CoreWeave just lost roughly 40% of its stock value over the past month, and Tuesday's earnings may determine whether the neocloud model survives contact with a rising cost of capital.
The numbers define the pressure. As of the second quarter, CoreWeave carried $35.6 billion in debt against $5.5 billion in cash, with a $104.2 billion contracted revenue backlog. Its debt-to-equity ratio stands at 702%, up from roughly 291% a year ago. The yield on its 9.75% bond due in 2031 hit 16.7% last week. Its five-year credit default swap spread closed at 1,278 basis points on Thursday — a level considered distressed — before pulling back to around 1,152 on Tuesday. The market is not questioning CoreWeave's customers; it is questioning whether the company can keep financing the infrastructure those customers needs.
The capex math is what makes this hard to paper over. CoreWeave guided for $35 billion to $39 billion in capital expenditures for the full year. Wells Fargo estimates 2027 spending could reach $66 billion based on an expected 1.5-gigawatt capacity addition, requiring at least $25 billion more in debt. The company owes roughly $15 billion in principal payments through 2028. Issuing equity to close the gap — the path Meta and Oracle recently took — isn't an option for CoreWeave, Explosive Options' Bob Lang argues. The stock has fallen more than 73% from its June 2025 peak above $183, recently trading near $47. What analysts want from Tuesday is a credible financing sequence, not just a revenue print against guided quarterly revenue of $3.4 billion to $3.6 billion.
Who actually gets to run AI at scale
That is the real question this earnings report forces. The neocloud model — borrow at scale, build ahead of demand, let the backlog justify the debt — assumed capital would stay cheap long enough for revenue to compound. It rhymes with the fiber overbuild, and with other capital-intensive infrastructure plays that front-loaded spend against long-dated commitments.
If CoreWeave steadies its financing story on Tuesday, it validates that contracted AI demand is bankable collateral and that purpose-built GPU clouds can compete with hyperscalers on more than price. If it can't, the market will read it as confirmation that only balance sheets the size of Microsoft or Google can absorb the capital cost of running frontier AI infrastructure — and the window for independent neoclouds narrows sharply. For PE partners and operators alike, that is the binary this earnings report is actually pricing.
Get the Daily Frontier
Stories like this one land in your inbox every weekday morning. Subscribe free at factua.com/blog.
Know someone who should be reading this? Send them to factua.com/blog to subscribe.
