Bitcoin mining company escapes again for the Nth time

Author: Zhou, ChainCatcher

Since late last year, publicly listed mining companies have kicked off a new wave of collective sell-offs.

Cango sold about 60% of its holdings—4,451 bitcoins—in February; Bitdeer liquidated all its bitcoin inventory in January; Riot Platforms sold 3,778 BTC in the first quarter; and Core Scientific had also previously planned to sell about 2,500 bitcoins in the first quarter.

Recently, MARA, a leading miner, disclosed in an announcement that in just three weeks—from March 4 to March 25—the company sold 15,133 bitcoins and raised more than $1 billion in cash. At the same time, it announced that it would cut about 15% of its employees as part of a strategic shift toward energy and digital infrastructure companies.

In fact, miners selling bitcoin is nothing new. During the bear markets of 2018 and 2022, mining firms also experienced large-scale liquidation and capitulation afterward, leaving behind more efficient players. But this time, what triggered the sell-off was not only a decline in coin prices—they also had a new destination to redeploy capital to: AI data centers.

1. Three layers of motives behind the sell-off

On the surface, it looks like a collective sell-off by mining companies. But if you break it down, their underlying motives are not uniform. Broadly, they can be categorized into three types of selling logic with different natures.

Mining itself has fallen into losses

The first—and most direct—one: cost pressure.

CoinShares’ latest mining report shows that the weighted average cash cost to mine one BTC among current publicly listed mining companies is about $79,995, while the BTC market price has been trading between $68k and $70k. On average, each coin is down by nearly $19,000, and the overall situation is about a 21% loss.

This is no longer just a matter of shrinking profit margins, but whether cash flow can actually hold up if they keep mining.

The report also shows that early in March, the hash rate price temporarily fell to $28 to $30 per PH per day, hitting the lowest level in history since the halving. At this level, for most active mining rigs to maintain cash profitability, electricity prices basically need to be pushed below $0.05 per kWh. Currently, about 15% to 20% of mining machines across the network are right on the edge of breakeven.

Meanwhile, geopolitical tensions in the Middle East are driving up energy prices, and power costs continue to face sustained pressure—an external variable that miners themselves can hardly control.

In its report, QCP Group pointed out that when the bitcoin price is significantly below the average mining cost, mining companies face considerable pressure, and liquidity priority is higher than a strategy of holding coins.

Against this backdrop, for some miners, selling bitcoin has become a realistic requirement to keep operations running.

AI provides a more stable revenue model

The second motive is more strategic, and also the most worth digging into in this round of sell-off.

Bloomberg’s analysis notes that unlike past sell-offs done to cover costs, the funds from this sell-off are being reallocated to the field of artificial intelligence.

The commercial logic behind it is straightforward: mining revenue is highly dependent on the coin price, mining difficulty, and electricity prices—so volatility is extremely high. By comparison, AI infrastructure is closer to long-term leases. CoinShares’ report indicates that its profit margins can reach 80% to 90%, and revenue is more predictably long-term.

More importantly, miners already hold ready-made resources: cheap power contracts, built data centers, mature cooling systems, and experienced operations teams.

Some analysts point out that the build cost of bitcoin mining infrastructure is roughly $700k to $1M per MW, while AI infrastructure is as high as $8 million to $15 million per MW. This huge cost gap is being monetized at scale by miners.

It’s also worth noting that behind this transition are a set of unexpected players: tech giants and traditional financial institutions.

Previously, Google provided credit backing for its AI cloud platform Fluidstack through lease obligations. The disclosed credit support it has provided has already exceeded $5 billion. It has successively underwritten AI transitions for miners such as TeraWulf, Cipher Mining, and Hut 8, and in exchange received corresponding equity; Microsoft signed a five-year, $9.7 billion AI cloud services contract with miner IREN; and Morgan Stanley provided a $500 million loan to Core Scientific, with a potential total limit of up to $1 billion.

Their entry provides far more solid capital backing for miners’ transition than anyone might have imagined.

Meanwhile, Core Scientific, TeraWulf, Hut 8, Cipher, and other miners have successively signed large AI/HPC contracts, with cumulative amounts already exceeding $70 billion. CoinShares’ report mentions that miners with AI/HPC contracts have valuation multiples about twice those of pure-play mining companies. The market is rewarding companies that complete the transition early with a valuation premium.

Even HIVE, a miner with the steadiest finances and the lowest leverage, has already proactively縮-backed its mining business and shifted to AI data center expansion. This shows that the pressure to transition is no longer a situation exclusive to heavily indebted miners—it is a directional choice facing the entire industry.

Proactively using BTC as a financial tool

The third logic is relatively shrewder and is the most主动.

Some miners choose to sell BTC not out of operational pressure, but as a tool to optimize their balance sheets—for example, MARA. The specific approach is: use proceeds from the sale to buy back previously issued convertible bonds at a discount below their face value. This both reduces the size of liabilities and lowers the risk of potential equity dilution.

For this kind of miner, BTC’s role on the balance sheet has quietly shifted—from a long-term holding symbolizing conviction to a strategic asset that can be flexibly allocated.

In addition, during this round of sell-offs, there has also emerged a relatively rare type of seller: sovereign states.

On-chain data shows that the BTC holdings of the Bhutan royal government have declined by about 66% from the peak at the end of 2024. The scale of a single transfer in March has risen to between $35 million and $45 million, and the pace of selling has continued to accelerate.

Unlike most countries that accumulate BTC through market purchases, Bhutan’s holdings come from its domestic hydropower mining business. This large-scale reduction may be related to the funding needs of its national-level development projects. It is also one of the largest government bitcoin sell-off actions on record.

With three layers of logic stacked together—mining losses, AI transitions, and debt optimization—plus sell pressure from the sovereign level, the market is absorbing structural supply pressure coming from multiple directions with different characteristics. Miners’ belief in bitcoin is being reshaped by more realistic business logic.

2. After exiting, each goes their own way

Of course, a sell-off does not equal full liquidation. The remaining holdings and subsequent strategies of each miner are now showing clear divergence.

Three paths, three choices

The first path: stay with mining.

Represented by CleanSpark and HIVE. They don’t follow the AI transition narrative, don’t add debt, and rely on a combination of low electricity prices, next-generation mining machines, and low leverage to try to win during the industry’s shakeout. The logic is that once high-cost production capacity exits one by one, the per-unit hash rate revenue of the remaining miners will rise accordingly.

CleanSpark has previously stated publicly that at the current hash rate price level, continuing to make large-scale investments in bitcoin mining is “not very economically reasonable,” but the company still chose to stick with its core business—betting that the cycle will eventually reverse.

A well-known crypto KOL, Blue Fox, pointed out that historically, after nearly every halving, miners have capitulated, and those who remain tend to be the more efficient players who then capture a larger share during the next rebound.

For this type of miner, sticking with mining is not stubbornness—it’s trust in the规律 of the cycle.

The second path: walk on two legs.

Represented by MARA, IREN, and Riot. Keep a fairly large BTC holdings base, and simultaneously build out AI/HPC. Use the relatively stable revenue from AI operations to hedge the cyclical volatility of mining revenue.

In essence, these companies are doing an asset allocation exercise; the answer differs by company. But the core logic is that the two lines of business support each other, dispersing the risk of a single area.

The third path: fully shift to AI.

Represented by Core Scientific, TeraWulf, and Cipher. BTC holdings have exited the position of core assets, and mining is gradually becoming an ancillary part of the data center business.

CoinShares expects that by the end of 2026, the share of AI revenue for some miners could be as high as 70%, while the share of mining revenue may drop from roughly 85% at the beginning of 2025 to less than 20%. These companies still look like miners on paper, but in reality they are becoming AI infrastructure operators with mining as the starting point.

The potential risk of this path is that a heavy-asset transition brings a massive debt burden; once AI demand cools, both ends of the business will come under pressure.

There are also views noting that the credit-guarantee structure carried out by Google via Fluidstack actually forms a highly concentrated counterparty risk: the entire cash flow chain depends on Fluidstack as an intermediary. If there is a major change in the AI lease market, this structure will become a single point of failure.

BTC price determines their fate

No matter which path they choose, they ultimately point to the same variable: where the price of BTC goes.

CoinShares lays out three scenarios:

● If BTC rebounds to $100k by the end of 2026, hash rate prices would rise to about $37 per PH per day, mining profits would recover, and overall industry pressure would ease;

● If it continues to stay below $80k, high-cost miners would accelerate their liquidations, and the traditional model of “mining and holding coins for a bull market” would become increasingly hard to sustain;

● If it breaks above the all-time high, hash rate prices could surge to $59 per PH per day, and the industry would enter a new round of expansion cycle.

Conclusion

Overall, miners are facing only two endgames: either the coin price rebounds, returning to their core business—everything happening now is merely a cyclical historical footnote; or prices remain persistently weak, and more and more miners complete their identity shift into AI data centers, making the company model of “mining and holding for a bull market” increasingly rare in this industry.

However, there is another issue worth probing beyond the business logic behind this transition. Miners are not ordinary public companies—the continued investment in hash rate is itself part of the bitcoin network’s security budget.

Sazmining CEO Kent Halliburton once said plainly that these companies “have power contracts, land, and infrastructure, yet they hand those resources over to Microsoft and Google, in exchange for rent check payments—transforming from protecting the Bitcoin network into storing rack space for mega-scale cloud service providers.”

When mining no longer generates sufficient economic returns, a rational business decision naturally is to shift resources. But if this trend continues to spread, it will become a question that cannot be ignored: who will bear the long-term cost of maintaining the security of the Bitcoin network?

History may have already provided an answer.

The Bitcoin network has gone through several large-scale miner liquidations, and each time after that, it operates with higher efficiency.

But this time, the miners that leave are not just shutting off machines.

The times have changed.