worth understanding, because it is the demand pulling power away from Bitcoin. Frontier AI labs need enormous, reliable, low-latency clusters to train and serve models, and a growing part of the field is trying to make that compute verifiable and trust-minimized rather than a black box, a problem we cover in our piece on whether verifiable AI needs its own blockchain. The mining industry’s bet is that it owns the one input all of that demand cannot manufacture on a reasonable timeline: energized, permitted power.

Reading a miner’s margins without getting fooled

If two of the largest miners can post nearly identical revenue and opposite narratives, the disclosures need care. A few traps recur.

First, cost-to-mine is not a standardized number. Riot’s $49,912 cash cost and MARA’s self-reported purchased-energy cost of $38,690 per coin for the same quarter are not directly comparable, because MARA’s figure captures only purchased energy at four-cent power while Riot’s cash cost is a broader measure. A lower headline cost-to-mine can reflect cheaper power, a narrower definition, or a more flattering accounting choice, and often all three.

Second, hashrate has more than one definition, and miners tend to quote the largest. MARA reported energized hashrate of 70.3 EH/s in the second quarter, but energized counts every rig plugged in, including idle ones, and overstates the capacity actually producing coins. Operating and realized hashrate are the honest numbers; the gap between energized and realized is where marketing lives.

Third, Bitcoin on the balance sheet runs through the income statement. Both MARA and Riot carry their mined coins at fair value, so a falling Bitcoin price produces large non-cash losses that have nothing to do with mining efficiency. MARA’s second-quarter net loss of $611.3 million was driven mainly by a $343.0 million unrealized mark-to-market on its Bitcoin, not by its power bill. Read a miner’s headline loss and you are often reading a leveraged bet on the Bitcoin price wearing a mining company’s clothes.

The balance-sheet twist: miners as leveraged Bitcoin bets

That last point deserves its own section, because it changes how a margin squeeze plays out. A miner that holds thousands of coins is running two businesses at once: a computing business with a real, if thin, operating margin, and a Bitcoin treasury that swings with the market. When the coin falls, the treasury loss can swamp the operating result, and the market often cannot tell the two apart.

The behavior this produces in 2026 is telling. Rather than sell equity into a depressed stock or borrow against machines the way the 2022 cohort did, several miners have sold Bitcoin from treasury to fund the AI pivot. MARA sold roughly $1.5 billion of Bitcoin to finance its transition, according to CoinDesk; Riot sold coins through the year to fund construction while still holding 11,380 BTC, of which 5,821 are pledged as collateral. Where custody and pledging of that treasury sits, whether in self-custody, at an exchange, or under a lender’s control, is not a footnote; as we have argued elsewhere, control is not the same as ownership, and a pledged coin is not a coin you can sell in a hurry. The miners that entered 2026 with unencumbered Bitcoin and low debt have optionality. The ones that pledged everything are back in the 2022 trap, squeeze or no squeeze.

Power strategy is margin strategy

Because energy is the floor, the miners with the best margins are usually the ones that treat their power contract as a profit center rather than a cost. Riot books grid credits by curtailing during peak demand: in the second quarter of 2026 it recorded $10.1 million in power curtailment credits, and $31.1 million over the first half. In Texas’s ERCOT market, a miner that can drop hundreds of megawatts within minutes gets paid for that flexibility, and unlike a factory it loses only foregone hashrate when it curtails, with no spoiled product and no restart penalty.

The AI pivot sharpens this into a direct trade-off. Every megawatt a miner commits to a fixed AI lease is a megawatt it cannot point at Bitcoin when hashprice spikes, and the reverse is also true. Riot’s lease delivers just 96 megawatts by December 2027 and the full 191 megawatts by June 2028, which means the company will run both businesses side by side for years, arbitraging power between a volatile mining revenue line and a fixed data-center one. That is the new margin game: not just cheap power, but the freedom to send each megawatt wherever it earns the most that day. It is a more sophisticated version of the same sourcing logic that has always separated the survivors from the casualties.

Fees, the halving, and the long-term margin math

Every margin projection eventually runs into two structural facts. The block subsidy halves again in the spring of 2028, from 3.125 BTC to 1.5625, cutting the largest input to hashprice in half unless price or fees make up the difference. And transaction fees, which are meant to eventually replace the subsidy as the network’s security budget, remain tiny, often well under 1 percent of the block reward in 2026.

That is why the long-run bull case for mining margins leans partly on Bitcoin becoming a busier settlement layer, not just a more valuable coin. Efforts to move dollars and other assets over Bitcoin, including the work described in our piece on dollars returning to Bitcoin through Taproot Assets, are, from a miner’s point of view, potential fee demand, the one revenue line that does not halve. Thiel’s framing of the 2028 endgame is blunt: “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.” Whether that is right or not, it explains why the smartest capital in mining is spending 2026 buying power plants and signing AI leases rather than buying more machines.

It is also worth remembering that mining is only one half of how crypto networks pay for their own security. Proof-of-stake systems replace hashrate and power bills with bonded capital and yield, an economic model we cover in our explainer on how staked ETH secures other systems. The contrast matters here because it frames what mining margins are compensation for: the real, energy-denominated cost of securing a proof-of-work chain, a cost that staking simply does not incur.

Where the SEC fits

For US-listed miners, the regulatory picture in 2026 is unusually settled on the core question. In March 2025 the Securities and Exchange Commission’s Division of Corporation Finance stated that proof-of-work mining on public, permissionless networks is not the offer or sale of a security, and that pooling hashrate does not change the analysis, because miners rely on their own efforts rather than those of a promoter. That removed a long-standing overhang: mining Bitcoin, and being paid for it, is not itself a securities activity.

What the SEC does still govern is disclosure. MARA and Riot are registrants that file quarterly and current reports, and their cost-to-mine figures, hashrate definitions, and Bitcoin fair-value accounting all sit inside that framework, which is why the numbers this article relies on are auditable rather than marketing. The AI pivot adds a new disclosure surface: a 20-year, multibillion-dollar lease is a material contract that must be described accurately, including who bears construction risk and what happens if a tenant walks. Separately, hashrate futures and other mining derivatives trade as commodities under the CFTC, a reminder that the same activity can sit under different regulators depending on whether you are doing it or betting on it. For miners, the practical upshot is that the margin story now has to be told to auditors and to a market that has learned, after 2022, to read past the headline number.

What survives the third shakeout

Put the three cycles side by side and the pattern is clear. Each squeeze culls the miners whose cost of production sits above the Bitcoin price and whose balance sheets cannot absorb the wait. 2018 killed the undercapitalized hosts. 2022 killed the over-leveraged operators. 2026 is culling the same profile, high cash cost, high debt, no cheap power, but for the first time it is not only killing them. It is sorting them.

The release valve changes the ending. A miner with cheap, flexible power and a clean balance sheet can now convert distress into a decades-long AI contract, as Riot just did, or run both businesses and arbitrage between them, as MARA is attempting with a power target raised toward 4.8 gigawatts. A miner without those things, high cost, fully pledged, no interconnection worth leasing, still faces the 2018 and 2022 outcome. The third shakeout is therefore less a bankruptcy wave than a bifurcation: the gap between the miners with an exit and the miners without one is widening into a canyon.

James Butterfill, head of research at CoinShares, expects some listed miners to earn as much as 70 percent of their revenue from AI by the end of 2026, up from roughly 30 percent today. If he is right, the companies that call themselves Bitcoin miners at the end of this shakeout may be a smaller, richer, and very different set than the ones that entered it, and the margin that matters most for their survival may no longer be the mining margin at all.

Frequently Asked Questions

Are Bitcoin miners profitable in 2026?

On a cash basis, the most efficient miners still are; on a full-cost basis, many are not. CoinShares put the weighted-average cash cost to mine one Bitcoin near $79,995, above the roughly $63,000 price in mid-August 2026, and estimated that 15 to 20 percent of older machines were losing money even before depreciation. Riot Platforms is typical: cash-profitable to mine in the second quarter, but 126.5 percent of production value once depreciation is included.

What is a good Bitcoin mining margin?

There is no single figure, because miners report gross, cash, and all-in margins. The number that decides whether a machine stays switched on is the cash margin, revenue minus mostly electricity; the number that decides whether the business creates value is the all-in margin, which adds depreciation and overhead. A modern rig on cheap power can be cash-positive while the same operation loses money on a full-cost basis.

What is hashprice in Bitcoin mining?

Hashprice is a miner’s expected daily revenue per unit of computing power, quoted in US dollars per petahash per second per day. It is set by the block subsidy, transaction fees, the Bitcoin price, and total network hashrate. In mid-August 2026 it sat near $31.73 per PH per day, about half its mid-2025 peak and close to a five-year low.

Why are Bitcoin miners pivoting to AI?

A megawatt of energized, permitted power earns far more under a long-term AI or high-performance-computing lease than it does mining Bitcoin at a five-year-low hashprice. Miners already control scarce power and grid connections, so leasing that capacity to AI tenants turns a thin, volatile margin into a fixed, multi-year one. The sector has announced more than $70 billion in AI and HPC contracts, and Riot signed a single 20-year lease worth about $9.1 billion in August 2026.

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

Cash cost counts the money that actually leaves the business to produce a coin, dominated by electricity; all-in cost adds depreciation of the machines and buildings plus corporate overhead. The gap is mostly depreciation. Riot mined Bitcoin for $49,912 in cash but $90,631 all-in in the second quarter of 2026, which is why the same coin can look profitable and unprofitable at the same time.

Yuki Tanaka is a senior markets writer at HOGE Wire covering Bitcoin mining, energy, and the economics of proof-of-work.