Abandoning Mining for AI: The Life-and-Death Race of Bitcoin Miners

Over the past six months, the hottest topic in the Bitcoin mining community is no longer “how much has it gone up again,” but rather “how long can this last?” Bitcoin prices hover around $60,000, yet the cost line is firmly above $80,000—no matter how you calculate it, it’s a loss.

As a result, a large-scale “exodus” has begun, but the direction is unexpected: instead of leaving the market, miners are pivoting into the AI computing power race.

1. Mining at a Loss of $20,000 per Coin, Miners Collectively “Shut Down”

● Consider some stark numbers. As of February 2026, the average total cost to produce one Bitcoin across the network has risen to $87,000, while the coin price is only about $64,000. This means that for each block mined, miners are losing nearly $20,000. This isn’t just thinning profit margins; it’s a situation where producing one unit results in a loss.

● Rosenblatt Securities analyst put it plainly: current mining revenue has fallen to less than $0.03 per terahash, with only the most efficient mining machines still profitable. The key metric measuring miners’ profitability, Hashprice, has dropped about 30% over the past three months, now only $28/PH/day—approaching historic lows.

● The consequence is widespread shutdowns. On February 9, Bitcoin’s network difficulty was adjusted downward by 11.16%—the largest reduction since China’s ban in 2021, nearly four and a half years ago. Difficulty adjustment is an automatic protective mechanism to make mining easier for remaining miners. But with a 45% cost-price inversion rate, this “relief” is barely enough.

● Industry insiders have a “Miner Profitability Sustainability Index,” with 100 being healthy. Currently, this number is 21. In plain terms: aside from a few top players with electricity costs below $0.05 per kWh and using the latest machines, most are operating at a cash flow loss.

2. Believers Liquidate: From “Holding Coins” to “Holding Cash”

● More provocative signals come from industry giants. In late February, Wu Jihan’s NASDAQ-listed miner Bitdeer did something shocking: it completely liquidated all its Bitcoin holdings. It sold all 189.8 BTC mined that week and zeroed out its reserve of 943.1 BTC, cashing out about $63 million.

● This company, which recently became the world’s largest listed self-mining firm with 63.2 EH/s of hash power, chose to hold no coins at all. This was unimaginable in the past—“mining equals hoarding” was a core principle in the mining world. But now, that belief has been broken.

● Bitdeer’s reasoning is pragmatic: with hash prices falling below $30, holding Bitcoin entails huge opportunity costs. Every coin held means less cash available for debt repayment, equipment upgrades, or pivoting to new ventures. They issued $325 million in convertible bonds, explicitly earmarked for debt buybacks, hedging dilution risks, and the rest for AI infrastructure.

● Similarly, Cango liquidated. In early February, it sold 4,451 BTC, cashing out $305 million, which was used to pay down debt and invest in AI computing. Former “mining industry leader” Bitfarms announced a complete exit from Bitcoin mining to focus entirely on AI.

3. The Other Side of the Madness: AI’s “Power Hunger”

While miners are fleeing, others are pouring money into AI companies.

● Morgan Stanley recently calculated that from 2025 to 2028, the US data center power demand will increase by 74 gigawatts (1 GW = 1 billion watts). But what about supply? New infrastructure being built adds only 10 GW, and the grid can provide about 15 GW of available capacity, totaling 25 GW—leaving a gap of 49 GW.

● This explains why mining farms are suddenly in high demand. What do miners have? Power lines, land, grid connection permits—what AI companies need most isn’t chips but “electricity,” specifically the speed of getting power online. How quickly you can connect power and set up data centers determines how fast you can seize market share.

● Morgan Stanley estimates that even if all available power in US and European Bitcoin mining farms were diverted to data centers, there would still be a power shortfall. However, transforming existing mining farms could reduce the gap by 10–15 GW, making it the best “quick fix.”

● Consequently, the story unfolds: mining farms are essentially large-scale, low-latency data centers with power capacity, cooling systems, and rack space. In Bitcoin’s bear market, these are cost burdens, but in the AI power shortage era, they are scarce assets capable of generating rental income.

4. Transition in Progress: Who Moves Fast, Who Gains Valuation

● On February 26, one of the largest US miners, MARA Holdings, announced a partnership with Starwood Capital to convert some of its Bitcoin mining facilities into AI data centers. The initial plan is for 1 GW capacity, expandable to 2.5 GW. Starwood handles design, leasing, construction, and operation, while MARA provides the site. The joint project caused MARA’s stock to jump 17% after hours.

● Interestingly, Morgan Stanley’s target price for MARA is only $8, about 3% below the then-current closing price. The reason: MARA isn’t “fully committed”—it still wants to mine coins and develop AI. The market prefers a “decisive” transformation story.

● What does decisive mean? TeraWulf is an example. The company recently secured financing for a 168 MW AI data center joint project with FluidStack, with Google as a guarantor. Analysts set a target price of $37, implying about 159% upside. All 13 analysts rated it a buy.

● Bitfarms has a more concrete plan: fully convert its 18 MW Bitcoin farm in Washington State to support Nvidia GB300 GPUs with liquid cooling, expected to be completed by late 2026. They estimate that this site, less than 1% of their total capacity, could generate net operating income surpassing all their previous mining profits once converted to GPU-as-a-Service.

● Valuation logic has completely shifted. Previously, miner stock prices fluctuated with coin prices and hash rates, like roller coasters. But with long-term leases and backing from reputable payers (like Google), cash flows become “monthly rent,” and the market values you as an infrastructure company—comparable to REITs like Equinix or Digital Realty, not other miners. Morgan Stanley calls this the “REIT Endgame.”

5. Hashrate Map: Who Takes Over, Who Gets Embarrassed

● Hashrate Index data shows that US miners currently hold about 37.5% of the global market share, Russia 16.4%, China 11.7%. If US miners shrink their Bitcoin operations to focus on AI, the network’s hashrate will become more concentrated in Russia and China. For a government that once promised to make the US the global crypto capital, this is somewhat awkward.

● But politics are unpredictable. Another possibility is selling to allies. For example, Canaan Technology recently spent nearly $40 million to buy a 49% stake in Cipher Mining’s three Texas farms. These farms have a total capacity of 120 MW, with electricity costs below $0.03 per kWh and wind power included. Capital is flowing, and the hash rate landscape is reshaping.

This round of “Mining to AI” is essentially a swap between two mathematical problems: on the Bitcoin side, halving reduces block rewards to 225 per day, with transaction fees unable to compensate; on the AI side, an additional 74 GW demand faces a 49 GW shortfall. The power, land, and grid permits held by miners are shifting from “cost of mining” to “hard currency of computing power.”

In the short term, shutdowns and conversions will continue. In the long term, those who can turn “mining volatility” into “rental cash flow” will survive as the next-generation infrastructure companies. This industry has never believed in sentiment—only in the price at which miners turn off.