Losing $19K per Coin Mined, Bitcoin Mining Firms Collective AI Defection

Original Article Title: Bitcoin miners are becoming AI companies and selling their BTC to fund the transition

Original Article Author: Shaurya Malwa, CoinDesk

Original Article Translation: DeepSea TechFlow

DeepSea Summary: CoinShares' latest mining report shows that the weighted average cost for a publicly listed mining company to mine one Bitcoin has risen to around $80,000, while the current price of BTC is in the $68,000-$70,000 range — meaning a loss of around $19,000 for each mined coin. The following is the content of the original article:

The industry is undergoing its most fundamental transformation since inception: over $70 billion in AI/HPC contracts have been signed, publicly listed mining companies have collectively sold over 15,000 BTC, companies like IREN and TeraWulf have taken on billions of dollars in debt. By the end of 2026, AI revenue share of some mining companies could reach 70%. They are transitioning from Bitcoin miners to data center operators that happen to still mine. The core contradiction is that the companies securing the Bitcoin network's security are the ones selling coins to transition to AI, with hash rate plummeting from a peak of 1,160 EH/s to around 920 EH/s.

· The Bitcoin mining industry is undergoing its most fundamental transformation since inception, with the clearest signal not being hash rate or difficulty adjustment, but the balance sheet.

· CoinShares' Q1 2026 mining report released this week shows that the weighted average cash cost for publicly listed mining companies to mine one Bitcoin in Q4 2025 rose to around $79,995.

· Bitcoin has been trading in the $68,000-$70,000 range, with a CoinDesk report last week estimating a loss of around $19,000 for each mined BTC.

· This figure is not sustainable, an industry well aware. The response is a full pivot to AI infrastructure — reshaping the very nature of these companies.

According to the CoinShares report, publicly listed mining companies have collectively announced over $70 billion in AI and High-Performance Computing (HPC) contracts. CoreWeave and Core Scientific's expanded agreement is valued at $10.2 billion over 12 years. TeraWulf has secured $12.8 billion in HPC contract revenue. Hut 8 has signed a $7 billion, 15-year AI infrastructure lease at the River Bend facility. Cipher Digital's agreement with Google-backed Fluidstack is worth billions.

By the end of 2026, publicly listed mining companies' AI revenue share could reach as high as 70%, up from the current approximately 30%. Core Scientific's AI hosting revenue already accounts for 39% of its total revenue. TeraWulf is at 27%. IREN is currently at 9% but is rapidly expanding, with a planned liquid-cooled GPU hash capacity of up to 200 megawatts.

This means these mining companies are increasingly resembling data center operators who happen to also mine Bitcoin.

The economics account for this shift. CoinShares data shows that the cost of Bitcoin mining infrastructure is roughly $7-10 million per megawatt, while AI infrastructure is around $8-15 million per megawatt. The difference is significant, but AI offers a more structurally higher and stable return.

Hash Price — a measure of a miner's revenue per unit of hash power — hit a historic low in early March post-halving, at around $28-30 per PH/day.

At this level, miners using older-generation machines need an electricity price below $0.05/kWh to sustain cash profitability. In contrast, AI infrastructure contracts promise a profit margin of over 85% with multi-year revenue visibility.

Where Is the Transition Money Coming From

A report from CoinShares highlights two clear sources of funding for this transition.

First, debt. The industry's leverage has undergone a qualitative shift. IREN now carries $3.7 billion in convertible notes across five series. TeraWulf has a total debt of $5.7 billion, consisting of convertible bonds and priority guaranteed notes from its hash power subsidiaries.

Cipher Digital issued $1.7 billion in priority guaranteed notes in November, leading to its quarterly interest expenses skyrocketing from $3.2 million in the first nine months to $33.4 million in just Q4. This is not mining-level debt burden; this is infrastructure-level stakes — betting that AI revenue can ramp fast enough to cover debt obligations.

Second, coin selling. Publicly listed mining firms have collectively sold over 15,000 BTC from peak levels. In January, Core Scientific sold about 1,900 BTC (worth $175 million) and plans to liquidate almost all remaining holdings by Q1 2026. Bitdeer zeroed out its holdings in February. Riot Platforms sold 1,818 BTC in December (worth $162 million).

Even the largest publicly traded holder Marathon (which holds 53,822 BTC) quietly expanded its policy in the March 10-K annual report, authorizing sales from the entire balance sheet reserve. Part of the reason is the pressure from its $350 million Bitcoin collateralized loan facility — as the price fell to $68,000, the loan-to-value (LTV) ratio has risen to 87%.

Who Guards the Bitcoin Network?

Those selling coins to do AI are precisely the companies that mine and secure the Bitcoin network. This poses the central paradox of this transformation. When mining is unprofitable and AI is very profitable, the rational economic decision is to move funds away from mining. But if enough miners do this, the network's security budget will shrink.

The hashrate data has already reflected this. The network's hashrate peaked at around 1,160 EH/s in early October 2025, then dropped to around 920 EH/s, experiencing three consecutive negative difficulty adjustments — the first since July 2022.

Valuation Divergence

The market has priced in this divergence. Mining companies with signed HPC contracts are currently trading at 12.3 times the next 12 months' revenue. Pure mining companies are only at 5.9 times. The market is paying over a two times premium for AI exposure, further reinforcing the transformation incentive.

The geographic landscape is also changing. The U.S., China, and Russia currently control around 68% of the global hashrate. In just one quarter, Q4, the U.S. increased its market share by about 2 percentage points. But emerging markets are also joining — Paraguay and Ethiopia have entered the top ten mining nations globally, driven by HIVE's 300 MW and Bitdeer's 40 MW facilities, respectively.

Hashrate Forecast

CoinShares forecasts network hashrate to reach 1.8 ZH/s by the end of 2026, rising to 2 ZH/s by the end of March 2027 (one month later than previously predicted).

But this forecast is premised on Bitcoin returning to $100,000 by year-end. If the price remains below $80,000, CoinShares expects hashrate prices to continue falling, hashrate to decline further, and more miners to exit. Persisting below $70,000 could trigger a larger-scale capitulation event — ironically, this would benefit survivors by lowering the difficulty.

The next-generation hardware offers a potential way out. Bitmain's S23 series and Bitdeer's in-house SEALMINERA3 have an energy efficiency of less than 10 joules/TH, with mass shipments expected in the first half of 2026. These miners, compared to the current mainstream mid-generation models, are projected to roughly halve the energy cost per bitcoin. However, deploying them requires funding, and many miners are currently redirecting funds towards AI.

When the Bitcoin mining industry entered this cycle, it was a group of companies securing the network and hodling bitcoin. It is now exiting this cycle in a different guise: a cohort of companies building AI data centers and selling bitcoin to finance themselves.

Is this merely a temporary response to unfavorable economic conditions, or a permanent structural shift? It depends on one variable: the price of bitcoin. If it returns to $100,000, mining profitability rebounds, and the pace of AI transformation slows down. If it stalls at $70,000 or lower, the transformation accelerates, and the mining industry, which has been mining-centric for the past decade, will continue to morph into something entirely different.

Original Article Link