$3.5B Debt, 9.875% Interest: Galaxy Digital’s AI Data Center Bet Is a High-Wire Act

Galaxy Digital just dropped a $3.5 billion time bomb on the AI infrastructure sector. And the fuse is a 9.875% annual coupon. That’s $346 million in interest every year – before a single GPU is ever racked. The debt is secured against a 260-megawatt data center campus in Texas. But here’s the kicker: principal repayments don’t start until the building is operational. That’s a minimum three-year gap. No cash flow. Just mounting interest. I’ve been dissecting high-yield crypto debt structures since the FTX crash. This one smells different. Not fraud – but over-leverage disguised as inevitability. Let me break down why. ⚠️ Deep article forbidden Context: Why Now? The AI inference boom is real. Everyone from OpenAI to Meta is scrambling for compute. CoreWeave, a cloud provider specializing in GPU clusters, already holds contracts with major labs. Galaxy’s CEO Mike Novogratz is betting his firm’s transformation from crypto trader to infrastructure financier. The vehicle? Galaxy Helios Data Centers II LLC – a special-purpose entity that isolates this risk from Galaxy’s main balance sheet. On paper, it’s a textbook asset-backed deal. Reality is messier. The debt is rated below investment grade. 9.875% is screaming "high risk" – not growth story. Compare that to typical investment-grade tech bonds yielding 4-5%. The premium tells you the market is pricing in serious execution risk. And they’re right. Core: The Numbers Don’t Lie. Let’s do forensic accounting. The $3.5B note is split into two tranches: a 2031 maturity with initial 4% annual amortization, and a floating-rate component tied to SOFR. The interest is payable semi-annually starting 2026. But the project’s first phase won’t deliver critical IT load until early 2027. That means Galaxy must fund $346M per year from its own cash reserves – or from other crypto earnings – for at least 12 months before any revenue hits. Their last 10-Q showed $800M in cash and equivalents. One year of interest alone consumes 43% of that. Add construction costs overruns (standard in data centers: 20-30% overruns are common) and the liquidity buffer vanishes fast. I ran a stress test using 2022’s bear market conditions. If Bitcoin drops to $30k and stays there, Galaxy’s trading revenue evaporates. They’d be forced to sell crypto holdings to service this debt. That’s the contagion channel no one is talking about. The deal’s collateral includes a first-priority lien on the project’s assets – land, buildings, equipment. But land in rural Texas isn’t worth $3.5B. The real collateral is the expected future cash flows from AI tenant contracts. And those contracts? Unstated. No public disclosure of committed pre-leases. That’s a red flag. In comparable real estate debt issuances, you’d see a minimum 60% pre-leased threshold. Here, we have silence. The bondholders are betting on a speculative demand curve. ⚠️ Deep article forbidden Contrarian Angle: The Real Blind Spot. The mainstream narrative is bullish: “Crypto capital building AI infrastructure – convergence at last.” I disagree. This is a classic late-cycle capital deployment. High interest rates + massive leverage + untested asset class = recipe for wipeout. The contrarian take? The 9.875% rate itself is a canary. It reveals that traditional fixed-income investors view Galaxy and CoreWeave as riskier than the average tech borrower. Why? Because crypto exposure taints the credit rating. Even if the project succeeds operationally, any downturn in digital assets could trigger a covenant breach or force a distressed sale. The other blind spot: regulatory creep. Texas’s ERCOT grid is fragile. Adding 400 megawatts of compute load could push the system over during summer peaks. If state regulators impose curtailments, the project’s uptime guarantees suffer. Revenue drops. Interest keeps accruing. That’s a death spiral. I’ve seen this pattern before – in 2018’s mining bubble, where cheap debt funded hardware that became worthless when Bitcoin dropped. The difference now is the interest burden is 5x higher. Back then, 6% was considered aggressive. Today, we’re at 9.875%. And the underlying asset is not a store of value – it’s a service that requires continuous demand. If AI model training hits a plateau in 2027, those GPUs become stranded assets. CoreWeave’s clients could walk. No long-term contracts means no cash flow. And bondholders get left holding a half-built shell in Texas. Takeaway: The Next Watch. I’m tracking three signals. First: the completion of Phase 2 foundation work by Q4 2026. Any delays will trigger bond price drops. Second: any announcement of a major tenant lease (e.g., Microsoft or OpenAI). If it doesn’t come by early 2027, the bull case collapses. Third: Galaxy’s crypto holdings. If they start moving BTC to exchanges, it’s a liquidity warning. This isn’t a bet on AI – it’s a bet on timing. One missed milestone, and the whole house of cards folds. Keep your stop-loss tight. ⚠️ Deep article forbidden