A viral chart of CoreWeave bond yields sends Oracle, Dell and neocloud shares down

SiliconANGLE's account covers share prices and bond yields. It reports no change in service to any CoreWeave customer. For a team that reserved a block of GPUs at a neocloud last spring, the question from finance is simpler: is the capacity safe?

So the supply-risk reading is an inference. We think it is a fair one for a specific group of buyers, and the chain behind it is short. CoreWeave grew from 3 data center facilities to 51 in four years ^[2], which is 48 added, or about 12 a year ^[12]. SiliconANGLE says it financed much of that buildout with debt ^[2] and reports that CoreWeave has $50 billion in outstanding obligations, a significant portion raised through bonds ^[3]. A yield increase on bonds is usually caused by falling prices, which indicates investors are selling ^[4]. In our view, the buyer's exposure sits in capacity that is promised but not yet built, because that is the part that depends on raising more money. The source does not say how much of CoreWeave's contracted capacity is still to be built.

CoreWeave's answer came from its chief executive. Mike Intrator reportedly wrote in a memo to employees: "when you are building a new type of technology company there will be periods of time when the market misunderstands and acts irrationally. We are confident in our backlog which more than supports all of our debt obligations." ^[9]

The backlog was $104 billion in June ^[10]. Set against the reported $50 billion in debt, that is 2.08 times ^[11]. But backlog measures expected revenue from cloud contracts ^[10], so the ratio does not account for what it costs to run the sites. Much of it is tied to infrastructure deals with Microsoft, Meta and OpenAI ^[10].

The record so far is a day of share moves across neoclouds, Oracle, Dell and AMD. A vendor's share price can still become a customer's problem. Whether it does for a given team depends on two things you already know about your own contract.

The first is how much of your spend is committed ahead of use, through prepayment, reserved blocks or multi-year terms, versus paid as used. The second is how long it would take to move the workload to another provider. Crossed, they give four cells.

Pay-as-used and quick to move: the chart is a news item, and nothing needs to change. Committed and quick to move: keep a second provider tested, which costs engineering time you may not want to spend. Pay-as-used and slow to move: the exposure is the day you have to leave, so the migration plan is the document to write.

Committed and slow to move is the cell where a provider's bond yield becomes your supply risk. We would ask that provider, in writing, for the delivery schedule on any capacity not yet live. The cost is a harder conversation at renewal.