Miners have stopped mining Bitcoin and are now selling electricity to AI.

Written by: Cathy , Blockchain in Plain Language

Mining one Bitcoin costs $87,000. Selling it, the market only gives you $67,000.

For every Bitcoin mined, there's a net loss of $20,000. This isn't a loss from transaction fees or electricity fluctuations; it's a solid, direct loss of $20,000 for every Bitcoin produced. This is the reality in March 2026. Data from Glassnode and MacroMicro both point to the same conclusion: Bitcoin mining, at current prices, is a money-losing business.

But the miners didn't sit idly by and wait to die. They made a choice that the entire market hadn't expected—they stopped mining and sold the electricity to AI.

To be precise, it's not that they "stopped mining," but rather that they emptied the Bitcoin treasury and poured all the funds into AI data centers, relegating mining to a secondary role.

Since Bitcoin turned downwards from its all-time high of $126,000 in October 2025, publicly listed mining companies have sold more than 15,000 Bitcoins. This is not a series of sporadic cash-outs, but an organized and strategic retreat.

Miners collectively liquidated their holdings; where did the 15,000 BTC go?

Core Scientific was the first and most decisive to take action.

In January 2026, it sold approximately 1,900 Bitcoins in one go, cashing out $175 million. The remaining assets were planned to be liquidated in Q1. This mining company, which had previously gone bankrupt and undergone restructuring, is now transforming its Texas mining farms into high-density AI hosting facilities, aiming to allocate its entire 1.3GW of power capacity to AI.

MARA went even further. This company, known for its "never selling its coins" policy, quietly changed its treasury policy in its March 2026 10-K annual report—authorizing all 53,822 Bitcoins for sale. At the time, this amounted to nearly $4 billion, transforming overnight from "strategic reserves" into "disposable funds." Immediately following this, MARA signed a joint venture agreement with Starwood Capital to deliver 1GW of AI data center capacity.

The most surprising case is Cango. This company, originally a Chinese auto finance platform, only entered the Bitcoin mining market at the end of 2024. By February 2026, it had sold 4,451 Bitcoins—60% of its reserves—causing $305 million to pay off debt and fund its AI transformation. It also poached former Zoom executive Jack Jin as its AI CTO and plans to cram containerized GPU computing nodes into its global mining farms. A car loan company transformed into a miner within two years, and then from a miner into an AI inference service provider—this kind of cross-industry speed is only seen in the cryptocurrency world.

BitDeer's choice, however, appears to be a calculated move. In February, it cleared out its entire Bitcoin holdings. Founder Jihan Wu's response was frank: zero holdings don't mean zero forever; they need liquidity to capitalize on opportunities to acquire electricity and land. Unlike other mining companies, BitDeer simultaneously accelerated its growth – its Bitcoin production in January surged 430% year-on-year, and its self-operated hashrate reached 63.2 EH/s, surpassing MARA to become the world's largest listed mining company in terms of self-operated hashrate. Clearing out its holdings resulted in a significant expansion of its hashrate and infrastructure. It demonstrated both the decisiveness of "cutting off one's own arm" and the ambition of "loading ammunition."

The same amount of electricity is used ten times more effectively by AI.

Why did the miners sell off their shares so uniformly? Because after doing the math, the answer was obvious.

Mining is unprofitable, but mining companies have one thing that the whole world is vying for: electrified land.

Following the 2024 halving, Bitcoin mining profit margins were squeezed from over 90% at their peak to the break-even point. However, during the same period, the demand for electricity and data centers from AI experienced explosive growth. MarketsandMarkets predicts that the global AI inference market will grow from approximately $106 billion in 2025 to nearly $255 billion in 2030.

Morgan Stanley calculated that shifting 1 megawatt of electricity from mining to AI hosting could result in a valuation premium of over 10 times.

This is no exaggeration. AI hosting contracts are typically 10 to 15-year long-term agreements, with clients including investment-grade giants like Microsoft and Meta, offering stable and predictable cash flow. In contrast, mining revenue depends entirely on the price of the cryptocurrency—and the price of the cryptocurrency, you know.

Wall Street has already voted with real money. Morgan Stanley provided Core Scientific with a $500 million loan facility, with an option to increase it to $1 billion. This isn't a loan to a "crypto company," it's a credit endorsement for a "digital infrastructure company." TeraWulf and Cipher Mining were rated "overweight" by Morgan Stanley due to their successful hybrid model, while MARA, which had previously clung to Bitcoin, was downgraded for excessive exposure to cryptocurrency price risks.

The signal from the capital markets couldn't be clearer: in the eyes of Wall Street, the value of these companies no longer depends on how much Bitcoin they hold, but on how much electricity they control.

On-chain metrics suggest it may be nearing its bottom.

Miners were selling off their holdings en masse, and the market was in a state of panic. But if you look at the on-chain data, you'll find a very interesting signal.

The hash ribbon has been inverted since the end of November 2025, lasting for a full three months until February 2026—one of the longest periods of miner capitulation in history. The last time a similar signal combination appeared was in December 2022, when Bitcoin bottomed out at $15,500. As of early March, the 30-day moving average is approaching above the 60-day moving average, suggesting a recovery signal is imminent.

The MVRV Z-Score remained between 0.43 and 0.49 in early March. This indicator measures the degree of deviation of market price from "real value." Historically, when the Z-Score falls into the 0-1 range, it almost always corresponds to a strategic entry point.

The Puell Multiple has dropped to around 0.6, meaning that miners' daily revenue has been compressed to about 60% of the annual average. It's not far from the 0.3 level seen at the bottom of the 2022 bear market, and miners' profit margins are being squeezed to historically low levels.

The most extreme signals came from the sentiment side. During the "Bitcoin polar vortex" in February, the Crypto Fear & Greed Index once fell to 5, and on February 5, the single-day loss after entity adjustment set a record of $3.2 billion.

Four independent indicators are flashing red simultaneously. The last time this happened, Bitcoin was forming a bottom.

Is it actually a good thing for miners to sell their coins?

This is the most counterintuitive part of the whole story.

In the past, miner sell-offs were always seen as a bearish signal—these are the "native sellers" of Bitcoin, selling it as soon as they mine it, creating continuous selling pressure in the market. But the nature of the sell-off in 2026 was completely different: after selling their Bitcoin, these mining companies turned to earning dollar revenue from AI.

Consider what this means. Previously, Core Scientific sold hundreds of Bitcoins each month to cover electricity and operating costs. Now, with a long-term contract with Microsoft and a credit line from Morgan Stanley, although it still plans to liquidate most of its remaining Bitcoin holdings (approximately 2,537 Bitcoins at the end of the year, with most already sold), this is no longer a passive "selling Bitcoin to survive," but a proactive liquidation to concentrate funds on AI infrastructure. Once the MARA and Starwood joint venture is completed, the dollar cash flow generated by the 1 GW data center will be sufficient to cover all costs.

In other words, mining companies that are transitioning to AI have transformed from structured sellers of Bitcoin into neutral or even potential buyers. The largest group of "natural short sellers" in the market is permanently exiting the market.

Bitcoin mining itself hasn't disappeared; it's simply changed its form. MARA's hybrid model has pointed the way: mining when electricity prices are low, and switching to GPU computing when AI demand peaks. Bitcoin becomes a "flexible load" and "insurance mechanism" for the power grid, with AI generating revenue and mining providing a safety net.

summary

In 2025, the Bitcoin network's hashrate will just surpass the milestone of 1 Zetahash. In the short term, some mining farms' transformation to AI will indeed lead to a slowdown in hashrate growth—for example, Cango took 31% of its hashrate offline for upgrades. However, this is actually a healthy process of capacity clearing: inefficient miners exit the market, and the remaining players become more efficient and focused, resulting in improved network security rather than decreased security.

This is not the miners' surrender, but the evolution of the mining industry.

When mining becomes a side job and AI becomes the main job, Bitcoin loses a group of miners who are forced to sell their coins, but gains a healthier supply structure.

The miners have sold all their Bitcoin, but the electricity is still available.