Nebius Group's $4.3B Convertible Bond: A Data-Driven Autopsy of the AI Infrastructure Bet
Ivytoshi
The $4.3 billion convertible bond raise by Nebius Group is not a signal of AI dominance. It is a desperate bet on GPU supply. The numbers don't lie. Follow the gas, not the hype.
Context: Nebius Group, the former Yandex AI infrastructure arm, announced a $4.3B convertible bond issuance to fund data center expansion. The market cheered. But I audited the data. The structure reeks of imbalance. Convertible bonds dilute equity at scale. The average conversion premium in tech debt is 20-30%. If Nebius shares trade down, the debt becomes a millstone. Based on my experience standardizing ICO ledgers, I learned that massive capital raises in hype sectors often precede market corrections. This is no different.
Core: The capital allocation is straightforward. At $30,000 per H100 GPU, $4.3B buys roughly 143,000 units. But network infrastructure, cooling, and construction eat 30% of that. Real GPU count: ~100,000. That is a large cluster, but not a game-changer. The real risk is timing. NVIDIA Blackwell B200 is already shipping. H100 assets will depreciate 40% by Q3 2025. Nebius is buying last-generation iron at peak prices. Quantify the manipulation: the bond terms are undisclosed. If the coupon is above 5%, the interest alone is $215M per year. Where is the revenue to cover that? No customer contracts were announced. No revenue disclosed. The data doesn't lie.
Contrarian: The conventional narrative says this is a vote of confidence in AI. I say it is a hedge against NVIDIA's monopoly. The bond investors are betting on GPU scarcity, not Nebius. If NVIDIA opens supply, GPU prices drop, and Nebius' asset base deflates. The contrarian angle: the real value is not in the GPUs, but in the energy contracts. Data center power purchase agreements (PPAs) are the new gold. Nebius may have locked in low-cost renewable energy, which drives long-term margin. But the article omitted that. I suspect the bond is a speculative play on energy arbitrage, not compute.
Takeaway: The next signal to watch is the first customer announcement. If it's a major AI lab, the thesis holds. If not, this is a distressed debt play. Data doesn't lie. But the market often does.