A $775 million debt raise. A claim of $40 billion in customer backing. A company with a complicated past.
Nebius Group just made a splash in the AI cloud market. The numbers sound big. The narrative is clear: GPU capacity expansion, debt not equity, no dilution. But I've been through enough ICO audits and DeFi yield chases to know that big numbers often hide bigger vulnerabilities.
Let me be blunt. The $40 billion figure is the first red flag. Not because it's impossible, but because it's too convenient. In 2017, I reverse-engineered a token distribution contract and found an integer overflow that allowed early whales to extract 20% of supply. The team had promised a fair launch. The code told a different story.
Here, the story is a debt facility. The code is the balance sheet. And I'm going to dissect it the same way: line by line, vulnerability by vulnerability.
Context: Who is Nebius, Really?
Nebius is the AI cloud spin-off of Yandex N.V., the Russian tech giant that has been restructuring since the Ukraine invasion. The company relocated its headquarters to the Netherlands. It now pitches itself as a European AI infrastructure provider. Its main offering: GPU-as-a-service, competing with CoreWeave, Lambda Labs, and the hyperscalers.
This funding round is a senior secured debt facility. That means it's backed by assets—likely its existing GPU fleet, data center real estate, and possibly receivables. The company says the funds will be used to expand its AI cloud platform and GPU capacity. No equity dilution. Clean, efficient, capital-light on the surface.
But senior secured debt is not free money. It comes with covenants, interest payments, and liquidation priority. If Nebius defaults, the lenders take the GPUs. That's a very different risk profile from equity funding.
Core: Deconstructing the Numbers
Let's start with the debt itself. $775 million. Assume a blended interest rate of 10-12% (typical for secured debt in the current high-rate environment). That's $77.5 to $93 million in annual interest expense. For context, Nebius's parent Yandex reported approximately $4.6 billion in total revenue in 2023, but much of that came from search and advertising. The AI cloud segment is smaller. We don't have exact EBITDA figures, but the interest burden is material.
Now, the $40 billion customer backing claim. This is where my applied mathematics background kicks in. Let's stress-test this number.
Global GPU cloud market revenue in 2024 is estimated between $10-15 billion. Nebius is a relatively small player. If they have $40 billion in customer backing, that implies they've captured several years of industry-wide demand. That's either: - A cumulative multi-year contract value (e.g., 5-10 year commitments), - A pipeline of signed letters of intent, - Or a total addressable market projection dressed up as "backing."
My experience with DeFi summer taught me that yield projections are always exaggerated. In 2020, I built a Python arbitrage bot that captured $18,000 in fees over three months. The theoretical APY was 140%. After accounting for gas spikes during the Sushiswap migration, my actual return was 40% lower. The same principle applies here: theoretical customer demand is not realized revenue.
If even 25% of that $40 billion converts to actual revenue over three years, that's $10 billion, or $3.3 billion per year. Compare that to Nebius's current AI cloud revenue (likely under $500 million). That's a 6x increase. Possible? Yes. But not without massive execution risk.
Let's look at the GPU math. $775 million in debt. After interest, legal, and operational costs, maybe $600-650 million is available for hardware. An NVIDIA H100 GPU costs roughly $25,000-30,000 at bulk pricing. That buys 20,000 to 26,000 GPUs. That's about 2% of the total H100s expected to ship in 2024 (around 1.5 million units). It's not negligible, but it's not game-changing either.
But wait—Nebius claims "over $40 billion in customer backing." If customers are truly committing to that level of capacity, why raise only $775 million in debt? That's a 2% debt-to-backing ratio. Either the backing is highly non-binding, or the debt is just the first tranche of a much larger capital plan.
Contrarian: The Retail vs. Smart Money Divergence
The market narrative around Nebius is bullish. Debt over equity is seen as disciplined. The $40 billion backing is treated as a validation. But smart money—specifically the lenders—are taking a different view. Senior secured debt is the safest claim on the company's assets. If they truly believed in the $40 billion opportunity, they would have offered unsecured debt or equity at a lower cost. Instead, they demanded collateral.
This is reminiscent of the Terra/Luna collapse. In 2022, I had shorted UST using CDPs because I modeled the algorithmic peg's fragility. The market believed in the 20% yield. I saw a death spiral. The lenders here are pricing in risk. They want first claim on the GPUs. That tells me they see downside scenarios.
Here's a counterintuitive angle: the $40 billion customer backing might actually be a liability. If those customers are large tech firms or sovereign funds, they hold leverage over Nebius. A few cancellations could trigger a liquidity crisis. And because the debt is secured, the lenders can seize the assets faster than Nebius can pivot.
Compare this to CoreWeave, which raised over $1 billion in debt and equity combined, and has publicly disclosed contracts with Microsoft. Nebius hasn't named a single customer. That's a red flag.
Also consider the geopolitical risk. Nebius's Yandex heritage means it's under scrutiny from EU and US regulators. The company says it has relocated, but the supply chain for NVIDIA GPUs is tightly controlled. The US government restricts advanced chip exports to certain entities. Even if Nebius is compliant today, a shift in sanctions policy could cut off its hardware supply. That's a single point of failure.
Takeaway: What This Means for AI Infrastructure Investors
The AI cloud market is a commodity business disguised as a growth story. GPU compute is becoming a traded good, like electricity or bandwidth. The margins are shrinking. The debt for Nebius is a bet that demand will outstrip supply for at least three to five years. But the same bet is being made by every hyperscaler and every GPU cloud startup. We're heading toward an oversupply crunch.
If you're invested in AI infrastructure through tokens like Render Network or Akash, or through equities like NVIDIA, watch Nebius as a canary. If it struggles to service its debt, that signal will ripple across the sector.
My take: the $40 billion number is plausible only as a cumulative, non-binding pipeline. The real question is conversion rate. And based on my experience stress-testing yield models, conversion rarely exceeds 20% for early-stage infrastructure. That puts effective demand at $8 billion over three years—still large, but not extraordinary.
The smart move? Monitor the debt covenants. If Nebius files for an extension or restructuring, that's the exit signal. Until then, this is a high-risk, high-leverage play in a market that rewards speed, not loyalty.
Code doesn't lie. The balance sheet will tell the truth soon enough.
Yield is just delayed volatility. And this yield comes with a debt bomb attached.
NFTs are illiquid promises. So are unverified customer commitments.
Survival beats speculation. Nebius will either survive the debt servicing cycle or become another cautionary tale.