Nebius Group just raised $4.3B in convertible bonds to build AI data centers. The market cheers. I run the numbers. The math doesn't add up.
Let me start with what matters: the capital structure. Four point three billion dollars in convertible bonds, zero equity dilution at the moment. Sounds like a smart move for a company that needs to scale fast. But every bond has a conversion price, and every conversion price tells a story about the issuer's confidence. The fact that Nebius Group chose debt over equity at this stage signals either a desperate need for cash without giving up control, or a calculated bet that their stock will rise enough to make conversion cheap. Either way, the risk sits squarely on the holders.
Context: Who is Nebius Group?
Nebius Group is the AI infrastructure spin-off of Yandex, the Russian tech giant. They build data centers optimized for GPU-heavy workloads. The company is not a protocol, not a DeFi platform, not a layer-2. They are a pure-play compute provider, competing with CoreWeave, Lambda Labs, and the hyperscalers. The $4.3B will go entirely into data center construction, likely in Europe and North America. The Crypto Briefing article that broke the news used the word "massive" four times. That should have been your first red flag.
Core: The Order Flow Analysis
Let me break down the numbers. A single NVIDIA H100 GPU costs roughly $30,000 at current market prices. $4.3 billion divided by $30,000 equals 143,333 units. That's enough to build a top-10 supercomputer. But the GPU market is not a spot market. Delivery times stretch 12 to 18 months. By the time Nebius Group gets those GPUs, NVIDIA will have shipped the B200 Blackwell, which is 2x to 4x faster for training. The H100 will depreciate like a rental car driven off the lot. That's impermanent loss at a macro scale, and no amount of institutional marketing can cover it.
"Ledgers do not lie, only the auditors do." The ledgers here show a $4.3B liability with no guaranteed revenue stream. Convertible bonds typically carry interest rates of 2% to 4% and a conversion premium of 20% to 30%. If Nebius Group's stock doesn't perform, they'll have to repay the principal in cash. Where does that cash come from? Not from GPU sales, because they haven't locked in any major customers yet. The article didn't mention a single enterprise contract. That's a red flag the size of a data center.
Now look at the competitive landscape. AWS, Azure, and Google Cloud collectively spend $200B to $500B on AI infrastructure annually. Nebius Group's $4.3B is a rounding error. They can't compete on scale. They can't compete on brand trust. They can only compete on price, and price wars in the GPU rental market are already brutal. CoreWeave recently cut prices by 20% to attract customers. Nebius Group will have to do the same, which means their unit economics will be squeezed before they even start.
"Beta is the tax you pay for ignorance." Retail investors see AI as the next gold rush. Smart money sees a commodity play with razor-thin margins. The real opportunity is not in owning the pickaxes but in shorting the overhyped narratives. Nebius Group's convertible bond is a leveraged bet on infinite GPU demand. If demand slowsโdue to model efficiency gains, smaller models, or a shift to inference rather than trainingโthe bondholders will be left holding debt on a depreciating asset.
Contrarian: Retail vs. Smart Money
Retail traders are piling into AI infrastructure stocks, treating every data center announcement as a catalyst. They see Nebius Group's $4.3B as a sign of strength. I see it as a sign of desperation. The company could have done an equity raise at a higher valuation. They chose debt because equity would dilute existing shareholders. That means insiders think the stock is overvalued relative to the bond's conversion price. Classic insider signal.
"Yield without due diligence is just borrowed luck." The yield on these bonds will be attractive, maybe 3% to 5%. But the risk is not the yield; it's the principal. If Nebius Group fails to generate enough cash flow to service the debt, they'll either have to issue more debt or default. Given the GPU supply risks and the competitive pressure, default is a real possibility within five years.
Moreover, the geopolitical dimension is ignored. Nebius Group has Russian roots. Even though they spun off from Yandex, the sanctions and regulatory scrutiny remain. US and European cloud customers may be reluctant to commit to a provider with murky ownership. That's a liquidity risk that no bond prospectus can capture.
"Liquidity is the only truth in a fragmented chain." In the AI compute market, liquidity means customer contracts. Without them, the data center is just a pile of expensive hardware. Nebius Group has not announced any anchor tenants. Compare that to CoreWeave, which signed a $1.5B deal with Microsoft. The difference is night and day.
Takeaway: Actionable Price Levels
If you're trading this narrative, watch the bond's terms. If the conversion price is set at a 30% premium to the current stock price, that bull case is already priced in. Any negative news on GPU delivery or customer acquisition will send the stock down, making conversion unlikely and increasing the debt burden. I'd short the stock if it breaks below the offering price of the bonds. Alternatively, if you want exposure to AI infrastructure, buy NVIDIA directly. They are the only party in this deal that profits regardless of who wins the data center war.
The algorithm executes, but the human decides. My decision is to stay out of Nebius Group's capital structure. The ledger shows a $4.3B liability with no revenue to back it. That's not a yield play; it's a gamble. And in a bull market, gambles get funded. But when the tide turns, the first to sink are those who ignored the balance sheet.
Sigma: The only safe bet in this deal is the bond underwriters. They collect fees upfront. Everyone else is betting on a GPU delivery schedule and a customer pipeline that doesn't exist yet. Read the code. Audit the contracts. Then decide.