I just noticed something interesting about Bitdeer's movements. Last week, they liquidated all their Bitcoin holdings—selling off 943 BTC completely, leaving their balance at zero. At first glance, it seems odd for a mining company, but behind this move is a much deeper transformation strategy. They are all-in on AI infrastructure, and their debt has already reached $1.3 billion.

Here's the story. Bitdeer started in 2018 as a mining machine sharing platform. Now, it’s one of the largest publicly listed mining companies, with 63.2 EH/s hashpower— the highest among public mining firms. But Wu Jihan seems to be tired of selling computing power. He wants to focus on more fundamental assets: land, electricity, and server space. These three things are called 'the hardest assets to imitate' in the AI industry. Bitdeer has been accumulating them over ten years of mining operations.

For this transition, they just completed a $325 million debt issuance in February. Combined with previous debt, their total debt book exceeds $1.3 billion. Tether has been the second-largest shareholder since May 2024 with a $100 million investment. Their convertible bonds are spread across 2029, 2031, and 2032—an intentionally structured buffer period. The annual interest they pay exceeds $650 million, even though their AI/HPC cloud revenue is still far below that. So, they are basically spinning debt to keep operations running.

Their main assets are located in three sites. First, Rockdale, Texas—563 MW already operational, focused on mining with stable cash flow. Second, Clarington, Ohio—570 MW, positioned as an HPC/AI center, scheduled for Q2 2027. Third, Tydal, Norway—175 MW with hydroelectric power, converting from a mine to an AI data center, scheduled for the end of 2026. Tydal is the lowest risk because the transformation cost is much cheaper than building from scratch.

Regarding energy efficiency, Bitdeer is also developing its own mining chips—series SEAL. SEAL03 has an efficiency of 9.7 joules per terahash, ranking among the top globally. SEAL04 aims for 5 joules per terahash—if achieved, it will surpass all existing mining machines on the market. The gross margin on chips alone is over 40%, much higher than traditional mining. This is a repeat of what they did at Bitmain: moving from buying tools from others to making their own tools.

Their GPU count has tripled in three months—from 584 to 1,792. But utilization rate has dropped from 87% to 41%. Why? Because they installed machines too quickly, while B200/GB200 are still in customer testing phases and haven't generated revenue yet. Power is installed, machines are being set up, but revenue hasn't caught up. Analysts estimate that if HPC capacity is fully utilized, it could generate $850 million annually. Management is more aggressive: if 200 MW are allocated to cloud AI, revenue could exceed $2 billion—three times the total mining revenue in 2025. But all of this depends on three conditions: on-time completion, long-term contracts with hyperscalers, and GPUs running at full capacity.

The biggest problem isn’t debt, but a steel factory. In Clarington, there’s American Heavy Plate Solutions, which signed a 30-year lease in 2018. They are now suing Bitdeer, claiming that the AI data center construction will disrupt shared electricity, roads, railways, and communication lines. They are asking the court to issue a permanent injunction. Clarington accounts for 42% of the pipeline under construction. If delayed, the entire timeline must be rewritten.

So, this is basically a high-wire act. Wu Jihan is buying this window with billions of dollars. He’s waiting for AI revenue to chase the debt. If everything goes smoothly: Tydal is operational by late 2026, Clarington wins in court and construction begins in 2027, and by 2028-2029, both core assets are fully operational with revenues approaching billions. When the first bonds mature in 2029, holders will see the stock price and likely choose to convert to equity rather than cash. But if the Clarington lawsuit drags on for two years, Tydal is delayed, GPU utilization remains at 41%, and the 2029 debt matures with insufficient cash—then forced refinancing, ongoing dilution, and higher conversion thresholds will make it even harder to reach their targets.

What’s interesting about all this: the mining industry always bets on the same thing—something in the future will be more expensive than current costs. Ten years ago, they bet on coin prices rising. Now, they bet on an explosion in demand for compute power. The object changes, but the logic of time arbitrage remains the same. Wu Jihan is basically buying the position of ‘whoever wins, has to pay my electricity.’ He’s not guessing which AI company will succeed; he’s just controlling the gateway. Amazon didn’t guess which internet company would win; they just rented servers to everyone. AT&T didn’t care what you said on the phone; as long as you made a call or not. The industry’s evolution is always one way: from selling products, to selling services, to collecting rent. The only difference is whether you walk there actively or are pushed by others. Wu Jihan bought this window with debt. Now he’s waiting for AI money to chase the speed of debt.