Why Bitcoin Miners Are Pivoting to AI and What It Means for BTC

Bitcoin miners are pivoting to AI because a megawatt of power now earns far more when it is rented to an AI company than when it runs mining machines. On estimates from the crypto asset manager CoinShares, the same megawatt earns roughly $1.5 million a year in profit hosting AI against about $0.5 million mining bitcoin. Converting a site costs far more than building a mine, but the rent is contracted for years, usually 15 or more, while mining barely covers its costs. CoinShares put the listed miners' weighted average cash cost of producing one bitcoin, excluding tax, at about $75,500 in the second quarter of 2026, above the roughly $71,700 a coin that Riot and CleanSpark actually received. BTC trades at $75,719 on CoinGecko as of 7 November 2026, and mining has become harder since.

So a growing share of the publicly listed miners, the mining companies whose shares trade on the stock market, are turning into data centre landlords. One, Keel Infrastructure, formerly Bitfarms, has already stopped mining, IREN plans to follow by the end of 2026 and Cipher Digital is expected to by the end of 2027. For a BTC holder the short answer is reassuring in one respect and worth watching in two others. The network has not lost its security, because hashrate, the total computing power working on Bitcoin and what makes it costly to attack, is still around 1,000 exahashes per second (EH/s). One EH/s is a million terahashes per second (TH/s), the unit a single mining machine is measured in, and machines switched off in Texas tend to be resold and switched back on somewhere else. What deserves watching is where that mining moves, and how much bitcoin the miners themselves are selling to pay for the conversion.

Below are the economics, the deals and what the shift means for Bitcoin, miner stocks and holders who once saw mining as the way to earn on their BTC.

Why mining stopped paying

A miner's income comes from two places, the block subsidy (the new bitcoin paid out with each block) and transaction fees, and both are weak. The April 2024 halving cut the subsidy from 6.25 to 3.125 BTC per block at block 840,000, as CoinGecko's halving page records, so the network pays out about 450 new bitcoin a day instead of 900. Fees have not filled the gap. Luxor's October review counted fees at 0.57% of total block rewards, the sixteenth month in a row below 1%.

Competition did the rest. Mining difficulty, the network setting that controls how hard it is to find a block, read 148.36 trillion on mempool.space on 7 November 2026, after what Luxor counted as an 11.76% rise in October alone, the largest monthly jump since February 2024. The same chart shows about 86 trillion just before the halving and a record of about 156 trillion in late October 2025. More hashrate chasing a halved subsidy means each machine earns less, and the BTC price has not compensated. Bitcoin peaked at about $126,000 in October 2025 and is roughly 40% below that level today.

The result shows up in hashprice, the revenue a miner earns per petahash per second (PH/s, a thousand TH/s) of computing power each day. It sat at about $32 on 7 November 2026 according to the mining data firm Luxor's Hashrate Index, close to the lows of June, when Luxor recorded a daily all-time low of $27.74 on 6 June and a record monthly average of $30.37. October averaged $37.97 and closed at $32.98. At those levels a large share of the global fleet runs at or below breakeven, which is why the listed miners started asking a different question about their sites. If you have wondered whether a home rig still makes sense, our mining profitability breakdown runs the numbers for retail miners, and they are worse than for the industrial operators described here.

None of this is unusual for the four-year cycle. Every halving squeezes miners. What is new this time is that the squeezed miners have a buyer for their most valuable asset.

What AI companies actually want from miners

They do not want the mining machines. Bitcoin mining runs on ASICs, chips built to compute SHA-256 hashes, the one calculation Bitcoin mining uses, and they cannot train or run AI models. What AI developers want is everything around those machines, the land, the substations and above all a working connection to the power grid.

That connection has become the scarce part. CoinShares, in its second-quarter mining report, notes that data centres account for 87% of a 410 GW queue of large loads waiting to connect to ERCOT, the Texas grid, and that transformer lead times now exceed 160 weeks, more than three years. A miner that spent years securing a substation and a power contract holds something a cloud company with plenty of cash still cannot buy quickly.

Turning that site into an AI data centre is not cheap, though. AI chips need liquid cooling, fast networking, backup power and far stricter uptime than a mine, where machines can be switched off whenever power gets expensive. CoinShares estimates the conversion at $8 million to $15 million per megawatt, against $0.7 million to $1 million per megawatt for mining infrastructure, and that covers the building only, before any GPUs, the chips AI actually runs on, are bought. Miners have taken two broad routes. Most lease the powered building to a tenant who brings its own computers, a model known as colocation. A few, the miner IREN above all, buy the GPUs themselves and sell computing time, which brings higher revenue and a much bigger hardware bill.

The main deals and how they are structured

The table below lists the largest deal each company had announced by early November 2026, per its own release, sorted by stated value. The values are totals over the full base term, not annual revenue, and most of the capacity is still being built. Fluidstack and CoreWeave, which come up below, are AI cloud companies that rent GPU capacity to AI developers.

Three details in the releases matter more than the headline totals.

The first is who carries the build cost. In Core Scientific's earlier CoreWeave agreements, about 590 MW over 12 years worth roughly $10.2 billion, CoreWeave funds most of the conversion, including part of Core Scientific's own share, and gets that share back as credits against its future hosting fees, capped at $1.5 million per megawatt. The tenant effectively helps fund the building and then pays lower rent for a while.

The second is the credit behind the rent. Fluidstack is far smaller than Google or Microsoft, and lenders financing a multibillion dollar build want a stronger name behind the payments. When TeraWulf leased about 360 MW at its Lake Mariner campus to Fluidstack in 2025, Google had agreed by 18 August 2025 to backstop about $3.2 billion of Fluidstack's obligations and received the right to buy a stake of about 14% in TeraWulf. Two of the 2026 deals lean on the tenant's own credit instead. TeraWulf expects the Anthropic lease to be supported by an investment grade credit, and Hut 8 says all of its AI capacity is leased to or backstopped by investment grade counterparties. Hut 8 also puts Beacon Point at up to $50.2 billion if every renewal option is used, which is a good example of why the headline number deserves a second look.

The third is the gap between signing and billing. To serve IREN's contract with Microsoft, IREN agreed to buy GPUs and equipment from Dell for about $5.8 billion. In its latest results IREN reported $200 million of AI cloud revenue for the quarter to 30 September 2026 against $28 million from bitcoin mining, but it also moved its target of more than $4 billion in annualised revenue from year end 2026 to January 2027, possibly February. Cipher's Barber Lake halls are now due between the fourth quarter of 2026 and early 2027, after tenant change orders. Across the companies it covers, CoinShares counted more than $100 billion of disclosed contracts and more than 4 GW of contracted capacity in its September report, but only around 550 MW actually billing, about $1.1 billion a year of revenue. That gap is closing slowly. Core Scientific alone reported 485 MW billing at the end of September in its third-quarter results, while Riot has since added 910 MW to the contracted side.

Riot's deal, announced on 4 November 2026, shows how far the model has moved. Its third-quarter release describes a lease of its entire Corsicana campus in Texas, where Riot delivers land, power infrastructure and powered pads, and the tenant installs, owns and operates the computing equipment. Rent on the full 910 MW is payable whether or not the tenant uses the space, and the first 400 MW is due for delivery in September 2027. At the other end, Bitfarms, renamed Keel Infrastructure in April, shut its last mining sites on 29 June 2026. CoinShares says that makes it the first listed miner headed for a quarter of zero mining revenue, and Keel still had no signed tenant as of that September report.

What the pivot means for Bitcoin's security

The worry you will see most often is that miners leaving for AI will leave Bitcoin less secure. So far the data does not show that. CoinShares counts a peak of about 1,160 EH/s in early October 2025 and a fall to about 850 EH/s by early February 2026, a drop of roughly 27% that it describes as economic capitulation rather than a regulatory shock, and the first half of 2026 was the first six-month decline since China's 2021 mining ban. The network has since recovered to about 1,000 EH/s on mempool.space as of 7 November 2026.

The reason is that hashrate changes hands rather than disappearing. When a listed miner converts a building, its machines are usually sold and plugged in somewhere with cheaper power. Cointelegraph's magazine report on the pivot found exactly that, "the machines get resold, redistributed and eventually switched back on." Luxor estimated about 1,390 EH/s of installed ASIC capacity in October, with about 323 EH/s sitting idle, close to the size of its three least efficient machine groups combined, which would need a much higher hashprice to switch back on.

Bitcoin's difficulty adjustment, which resets every 2,016 blocks, roughly every two weeks, handles the rest. If machines leave, blocks slow down, difficulty falls and the remaining miners earn more per machine, which pulls some idle hardware back in. That self-correction is why a slow exit of listed miners shows up as a reshuffle rather than a collapse.

Two changes are still worth tracking. The first is where hashrate goes. CoinShares expects at least 35 EH/s, about 3.5% of today's hashrate, to leave the listed miners as IREN and Cipher finish their exits, and CoinDesk reported in March that the United States, China and Russia control roughly 68% of global hashrate, with countries such as Paraguay and Ethiopia joining the top ten. Hashrate moving from transparent, audited US companies to private operators in other jurisdictions is not a security failure, but it is a change in who sits behind the network. The second is the security budget, meaning the total paid to miners each day, which is what keeps enough machines running to make an attack too expensive. It is set by the subsidy, fees and the BTC price, not by who mines. The subsidy halves every four years and fees are below 1% of rewards, so unless fees or the price rise, that budget shrinks. That is the long-run question for Bitcoin, AI or no AI, and it sits alongside other slow-moving questions like quantum computing.

There is also a counterargument. Charles Schwab's research team argued in May that hybrid miners, which keep mining while hosting AI, are less likely to be forced to sell bitcoin in a downturn because the AI rent keeps paying, which supports long-term network security. That point is about forced selling. The selling described below is a different kind, done by choice to fund construction, and it is happening anyway.

Miners are selling their bitcoin to fund the build

The pivot needs capital, and the cheapest capital many miners had was their own bitcoin. Riot sold 20,264 BTC in the first nine months of 2026 for about $1.5 billion, mainly to fund data centre development, and ended September with 2,418 BTC, according to its third-quarter release. On our own arithmetic, that is about 74 BTC a day over those nine months, roughly a sixth of the 450 or so new bitcoin the network issues daily, from a single company. Keel sold 1,085 BTC between April and early August and, per CoinShares, plans to sell the rest before the year ends. CoinShares also reports that Cipher Digital sold $123.4 million of bitcoin in the first half at a realised loss and that IREN holds no bitcoin at all.

This is supply that used to sit on company balance sheets and is now reaching the market, part of a wider treasury shakeout among listed BTC holders this year. It also removes a group of companies, some of which once kept most of what they mined, from the ranks of long-term holders. Our look at corporate treasuries under macro pressure covers the other side of that ledger.

What it means for investors in miner stocks

Nothing here is investment advice, but the shift changes what a miner stock is. In its September report, CoinShares found that companies with contracted AI capacity traded at an average of 12.9 times their expected sales for the next twelve months, counting debt as well as shares, against 3.7 times for miners without contracts, so the market now prices them as data centre developers whose value depends on delivering buildings, collecting rent and refinancing debt. That premium cuts both ways, as the average 35% fall in these shares in October 2026, a VanEck figure reported by CoinDesk, showed, and if what you wanted from a miner stock was exposure to bitcoin, the pivot has largely removed it.

If you hold BTC and wanted mining income

The pivot is a useful reality check for anyone who still pictures mining as a way to earn on bitcoin. Assume a modern machine producing 200 TH/s and drawing about 3.5 kW, roughly 17.5 joules per terahash. At today's network hashrate of about 1,000 EH/s it would earn roughly 0.00009 BTC a day through a mining pool, so one whole bitcoin would take around 30 years even at constant difficulty, and longer in practice, since the next halving, expected in 2028, cuts the block reward in half again. At a hashprice near $32 it brings in about $6.50 a day before electricity. Power decides the rest. On Luxor's October figures, machines in the 14 to 19 joules per terahash range earned about 9.6 cents for every kWh they used at the month's average hashprice, the newest machines below 14 joules per terahash about 13 cents, and less by month end, while older machines at about 32 joules per terahash broke even only below about 4.3 cents at the October close. US households paid about 18 cents per kWh on average in August 2026, according to the EIA. On our arithmetic, the example machine uses about 84 kWh a day, roughly $15 of household power against $6.50 of revenue.

Bitcoin itself offers no native way to earn on coins you already hold, since it runs on proof of work and has no staking, as our explainer on staking bitcoin covers. Holders who want BTC income without hardware look at yield strategies instead, and our BTC yield guide compares the main routes.

The listed miners swapped income that swings with hashprice for contracted rent. A holder who wants income from BTC faces a smaller version of the same choice, between staying fully in the coin and accepting a strategy that holds part of it in stablecoins. EarnPark tags each strategy with a Risk Label describing the kind of risk it takes. BTC Maker Core targets 5% APY as of November 2026 under the Market-Neutral label. That target is paid only when earned, and Market-Neutral describes the trades rather than your deposit, which stays in BTC and fully exposed to its price. Withdrawals go through a 30-day bonding period, meaning funds stay in the strategy, still earning, for 30 days after you ask to withdraw. The trades behind it are similar to the ones in our funding rate explainer.

The higher published rate, up to 10% APY as of November 2026, comes from BTC Alpha Vault under the Market-Exposed label, and it sits closer to the miners' trade. It holds roughly half BTC and half stablecoins, rebalances between them, and earns mainly from trading fees and from acting as the counterparty to traders on perpetual futures, crypto futures with no expiry date. That design gives up part of the upside in a strong rally, and counted in bitcoin it can leave you with fewer coins than you started with, yield included. Its rate is an estimate, not a promise, and withdrawals settle monthly, with requests due at least 48 hours before the settlement date.

Both are the published rates any account gets holding zero PARK, EarnPark's own token, so reaching them does not depend on buying it. The strategies use leverage, run on large centralised exchanges and onchain venues, carry no deposit insurance and can produce negative results, so read the risk statement before choosing between them.

What to watch next

The pivot will be judged on delivery, not announcements. The numbers that tell you how it is going are the megawatts actually billing against the more than 4 GW CoinShares counted as contracted in September, IREN's revised January 2027 target, the first rent from Riot's Corsicana campus in late 2027, and whether miners keep selling their remaining bitcoin. On the network side, watch hashprice and the difficulty adjustments, starting with the one due in mid-November. A BTC price recovery would make mining attractive again for the operators who kept their fleets, while a long stretch at today's prices would push more of them toward AI or out of the business.

For Bitcoin, the most likely outcome is the one the data already shows. Hashrate migrates to whoever has the cheapest power, the difficulty adjustment absorbs the shift, and the listed miners become power and data centre companies with a smaller mining side business. For holders, the takeaway is simpler. The listed miners are leaving because they found a better tenant for their grid connection, a choice no home miner has, and at household power prices mining is not a good way for an individual to earn on bitcoin either.

If you would rather put idle BTC to work than buy a rig, you can Start earning once you have read the strategy terms.

FAQ

What is the future of bitcoin miners?

The largest US listed bitcoin miners are becoming power and data centre companies with a smaller mining side business. Riot, Hut 8, TeraWulf, Core Scientific, IREN and Cipher Digital have all signed long-term AI contracts, Keel Infrastructure stopped mining in June 2026, and several are selling their bitcoin to fund the build.

Can bitcoin mining machines be used for AI?

Bitcoin mining machines cannot be used for AI, because they are ASICs built only to compute the SHA-256 hashes Bitcoin needs. What miners repurpose is the site, meaning the land, substations and grid connection. Converting it to an AI data centre costs $8 million to $15 million per megawatt on CoinShares estimates, and the GPUs that do the AI computing are bought on top of that, by the tenant or the miner.

How long does it take one miner to mine one bitcoin?

One mining machine producing 200 TH/s would take around 30 years to mine one bitcoin at the network hashrate of about 1,000 EH/s seen in November 2026, assuming difficulty stayed constant. In practice small miners join a pool and receive tiny daily payouts instead of waiting for a whole block. Difficulty usually rises over time, and the next halving, expected in 2028, cuts the block reward in half again, so the real wait is far longer.

Is it still worth it to mine bitcoin in 2026?

Mining bitcoin in 2026 is worth it only for operators with very cheap power and efficient machines, and even many of those are turning to AI. On Luxor's October figures even the most efficient machines earned only about 13 cents per kWh of power used, while US households paid about 18 cents on average, so a home miner usually loses money on electricity alone.

Does the AI pivot make Bitcoin less secure?

The AI pivot has not made Bitcoin less secure so far, since network hashrate recovered to about 1,000 EH/s by November 2026 after falling earlier in the year. Machines from converted sites are typically resold and switched on elsewhere. The longer-term questions are where hashrate concentrates and how the security budget, the total paid to miners, holds up as the block subsidy keeps halving.

What happens when all 21 million bitcoins are mined?

When all 21 million bitcoins are mined, around the year 2140 on CoinGecko's schedule, miners will be paid by transaction fees alone. That is why the fee share matters. In October 2026 fees made up only 0.57% of block rewards, so for Bitcoin to stay as costly to attack, fees or the BTC price would have to rise a long way as the subsidy keeps halving.