Miners are no longer mining Bitcoin; they are selling electricity to AI.

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

By Cathy

Mining one Bitcoin costs $87,000. When sold, the market only pays $67,000.

For each Bitcoin mined, there’s a net loss of $20,000. It’s not just fees or electricity price fluctuations—it’s a real, tangible loss of $20,000 per Bitcoin produced. This is the reality as of March 2026. Data from Glassnode and MacroMicro point to the same conclusion: Bitcoin mining, at current prices, is a losing business.

But miners haven’t just sat around waiting to die. They’ve made a surprising choice—stop mining and sell electricity to AI.

More precisely, it’s not “stop mining,” but rather emptying the Bitcoin treasury and pouring all funds into AI data centers, relegating mining to a side business.

Since Bitcoin’s peak of $126,000 in October 2025, publicly listed mining companies have sold over 15,000 Bitcoins. This isn’t sporadic cash-out; it’s an organized, strategic large-scale retreat.

Where did the 15,000 BTC go?

Core Scientific was the earliest and most decisive mover.

In January 2026, it sold about 1,900 Bitcoins, cashing out $175 million. The plan is to liquidate all remaining holdings in Q1. This once-bankrupt and restructured mining company is now transforming its Texas mines into high-density AI hosting facilities, aiming to allocate all 1.3 GW of total power capacity to AI.

MARA is even more aggressive. Known for “never selling coins,” it quietly changed its treasury policy in its March 2026 10-K report—53822 Bitcoins, fully authorized for sale. At the then-current price, nearly $4 billion worth of assets shifted from “strategic reserves” to “liquid funds” overnight. Soon after, MARA signed a joint venture with Starwood Capital to deliver 1 GW of AI data center capacity.

The most surprising is Cango. Originally a Chinese auto finance platform, it entered Bitcoin mining only at the end of 2024, and by February 2026, it sold 4,451 Bitcoins—60% of its reserves—raising $305 million to pay debts and fund AI transformation. It also hired former Zoom executive Jack Jin as CTO of AI operations, planning to deploy containerized GPU compute nodes across global mines. A car loan company turned miner, then AI inference service provider—such cross-industry shifts are unique to the crypto world.

BitDeer’s move appears to be a calculated chess move. It cleared its Bitcoin holdings in February. Founder Wu Jihan’s straightforward response: holding zero doesn’t mean forever; right now, liquidity is needed to seize opportunities in electricity and land acquisitions. Unlike others, BitDeer is both clearing out and accelerating—its Bitcoin production in January surged 430% year-over-year, with self-operated hash rate reaching 63.2 EH/s, surpassing MARA to become the largest publicly listed miner by hash rate. Selling off coins has fueled massive expansion in hash power and infrastructure—decisiveness with a bold, ambitious edge.

The same electricity, now worth 10 times more for AI

Why are miners so uniformly selling off? Because the numbers are clear once you do the math.

Mining is losing money, but miners hold one valuable asset that everyone is competing for: electrified land.

After the 2024 halving, Bitcoin mining profitability shrank from over 90% at its peak to breakeven. Meanwhile, demand for electricity and data centers for AI exploded. According to MarketsandMarkets, the global AI inference market is projected to grow from about $106 billion in 2025 to nearly $255 billion in 2030.

Morgan Stanley calculated that shifting 1 MW of power from mining to AI hosting could yield a valuation premium of over 10 times.

This isn’t an exaggeration. AI hosting contracts are typically long-term—10 to 15 years—with clients like Microsoft and Meta, offering stable, predictable cash flows. In contrast, mining income depends entirely on coin prices—and you know how volatile they are.

Wall Street has already spoken with real money. Morgan Stanley extended a $500 million loan to Core Scientific, with an option to increase to $1 billion. This isn’t a loan to a “crypto company,” but a credit backing for a “digital infrastructure company.” TeraWulf and Cipher Mining, thanks to their successful hybrid models, are rated as “overweight” by Morgan Stanley, while MARA, once a steadfast Bitcoin holder, was downgraded due to overexposure to coin price risks.

The message from capital markets is clear: in Wall Street’s view, these companies’ value is no longer based on how many Bitcoins they hold, but on how much electricity they control.

On-chain indicators suggest the bottom may be near

Miners’ collective sell-off has caused market distress. But on-chain data reveals some interesting signals.

Hash Ribbon indicator inverted starting late November 2025, and by February 2026, it had persisted for three months—one of the longest miner capitulation periods in history. The last similar signal was in December 2022, when Bitcoin bottomed at $15,500. As of early March, the 30-day moving average is approaching above the 60-day moving average, signaling a potential recovery.

The MVRV Z-Score remained between 0.43 and 0.49 in early March. This metric measures the deviation of market price from “realized value.” Historically, when Z-Score drops into 0–1, it almost always signals a strategic buying window.

Puell Multiple dropped to around 0.6, indicating miners’ daily revenue has been compressed to about 60% of the annual average. It’s approaching the 2022 bear market bottom of 0.3, with profit margins squeezed to historic lows.

The most extreme signals come from sentiment indicators. During February’s “Bitcoin Polar Vortex,” the Crypto Fear & Greed Index plunged to 5, and on February 5, a single-day correction resulted in a $3.2 billion loss, a record.

All four independent indicators flashing red—when this last happened, Bitcoin was forming a bottom.

Miner selling might actually be bullish?

This is the most counterintuitive part of the story.

In the past, miner sell-offs were seen as bearish signals—these “original sellers” would mine and immediately sell, exerting persistent downward pressure. But in 2026, the nature of these sales has changed: after selling Bitcoin, these miners are now earning US dollars from AI.

Think about what this means. Previously, Core Scientific needed to sell hundreds of Bitcoins monthly to cover electricity and operational costs. Now, with long-term contracts with Microsoft and Morgan Stanley’s credit lines, they plan to liquidate only a portion of their remaining holdings (about 2,537 BTC by year-end, most already sold). This isn’t passive “selling to survive,” but active liquidation and reinvestment into AI infrastructure. Once the MARA-Starwood joint project is operational, the $1 billion+ cash flow from the 1 GW data center will cover all costs.

In other words, AI-transformed miners are shifting from being structural sellers of Bitcoin to neutral or even potential buyers. The largest “natural shorts” in the market are exiting permanently.

Bitcoin mining itself hasn’t disappeared; it’s just changing form. MARA’s hybrid model points the way: mine when electricity is cheap, switch to GPU computing during AI demand peaks. Bitcoin becomes a flexible load and insurance mechanism for the power grid—AI makes money, mining provides a safety net.

Summary

In 2025, Bitcoin’s network hash rate just surpassed 1 Zettahash. In the short term, some mines transitioning to AI will slow hash rate growth—Cango, for example, has 31% of its hash rate offline for upgrades. But this is a healthy capacity cleanup: inefficient miners exit, leaving more efficient, focused players, which actually enhances network security.

This isn’t miners surrendering; it’s the evolution of mining.

When mining becomes a side business and AI takes the main stage, Bitcoin loses a group of miners forced to sell, but gains a more sustainable supply structure.

The miners’ Bitcoin holdings are exhausted, but electricity remains.