Recently, there was a decision that quite surprised the mining industry. Bitdeer, one of the largest Bitcoin mining companies in the world, decided to sell all their Bitcoin holdings. Last week, they posted an update: 189.8 BTC mined and sold, and the remaining 943.1 BTC also sold out. Their Bitcoin balance is now zero.

Why would a mining company do this? This is not a reckless decision. From the beginning, Bitcoin mining was essentially a simple arbitrage game: take electricity and machines today, exchange for Bitcoin tomorrow. No need for factories, no need for customers, no need for a fancy brand. The invested capital is the current cost, and the future price is at stake. This logic has been running for twelve years.

But now Wu Jihan, founder of Bitdeer, is changing his game target. It’s no longer about coin prices, but about long-term prices for computational power demand in the AI era. The method is also changing: from using electricity to earn coins, to borrowing money to buy land. The arbitrage object is changing, but the arbitrage structure remains the same.

In the same week as the Bitcoin sale, Bitdeer also completed a new debt issuance of $325 million. According to their financial reports, by the end of 2025, Bitdeer’s debt will reach $1 billion. Adding the new debt, total liabilities are now about $1.3 billion. The money is real, the land purchases are real, but will this aggressive strategy succeed? Maybe we’ll find out only in 2029.

Bitdeer was originally founded in 2018 as a mining machine sharing platform. Now they are one of the largest publicly listed mining companies, with a mining hash rate of 63.2 EH/s, the highest among all public mining firms. That’s about 6 percent of the total Bitcoin network hash rate. But Wu Jihan no longer wants to sell mining capacity. He wants to focus on the electricity business.

Bitdeer’s global power capacity now reaches 3,002 megawatts, with 1,658 MW operational and 1,344 MW still under construction or awaiting approval. For comparison, giant data centers owned by Microsoft and Google are usually only 100 to 300 MW. So, 3,002 MW is equivalent to gathering the electricity needs of 10 to 30 giant Google data centers in one company. That number is impressive in theory.

This $1.3 billion debt is mostly used to secure global land assets and prepare for the transition into AI data centers. There are several major projects. First, Rockdale in Texas with 563 MW, already operational and focused on mining with stable cash flow. Second, Clarington in Ohio with 570 MW, which is the core of their entire AI transition plan. Power contracts have been signed, scheduled to complete in Q2 2027. But this is also the biggest current risk, as we will discuss later.

Then there’s Tydal in Norway, 175 MW, converting a mine into an AI data center, scheduled to finish by late 2026. Because it uses hydroelectric power, energy costs are very competitive. This is the most advanced and lowest-risk option at the moment.

In the AI industry, there are terms for the three hardest things to imitate: land, electricity, and server space. Bitdeer has accumulated all three through ten years of mining operations. But there’s one rarely discussed but noteworthy point: they are also developing their own mining chip called SEALMINER.

The SEAL series has reached the third generation. SEAL03 has an energy efficiency of 9.7 joules per terahash. To understand what a joule is, simply put, it’s a unit of energy measurement. In mining, the lower the joules per terahash, the more energy-efficient the machine is at performing calculations. The A3 Pro, which will be mass-produced in September 2025, is already among the top globally. SEAL04 aims for 5 joules per terahash, and if achieved, will surpass all mass-market mining machines available. The gross margin of their self-developed chips exceeds 40 percent, much higher than mining itself. This is a repeat of what Wu Jihan did at Bitmain: from buying other people’s shovels to making their own.

Now let’s see how much profit AI could bring. In less than two years, Bitdeer has gone through several funding rounds. In May 2024, Tether invested $100 million and became the second-largest shareholder. Three months later, the first convertible loan of $150 million with an 8.5 percent annual interest rate was realized. In November of that year, the second convertible loan of $360 million with a reduced interest rate of 5.25 percent. By November 2025, a combined offering of $400 million in convertible bonds and 148.4 million additional shares. In February 2026, again with $325 million in convertible bonds and 43.5 million additional shares, plus repurchasing $135 million of old bonds maturing in 2029, extending to 2032.

Each time bonds are issued, Bitdeer’s stock price drops 10 to 17 percent. This has become a market reflex. The company’s profits always manage to secure funding. The core of the debt structure is convertible bonds. The 2032 bonds have an initial conversion price of about $9.93, a 25 percent premium over the offering price of $7.94. When the stock price reaches that level, bondholders will convert into shares instead of cash. The company doesn’t need to repay the debt as long as the stock price rises. It’s essentially a bet on whether the AI narrative will be accepted by the market.

The annual interest expense, assuming an average interest rate of 5 percent and principal of $1.3 billion, results in over $650 million in interest payments per year. Meanwhile, AI or HPC cloud revenue throughout 2025 will not reach even one-tenth of the six-month interest. Currently, this interest depends entirely on continuous debt issuance. The pressure is indeed significant.

Bitdeer’s AI business now generates $10 million annually, less than 2 percent of total revenue. For a company with a market cap approaching $2 billion, this figure is almost negligible. Their GPU count increased from 584 to 1,792 in three months, tripling. But utilization rates dropped from 87 percent to 41 percent, mainly because machines were installed too quickly while the B200 or GB200 are still in testing and haven’t started generating revenue.

Electricity is in place, machines are being installed, but revenue has yet to follow. How high is the ceiling? Roth or MKM estimate that if HPC capacity is fully realized, annual revenue could reach about $850 million. Management is more aggressive: allocating all 200 MW to AI cloud, annual revenue could exceed $2 billion, three times the mining revenue in 2025. But both figures depend on three conditions: project completion on time, securing long-term contracts from hyperscalers, and full GPU operation. None of these are currently met.

This is the war Bitdeer is playing: mining to support AI, while AI is shaping a dream. Whether that dream can be realized still depends on execution in the next two to three years.

Bitdeer’s debt structure is designed to be more stable than it appears. High-leverage companies usually fail for the same reason: maturing debt concentrates, insufficient cash, forced asset sales. Bitdeer has set maturity dates for three series of convertible bonds in 2029, 2031, and 2032. This can be seen as an intentionally created buffer. When the first batch matures, Tydal and Clarington are theoretically already in operation. When the second batch matures, AI revenue should be able to speak for itself. When the third batch matures, the market will judge what this company really is.

But Wall Street doesn’t buy this story. Keefe Bruyette lowered their target price from $26.50 to $14. The current stock price is around $8. The market signals are very clear: the transformation story must show real revenue.

But all this pressure gives Wu Jihan what he needs most: time. The smooth scenario might look like this. By the end of 2026, Tydal renovations are complete, the 164 MW hydroelectric data center in Norway begins operation, and European customer contracts start coming in. In 2027, Clarington wins in court, the 570 MW project in Ohio officially begins construction, major American clients follow. From 2028 to 2029, two core assets operate fully, revenue reaches the billion-dollar level, and analysts change Bitdeer’s label from a mining discount to AI infrastructure premium. In 2029, the first debt matures, and debt holders may choose to exchange shares rather than receive cash.

But there’s something more urgent. In the same Ohio industrial park, there’s a steel producer called American Heavy Plate Solutions that signed a 30-year lease for 9.9 acres in 2018. They sued Bitdeer: the construction of the AI data center will interfere with electricity, roads, railways, and communication lines, violating restrictions. Their demand is for the court to issue a permanent injunction preventing Bitdeer from starting construction. Clarington accounts for 42 percent of the pipeline under construction. If delayed, the entire schedule must be rewritten. So, Bitdeer’s biggest single risk isn’t debt or stock price, but a steel plant.

In the mining side, there’s no time to rest either. February 2026 saw Bitcoin network difficulty jump 14.7 percent, the largest single increase since May 2021. With the same electricity, fewer coins are mined. Q4 gross margin dropped from 7.4 percent a year ago to 4.7 percent. This mining branch is gradually becoming thinner.

The worst-case scenario is also clear: a prolonged two-year lawsuit with Clarington, halted construction; Tydal delayed, GPU utilization remains around 41 percent; the first debt matures in 2029, cash on the balance sheet is insufficient, refinancing is forced, shares continue to dilute, and the conversion threshold becomes even harder to reach.

Both paths are very real. What Wu Jihan is truly betting on with billions is that whoever wins, they will pay for his electricity. No need to guess the route, just control the entry point. Amazon doesn’t guess which internet company will win, they just rent servers to everyone. AT&T doesn’t care what you say on the phone, only whether you call or not. From selling products, to selling services, to collecting rent, the industry’s evolution always follows one direction. The only difference is whether you actively move closer or are pushed by others. Wu Jihan is waiting for AI money to accelerate his debt repayment.