Mining one Bitcoin results in a loss of 20,000! Miners collectively shift to AI, rewriting the Bitcoin mining model

In March 2026, Bitcoin mining faced a brutal inverted profitability crisis: the total cost to mine one Bitcoin was about $87,000, while the market price was only around $67,000, resulting in a net loss of $20,000 per Bitcoin produced. The response from mining companies was unexpected—they didn’t wait for the price to rebound but systematically emptied their Bitcoin reserves, shifting their core assets toward AI.

Who’s Selling, Where Are They Going: Four Mining Companies’ Transformation Paths

Core Scientific was the earliest and most decisive mover. In January 2026, it sold approximately 1,900 Bitcoins, cashing out $175 million, and planned to liquidate its remaining holdings in the first quarter. This company, which had previously gone through bankruptcy restructuring, is transforming its Texas mines into high-density AI hosting facilities, aiming to switch all 1.3 GW of total power capacity to AI—Morgan Stanley provided a $500 million credit line (expandable to $1 billion).

MARA (Marathon Digital Holdings) has the most symbolic transformation. Known for “never selling coins,” the company quietly amended its treasury policy in its 2026 annual report to allow the sale of all 53,822 Bitcoins (worth about $4 billion at the time), and signed a joint venture with Starwood Capital to deliver 1 GW of AI data center capacity.

Cango represents a more thorough identity shift. Formerly a Chinese auto finance platform, it entered Bitcoin mining only at the end of 2024. By February 2026, it had sold 4,451 Bitcoins (60% of its reserves), cashing out $305 million to pay down debt and fund AI transformation. It also hired former Zoom executive Jack Jin as CTO of AI operations, planning to deploy containerized GPU computing nodes across global mines.

Bitdeer’s choice reflects another logic—liquidating its on-paper holdings while accelerating capacity expansion. Founder Wu Jihan admitted that zeroing out holdings was to free liquidity to seize opportunities in power and land acquisitions; in January, Bitcoin production surged 430% year-over-year, with self-mining capacity rising to 63.2 EH/s, surpassing MARA to become the largest publicly listed self-mining company globally. This is a dual act of “biting the bullet” and “loading the guns.”

Why Is the Same Kilowatt in AI Worth 10 Times More?

The business logic behind mining companies shifting to AI is the re-pricing of computational resources. The AI inference service market is projected to grow from about $106 billion in 2025 to approximately $255 billion by 2030. More importantly, AI data center contracts are fundamentally different: long-term agreements of 10–15 years, enterprise clients like Microsoft and Meta, stable and predictable dollar cash flows, completely decoupled from coin prices. Meanwhile, Bitcoin mining revenue, affected by equipment depreciation, difficulty adjustments, and coin price volatility, has been compressed—after the 2024 halving, profit margins shrank from over 90% at peak to break-even levels.

Capital markets have signaled their stance through actions: Morgan Stanley rated mining companies like Core Scientific and TeraWulf, which successfully integrated AI models, as “Overweight,” while those overly dependent on coin prices were downgraded. The valuation basis is shifting from “how many Bitcoins held” to “how much power is controlled.”

On-Chain Indicators Hint That the Bottom May Be Near

While miners are collectively liquidating, on-chain data is forming a set of signals indicating a potential bottom:

Hash Ribbon: Since inversion in late November 2025, nearly three months ago, it’s one of the longest miner capitulation periods in history; the last similar signal appeared in December 2022, after which Bitcoin bottomed at around $15,500.

MVRV Z-Score: Maintained between 0.43 and 0.49 in early March, historically this range has almost always corresponded to strategic accumulation opportunities.

Puell Multiple: Dropped to about 0.6, meaning miners’ daily revenue is only 60% of the annual average; close to the 0.3 bottom during the 2022 bear market.

Market Sentiment Extremes: During the “Polar Vortex” in February, the Crypto Fear & Greed Index hit 5; on February 5, a single-day realized loss set a record of $3.2 billion.

All four independent indicators simultaneously reaching extreme zones is a rare occurrence in history.

Frequently Asked Questions

Why might large-scale miner selling actually be bullish for Bitcoin?

Historically, miners have been the largest “structural sellers” of Bitcoin—they mine and need to sell coins to pay for electricity and operational costs. After transitioning to AI, their dollar cash flow comes from AI hosting contracts, so they no longer need to sell coins continuously to sustain operations, turning from forced sellers into neutral or even potential buyers. The largest “natural shorts” in the market are systematically exiting, which is a substantial long-term positive for Bitcoin’s supply structure.

How does MARA’s hybrid mining model switch between AI and Bitcoin?

MARA’s joint venture with Starwood Capital hinges on flexible use of the same power infrastructure: when electricity prices are low, capacity is used for Bitcoin mining; during AI demand peaks, computing resources switch to GPU services. This transforms Bitcoin mining from a core business into a “flexible load balancer,” covering electricity costs when AI isn’t needed and earning higher profits when AI demand is high.

Will the Bitcoin mining industry decline because of this?

In the short term, some mines transitioning to AI will slow capacity growth (e.g., Cango’s 31% capacity reduction for upgrades). But from an industry evolution perspective, this is a healthy capacity clearing: inefficient miners unable to afford high electricity costs exit, leaving more efficient and specialized players. Overall network security does not decline—in fact, it improves. Bitcoin mining is evolving from “must be profitable to survive” into a “grid flexibility mechanism,” representing a more resilient long-term model.