CoreWeave - The Cost of Borrowed Time? - Initiation
All,
Please find our initial analysis of CoreWeave here.
Note: Beta testers can check CoreWeave on the portal here.
CoreWeave’s credit comes down to one question: can a business that borrows to build reach the end of its build before the contracts behind it run out? The next two years look well covered, with the debt raised against specific GPUs and customer contracts and management guiding to at least 8GW of active power, so there are no signs of slowing down. Around 2029–31 most of the capacity has to be re-sold, debt will be at its peak, and the first generation of chips needs replacing, all at once. Will customers renew at today’s prices, and will management’s guidance on attractive pricing for older chips hold when the real test arrives, given the six-year life they assume for these chips? Will funding markets stay open to finance the rest of the build, and how much further does the push go beyond 8GW? And if demand drops, how do management and its investors react: do they keep building, or do they step in with fresh equity? The answers decide whether the 2030–32 maturities are refinanced at par or restructured, which is why the outcome is so binary.
Investment Rationale
CoreWeave is winning contracts, but not yet cash. Q2 26 revenue rose 112% YoY to $2.6bn against a $104bn backlog, yet operating cash flow of $679m covered about a tenth of $6.4bn of capex, with $13.5bn of gross debt raised to bridge the gap. With c.$42bn of debt, the SUNs sit behind c.$21bn of senior secured debt, and the build to 8GW must be finished before contracts start rolling off in 2029–31. At 78 (14.3% YTM), the 8.50% € 2032 SUNs are priced for that to succeed. Fundamentally, this is an equity story, and in the SUNs we take senior equity risk, so we want more than the usual 20% for a long-term AI bet with market-wide as well as company-specific risk. A 14% yield doesn’t pay for that, and a return of c.30% would mean buying at around 50, well off the market. We are therefore not investing. The outcome hinges on whether customers renew at today’s prices in 2029–31, and we would look again at a price that pays us for that risk, or on evidence that renewals are holding or that partners are funding more of the build.
If AI demand holds and the 8GW build is funded and delivered, the 2030–32 maturities are refinanced and the SUNs pull to par, with the coupon on top. This depends on a number of unit economic factors lifting the credit profile and allowing the business to reach its potential.
The downside is where the risk truly sits: if renewals come in weaker and cheaper while funding markets tighten, the business can only support part of the debt it has raised. Secured lenders are protected by the GPUs and contracts behind them, but the SUNs sit behind c.$21bn of secured debt plus c.$16bn of operating leases and would have to be restructured. In a recapitalisation, the new cash mainly benefits those providing it and the senior lenders. With Nvidia’s support the SUNs recover c.38% of face; without a rescue, nothing is left for them.
We are not considering a short position, because there is no fundamental short-term rationale that we can focus on, without a near-term catalyst shorting the bonds here would be purely a trade on sentiment, which may be fair in this environment, but is not the fundamental strategy that we are pursuing the main catalyst that we are anticipating is 3-5 years away.
We would turn more cautious if new secured debt keeps pushing the SUNs further down the capital structure, or if the build continues without contracts behind it.
Key Conclusions
The next two years are largely secured by contracts, and the debt raised against them is backed by specific GPUs and customer cash flows. Management guides to attractive pricing for older chip generations, but with a six-year chip life the real test comes in 2029–31.
If the build lands and renewals hold at today’s prices, the business grows into its debt. EBITDA rises faster than borrowing, interest cover improves once growth capex slows, and the 2030–32 maturities can be refinanced on the strength of a completed fleet. In that world the SUNs are repaid at par, and holders collect a high coupon for their risk.
If renewals come in weaker, the picture changes but the bonds hold. Customers renew a little less often and at slightly lower prices, so revenue settles below plan while the cost base and debt stay much the same. The business still covers its interest and can refinance, but with less headroom and at a higher cost, and most of the hit falls on equity rather than debt. The SUNs are still repaid, but the cushion beneath them is much thinner, giving a weaker outlook for potential growth.
A hard downturn, with sharply lower renewals, the build cut short and funding markets closing, is where the SUNs are restructured. With Nvidia or other equity support they recover c.38% of face; without it, the secured debt and any new money absorb all the value and the SUNs are left with nothing.
The credit is therefore less about the next few quarters and more about evidence on renewal pricing, contract length and who funds the rest of the build. More customer prepayments or partner capital, rather than more secured debt, would be the clearest sign the risk to unsecured holders is falling.
Happy to discuss,
Eashan
E: ekapoor@sarria.co.uk
T: +44 203 744 7055