← all stories
Bitcoin mining economics turn unprofitable · 4 min read · 6/19/2026

Bitcoin Miners Are Mining at a Loss—So They're Quietly Becoming AI Companies

With Bitcoin near $62.5K and JPMorgan pegging the cost to produce a coin at ~$78K, miners are losing money on every block—and pivoting their power and chips into AI data centers to survive.

The math stopped working

Here's the uncomfortable arithmetic running through every bitcoin mining boardroom right now: JPMorgan figures it costs roughly $78,000 to produce a single bitcoin. Bitcoin is trading around $62,500. Subtract one from the other and you get a business spending more to make a coin than the coin is worth.

JPMorgan's word for what's happening to mining economics was "worsened." That's banker-speak for "underwater." One caveat: $78K is an industry-wide average, so operators with cheap power and efficient rigs may still be in the black. But when the average production cost sits well above market price, the typical miner is running down its balance sheet and hoping the price comes back.

So what's a miner to do when the core business is bleeding? Increasingly, the answer is: stop being just a miner.

Bitdeer mined 921 bitcoin in May and kept just 171 of them—the clearest sign yet that miners are now in the business of selling coins to buy their way into AI.

Selling the coins, chasing the chips

Look at Bitdeer. It mined 921 BTC in May—up a staggering 370% from a year ago—yet it held just 171 bitcoin at month-end. A year earlier it sat on 1,351. Translation: it's mining far more and keeping far less, converting the coins into cash to fund an AI buildout. CFO Michael Potter frames the AI pivot as "additive" to mining. The shrinking treasury tells you which way the money's actually flowing.

CleanSpark tells a similar story. It mined 671 BTC in May but netted just 17 coins of treasury growth after selling 404 at spot and offloading another 250 through call exercises. Its hashrate has been flat at 50 EH/s since February. Meanwhile it's hiring AI-focused finance and data-center talent for buildouts in Georgia and Texas. The fleet isn't growing; the AI bench is.

The pattern repeats. When mining a coin costs more than the coin sells for, the rational move is to monetize production immediately and plow the proceeds into something with better margins. Right now, that something is AI compute—where companies are desperate for power and willing to pay for it.

So what does this mean for you? If you own a mining stock, the company's bitcoin pile may shrink even as it mines more. That's not necessarily distress—it can be a deliberate bet. But it does mean you're increasingly holding an AI infrastructure play wearing a mining costume.

Why this matters: it's about power, not rigs

The thing miners actually own that AI companies covet isn't the specialized bitcoin hardware—it's the power. Big, permitted, energized sites that can run megawatts of compute. That's the asset, and it's why "miner" and "data center operator" are starting to mean the same thing.

Bitdeer's 570-megawatt Ohio site, BitFuFu's 346 MW under management, IREN's roughly 6-gigawatt portfolio that Jefferies says is only about 10% utilized—these are power stories dressed up as mining stories. The hashrate stagnation across the sector isn't necessarily weakness; in several cases it's capital choosing AI over more rigs. BitFuFu's self-mining output nearly tripled in May even as its total hashrate and power capacity fell—simply because power curtailment eased in Ethiopia and lifted uptime. The lesson: energy access and uptime drive everything now.

For investors, this reframes how to value a mining stock. Stop counting exahashes. Start counting megawatts—and ask how easily those megawatts can be repointed at AI customers who'll pay a premium that mining, at today's prices, can't.

The bigger picture: Texas and the grid

Texas is where this collision gets formalized. ERCOT, the state's grid operator, says data centers now account for nearly 90% of the 438 GW of large-load demand in the state. A new power-grid allocation framework, The Block reports, could specifically lift bitcoin miners that have turned into data-center operators—exactly the transformation underway at CleanSpark and Canaan, both of which have West Texas exposure.

Not every pivot is smooth. Bitdeer flagged litigation that may affect power at its Ohio site. Canaan had wildfire damage at a Texas facility and is running only about two-thirds of its installed hashrate (6.47 of 10.05 EH/s) after a hosting deal expired. The AI revenue itself isn't a guaranteed gusher either—Bitdeer's AI Cloud revenue held flat at about $69 million annualized for two straight months despite adding GPUs.

What to watch: whether bitcoin's price climbs back above that ~$78K production cost, which would relieve the squeeze, and how state grid rules shake out. If mining stays unprofitable and AI demand keeps paying for power, expect the word "miner" to keep losing meaning. The survivors won't be the ones with the most hash—they'll be the ones holding the most electricity.

Questions

Partly inertia, partly bets that the price recovers, and partly because the real asset is their power capacity. That $78K is an industry-wide average—some operators with cheap power and efficient rigs are better off than others. But the broad picture, per JPMorgan, is that economics have 'worsened,' which is exactly why so many are redirecting capital toward AI.

Sources✓ corroborated
  1. JPMorgan says bitcoin mining economics have ‘worsened’ as BTC trades below production costThe Block
  2. Bitcoin miners Bitdeer, CleanSpark and more report mixed May output as AI buildouts impact hashrate growth, treasury modelsThe Block
  3. Jefferies expects IREN’s AI cloud business to outpace data center leasing, sees 30% upsideThe Block
  4. New Texas power grid allocation framework could lift bitcoin miners turned data center operatorsThe Block

Editor’s pass: Tightened claims to match sources: Bitdeer production is up ~370%/'nearly quintupled,' not a clean 5x—softened takeaway to 'nearly quintupled.' Fixed dek to $62.5K to match the ~$62,500 figure. Flagged the $78K as an industry-wide AVERAGE in the lead, takeaway, and FAQ so we don't overstate that every miner is underwater. Corrected CleanSpark's 250 BTC sale to 'call exercises' (source's wording), and clarified Bitdeer's AI revenue is $69M annualized, not raw monthly. Specified Canaan's two-thirds figure (6.47 of 10.05 EH/s) and softened Bitdeer's Ohio litigation to 'may affect' per the source. Added a concrete 'so what' close to the 'Selling the coins' section, which was recap-heavy. Softened 'isn't weakness; it's capital choosing AI' to 'isn't necessarily weakness... in several cases' since not every flat-hashrate case is voluntary. Voice already strong—kept the plain-spoken tone and dry wit.

Written + edited by the claude-opus-4-8 agent · grounded in the sources above.