Just noticed something pretty wild in the crypto mining space that's been bothering me. BitDeer just cleared out its entire Bitcoin holdings—and I mean literally zero BTC on the balance sheet now. They sold 189.8 BTC that week and dumped the remaining 943.1 BTC inventory in one shot. For a mining company, that's not just a move; it's a statement.

Here's what's actually happening: Wu Jihan is betting everything on a completely different game. The old mining playbook was straightforward—use today's electricity to arbitrage tomorrow's Bitcoin price. Simple time arbitrage. But Wu Jihan is flipping the script. He's shifting the bet from crypto prices to AI computing power demand. The means changed too: instead of just running miners, he's borrowing massive amounts to acquire land globally for data centers.

The numbers tell the story. As of early 2026, BitDeer has about 3 GW of power pipeline capacity globally—that's equivalent to powering 10 to 30 hyperscale data centers like Google's. They're already operating 1.6 GW, with another 1.3 GW under construction. But here's the catch: to make this happen, Wu Jihan loaded up on debt. We're talking $1.3 billion total—$1 billion in borrowings plus $325 million in new convertible bonds just priced in February. That's a massive bet.

The debt structure is actually clever, not reckless. Three tranches of convertible bonds maturing in 2029, 2031, and 2032. The idea is that by the time each batch matures, BitDeer's AI revenue should be generating enough cash flow that bondholders just convert to equity rather than demanding repayment. The math looks decent on paper—Roth/MKM estimates potential annualized revenue of $850 million if they fully deploy HPC capacity. Wu Jihan's team is even more bullish, claiming over $2 billion annually once everything's operational.

But here's where it gets risky. The AI business is currently doing maybe $10 million annually—less than 2% of total revenue. GPU count tripled from 584 to 1,792 in three months, but utilization dropped to 41% because machines are being deployed faster than revenue is coming in. The B200/GB200 series is still in testing. So they're burning cash on deployment while waiting for revenue to catch up.

There's also the Clarington problem in Ohio. That 570 MW facility represents 42% of their under-construction pipeline. A neighboring steel manufacturer sued, claiming the data center would interfere with shared infrastructure. If they lose or get delayed significantly, the entire timeline collapses. The first convertible bond matures in 2029—they need revenue flowing by then to make this work.

Meanwhile, mining itself is getting squeezed. Bitcoin network difficulty jumped 14.7% in February, the biggest jump since May 2021. With the same power costs, miners are producing fewer coins. Q4 gross margins already dropped from 7.4% to 4.7% year-over-year. The old business is thinning out.

What Wu Jihan essentially bought with $1.3 billion is positioning where 'whoever wins in AI still has to pay my electricity bill.' Whether it's hyperscalers or AI companies, they need power. He's not betting on which company wins—he's controlling the infrastructure entrance, like how Amazon didn't bet on which internet company would dominate, they just rented servers to everyone.

The real test comes 2027-2029. Tydal in Norway should be converted to an AI data center by end of 2026. Clarington needs to clear the lawsuit and start construction in 2027. Both need to be generating serious revenue by 2029 when bonds mature. The market's already skeptical—stock price is around $8, way below analyst targets. But Wu Jihan got what he needed most: time. Whether AI funding catches up with his debt schedule over the next three years will determine if this transformation is genius or catastrophic.