A 15% Bond Yield and a $52 Billion IOU: The Math Wall Street Is Quietly Running on CoreWeave

Posted on by

(SeaPRwire) –   By: Reginald Vance

The Panic Has a Number, and That Number Is 15

Markets don’t panic over press releases. They panic over yields. On Wednesday, CoreWeave shares dropped 15.46%, with another 5.62% shaved off in post-market trading. The trigger wasn’t an earnings miss. It was Chamath Palihapitiya posting a chart on X showing CoreWeave’s bond yield touching 15%, paired with a question that landed like a grenade: is the AI buildout creating a private credit bubble? That same day, the Financial Times ran an interview with co-founder and finance lead Brannin McBee, who said he wants to borrow even more, this time from insurance companies. His framing was striking. The insurance sector’s access to capital “is unlimited,” he said. “Like, that’s all the money in the world that sits in the insurance tranche.” Meanwhile, Deutsche Bank’s Jim Reid mapped the damage with precision. AA-rated AI names like Meta, Alphabet, Nvidia, and Amazon borrow at roughly 5–5.5% at the front end. BBB names like Oracle and SpaceX cluster around 8–8.5% at 10–12 years’ duration, already at the top of the investment-grade distribution. Then there’s CoreWeave, rated BB+, sitting in the red box at the very top of the chart. The capital markets have rendered a verdict on hardware scaling economics. Building GPU capacity on borrowed money is no longer treated as infrastructure. It’s treated as speculation. When the market demands 15% to hold your paper while Microsoft pays half that, the bottleneck isn’t silicon supply. It’s credibility.

The Ledger: Debt, Leases, and the RPO Shield

Now the numbers that matter. According to CoreWeave’s Q2 earnings release, the company carries operating lease liabilities of $16.3 billion and total debt of $35.6 billion. Call it $52 billion in combined obligations. Against that, it produced $4.6 billion in quarterly revenue, and operating expenses exceeded sales, leaving a net loss of $1.4 billion. On paper, that’s an ugly picture. But Evan Schlossman, principal at Neostellar Capital Corp., a publicly traded VC fund that bought a $25 million stake in CoreWeave in 2024 and sold 40% of it for a $15 million gain in Q2 2025, argues the critics are reading the wrong page of the filing. The counterweight is $103.7 billion in remaining performance obligations, 41% of which is scheduled to be paid through 2028. These RPOs are non-cancellable, take-or-pay contracts. The customer pays whether or not they use the compute. And the customers are Microsoft and OpenAI, with the top three clients generating 72% of Q2 revenue. CoreWeave itself states that 98% of revenue comes from committed take-or-pay contracts, and that financing is structured so customer cash flows repay the associated debt. Schlossman’s point about OpenAI is subtler. Its balance sheet is thinner than Microsoft’s, but its commitments to CoreWeave are backstopped by RPOs OpenAI holds from other tier-one counterparties. The contracts form a chain, and the chain anchors on balance sheets that won’t fail. Even if OpenAI vanished tomorrow, he argues, the contracted compute capacity is worth more today than when the deals were signed. It could be resold. That’s the bull case, stated cleanly: the debt is collateralized not by GPUs, which depreciate brutally, but by contractual claims on the strongest credits in technology.

Cash Flow Gravity and the Coming Consolidation

Here’s where the hardware wargame gets honest. The bear case isn’t theoretical. Gartner estimates 80% of neocloud providers will cease independent operations by 2030 because revenues won’t cover outgoings. Oracle already issued a force majeure citation to Blue Owl over a massive New Mexico data center slated to come online in 2028, a signal that customers are checking their legal exits if delivery slips. And Francine McKenna, the accountant behind The Dig, flatly doesn’t believe Microsoft will honor CoreWeave’s RPOs if compute demand softens. That skepticism matters because Schlossman himself concedes the only clean escape hatch in the Microsoft agreement is exactly an Oracle-style force majeure event. So the whole structure rests on one premise: delivery. If CoreWeave builds on time, tier-one cash flows service the debt and generate free cash on top. If it doesn’t, the take-or-pay shield evaporates and 15% paper becomes self-fulfilling. Schlossman’s own caveat is the tell. “If there is something that goes wrong, I think you’re going to see that it’s going to matter significantly who owes you money,” he said. Read that as a consolidation map. Neoclouds whose receivables trace back to Microsoft survive. Neoclouds backed by venture money, selling to seed-stage customers, get absorbed or buried. This is the same endgame I’ve watched in semiconductor capacity cycles for twenty years. Capital intensity sorts the field. Cheap credit built the fleet, expensive credit culls it, and the hyperscalers end up owning the survivors’ capacity at distressed prices. My practical read for anyone underwriting this space: stop valuing GPU fleets and start valuing counterparty quality. Trace every dollar of contracted revenue to its ultimate balance sheet. If the chain ends at a foundation model lab burning venture cash, discount it hard. If it ends at Redmond, the yield is the opportunity, not the warning. CoreWeave isn’t a bubble story or a redemption story. It’s a stress test of whether contractual paper can substitute for profit, and the answer arrives in 2028, one data center delivery at a time.

Author bio: Reginald Vance is a venture partner specializing in semiconductor valuation and advanced materials, with two decades of experience underwriting capital-intensive compute infrastructure and advising institutional investors on hardware cycle risk.