Author: Heart of Computing Power
Recently, crypto mining companies in the U.S. stock market have been having a tough time.
At the close of U.S. stocks on October 14, 2026, the share price of former "mining leader" MARA Holdings fell to $8.23, a single-day drop of 9.66%.
If you include the past four weeks, it has already fallen 26.78%.
If you extend the time frame to a full year, the decline is even greater, reaching 63.97%.
Just the day before, on October 13, another mining company, Cipher Digital, announced an $850 million zero-coupon convertible bond financing. Not only did it fail to boost confidence, but the stock price plunged as much as 15% intraday that day.
And a few days earlier, on October 9, investment bank Morgan Stanley downgraded the target prices of four mining companies—MARA, Riot, Galaxy, and Cipher—all at once, giving a very direct reason: the borrowing environment for mining companies to do AI business had tightened.
Previously, everyone thought it was only natural for mining companies with large amounts of power to transition into AI computing power.
But by this autumn, the situation was different.
I. Borrowing to Transform, and the Stock Price Fell First
The market selloff triggered by Cipher's bond issuance this time actually had a very clear chain of events.
On October 13, the plan Cipher put forward was to issue $850 million in zero-coupon convertible bonds due in November 2032, with an additional $150 million over-allotment option, bringing the total scale to as much as $1 billion; pricing was completed on October 14, with the conversion price set at $15.31, 35% higher than the closing price of $11.34 the day before pricing.
On the surface, the company would not have to pay regular interest and could obtain a large amount of cash all at once, making the terms look very favorable.
But the problem is, why was it suddenly in such a hurry for so much money?
Looking at the investor presentation materials they submitted to the SEC on October 13, it becomes clear that Cipher's core AI data center project, Barber Lake, went over budget.
The project's initial budget was $2.157 billion, and it has now been directly adjusted to $2.941 billion to $2.966 billion, an increase in cost of more than 36%.
The company's explanation was that tenant Fluidstack raised new design adjustment requirements.
This put reality on the table: the major client changed the plan, engineering costs rose sharply, and Cipher had to front the first $359.3 million of the overrun itself, with the tenant only reimbursing half of the excess over the 20-year lease term. The company itself could not come up with so much cash to front, so it had no choice but to borrow from the market.
But convertible bond investors look at stock price performance. Seeing project costs spiral out of control and the company still needing to borrow huge sums to cover the shortfall, the market's first reaction was to vote with its feet, and the stock fell 15% intraday on the day the news was announced.
II. Short by Tens of Billions, Delivery Cannot Keep Up
The difficulties Cipher encountered are a microcosm of the entire group of mining companies undergoing transformation.
In a June report, investment institution VanEck did the math: for the entire mining industry to complete all the AI infrastructure previously announced, the short-term funding gap is about $50 billion, and long-term capital demand is close to $2.21 trillion.
Based on VanEck's data, consulting firm BlocksBridge Consulting also provided specific figures for several leading companies: IREN's funding gap reached $21.1 billion, Riot was short $7.2 billion, and HIVE was also short $4.6 billion.
While money is lacking, the cost of borrowing outside is still getting more expensive.
Take IREN as an example. In June, the company could still obtain $3.65 billion in GPU financing at a blended interest rate of 6%; just two months later, for a separate $2.4 billion transaction, the fixed rate directly rose to 9%.
The reason lending institutions have become more cautious is that everyone's focus of examination has changed.
Last year, everyone cared about how much power quota mining companies held and how many letters of intent for cooperation they had signed, but this year everyone only looks at actual project progress.
According to VanEck's mid-year statistics, the industry's actual delivered computing capacity is currently only about 25% of the capacity already signed and leased.
The blueprint is drawn very large, but not many computer rooms have actually been delivered on the construction site.
III. The Old Business Is Losing Money, and Even the Mining Pool Has Shut Down
Since transforming into AI is so expensive, can't they just honestly mine Bitcoin?
The answer is that in the first half of this year, the old business really could not make money.
On October 13, 2026, Binance, the world's largest crypto exchange, announced that it would officially shut down its mining pool business starting November 10, with all supporting services including Pool Savings, cloud mining, and transaction accelerators discontinued, and after that miners would no longer receive rewards for submitting hash rate.
Binance Pool launched in April 2020 and operated smoothly for more than six years.
The official reason given for the shutdown was a routine product evaluation, but the broader context is not hard to understand: the economics of mining have long been difficult to balance.
CoinShares data shows that the weighted average cash cost for listed mining companies to mine one Bitcoin was about $80,000 in the fourth quarter of 2025, and fell to about $75,500 in the second quarter of 2026.
When Bitcoin fell to $69,000 in March this year, based on the production cost of about $88,000 at the time, miners lost about $19,000 for every coin mined.
It was not until Bitcoin recently returned above $80,000 and the hash price recovered to around $40 that the industry barely returned to near the break-even line.
It is precisely because the main business fell into sustained losses in the first half of the year that mining companies had to turn to AI to seek a second growth curve.
But the awkward reality is that the old business not only cannot supply ammunition to the new business, but has instead become a drag.
IV. Some Are Holding On Desperately, Some Cut Losses First
Facing industry-wide funding pressure, mining companies have taken two completely different paths in their response strategies.
The first path is to hold on desperately, such as the giant MARA.
MARA was unwilling to give up its Bitcoin position and chose to pledge 54% of its Bitcoin holdings, adding $600 million in new borrowing, which directly pushed its current debt scale to 70.8% of the company's total market value.
The result of propping things up with high leverage is being held hostage by high interest expenses. In Morgan Stanley's October 9 research report, MARA's new target price was capped at $10, leaving only 2.4% potential upside, directly ranking last among the companies covered.
Bitdeer chose another path: not betting on a Bitcoin price rebound, but completely cutting losses to preserve liquidity.
Starting with Tether's investment in 2024, Bitdeer also replenished its ammunition substantially through convertible bonds. By the end of the first quarter of this year, total borrowing was about $1.9 billion, triggering concerns in the capital market about its debt scale.
But compared with other peers still hesitating, it made an extremely rare yet highly pragmatic decision—
in February this year, it directly liquidated all of its own Bitcoin (holdings dropped to zero on February 20), and afterward sold all newly mined coins immediately, completely converting them into cash.
During the industry's freezing point in the first half of this year, when mining was generally unprofitable, stubbornly clinging to the "faith in hoarding coins" often led to being dragged down by high interest expenses.
This step by Bitdeer was equivalent to directly cutting off the continuous bleeding of the old business and pinning all monetized funds firmly on AI infrastructure.
This decisive strategy has bought it a substantial lead in delivery progress.
In an industry delivery quagmire where the delivery rate is only about 25%, Bitdeer's 180 MW AI data center in Tydal, Norway, has been advancing on schedule, with the first phase targeted for delivery in December this year.
An even more critical set of data is that in April this year, the annual recurring revenue (ARR) of its AI cloud business had already reached about $69 million.
Although dragged down by the broader plunge in the mining sector, Bitdeer's stock price fell back to $8.40 on October 15, the two are fundamentally different—
its peers were dragged into a liquidity quagmire by old burdens, while Bitdeer completed a heavy-asset lane change in an extremely short time, becoming one of the few mining companies that has truly seen money coming back from its AI business.
▌▌▌▌▌▌▌▌▌▌
From being a hot favorite with power resources in hand to now facing stock price adjustments, the change in mining companies' situation is actually a shift in logic.
Previously, the market gave mining companies high valuations because it valued the scarce power quotas in the AI era and the grand contracts on paper.
But as projects successively broke ground, investors began to assess actual engineering progress, debt ratios, and interest expenses.
Facing a $50 billion short-term industry gap, along with the objective facts of losses in the old business and the heavy-asset nature of the new business, as the cost of capital gradually rises, the market is waiting for mining companies to prove themselves again with real delivery performance.