Serenity Capital just published a public short thesis on CoreWeave (CRWV). The headline: 42% of EBITDA goes to interest payments. That’s not a margin squeeze. That’s a structural debt trap disguised as a growth story.
Context: Why Now? CRWV is the poster child of the neocloud sector—GPU-as-a-service, over $100 billion in backlog, 59% adjusted EBITDA margins. The market loves the narrative: AI demand infinite, hardware scarce, first-mover advantage. But Serenity is betting the opposite direction. They’re not arguing demand is fake. They’re arguing the capital structure is unsustainable. And the numbers back it up.
Core: The Numbers Don’t Lie Let’s deconstruct the math. From Serenity’s data: interest expense $640M, 42% of EBITDA. That implies EBITDA ~$1.524B. At 59% adjusted EBITDA margin, implied revenue ~$2.58B. A $100B backlog means 39x annual revenue in future commitments. That’s not a SaaS backlog—it’s a long-term capacity contract stack. Every contract is a promise to deliver GPUs. To deliver, CRWV must borrow more to buy more NVIDIA chips. The debt spiral is built into the business model.
I’ve seen this pattern before. In my Terra/Luna recon, the algorithmic stablecoin model looked sustainable on paper—high yields, growing TVL—until the leverage cascade hit. CRWV is not a stablecoin, but the same dynamic applies: when the cost of capital exceeds the return on deployed assets, the house of cards trembles. The ledger never sleeps, only updates. And right now, CRWV’s ledger shows a debt-to-EBITDA ratio that would make a junk-rated telecom blush.
Let’s talk about interest coverage. 42% of EBITDA goes to interest. Traditional investment-grade thresholds require coverage above 3x (interest/EBITDA <33%). CRWV is at 42%—that’s high-yield territory. Any drop in EBITDA (say, from a client delaying deployment or a price war on GPU compute) pushes coverage past 50%. Then debt refinancing costs spike, and the loop tightens. The balance sheet is the bottleneck.
Contrarian: The Backlog Mirage Here’s what the bulls miss: $100B in backlog is both a moat and a millstone. It locks in revenue, yes. But it also locks in capital expenditure commitments. CRWV must build data centers, buy H100s and B200s, and pay for power and networking. Every dollar of backlog is a future liability on the asset side. If a single top-3 client (say, a certain AI lab) decides to build its own compute or shift to a cheaper provider, that backlog frags. Customer concentration is the silent killer. Based on my audit of Uniswap V2’s contract, I learned that what looks like a safety net often hides a trapdoor. Here, the trapdoor is the “take-or-pay” clause. If clients can reduce usage under contract terms, the backlog’s quality degrades.
And the moat? It’s shallow. NVIDIA is both supplier and potential competitor (DGX Cloud). AWS, Azure, GCP have cheaper capital and can replicate the same GPU clusters. CRWV’s edge is speed and niche focus—but speed is only a temporary moat in a borderless war. The real question: can CRWV refinance its debt at favorable rates when the next rate hike or credit crunch hits?
Takeaway: The Next Watch Serenity’s thesis is not about demand—it’s about capital stack fragility. The next catalyst is not a revenue miss but a debt covenant breach or a downgrade by Moody’s/S&P. Watch the interest coverage ratio quarterly. If it creeps above 45%, the short thesis accelerates. The truth is hidden in the block height—or in this case, the debt maturity schedule. Adapt or get front-run by your own assumptions.
Chaos is just data waiting to be indexed. And the data on CRWV’s balance sheet is screaming for a repricing.