Neocloud (GPU cloud)
A cloud provider built around renting GPU (graphics processing unit) clusters for AI workloads, such as CoreWeave, Lambda or Nebius, funded by debt secured on the hardware and by multiyear contracts with a handful of large customers.
A neocloud is a cloud provider built around renting GPU (graphics processing unit) clusters for AI workloads, funded by debt secured on the hardware and by multiyear contracts with a handful of large customers. CoreWeave listed on Nasdaq on March 28, 2025 with one customer, Microsoft, accounting for 62% of its 2024 revenue. The whole neocloud model sits in that number. The operator buys Nvidia GPUs by the tens of thousands, houses them with wholesale colocation providers or in its own sites, and rents them by the hour or, mostly, under multiyear take-or-pay contracts: the customer pays for reserved capacity whether it uses it or not. Those contracts back GPU-collateralized debt (CoreWeave raised $7.5 billion in May 2024 from a group led by Blackstone and Magnetar, secured on the hardware). For the property underwriter, the change in scale is abrupt. A GB200 NVL72 rack is worth several million dollars, and hardware value per MW of IT capacity far exceeds the value of the building around it: a data hall loss becomes an equipment loss. Accumulation has to be read at the operator rather than at the site owner, since the same neocloud leases halls on ten or so campuses. For the business interruption underwriter, revenue concentrated on two or three customers matters as much as the engineering. One caveat: obsolescence is fast, and the value of a chip generation drops as soon as the next one ships, which complicates insured values and the basis of settlement after a loss.
In March 2025, OpenAI signed a contract with CoreWeave worth up to $11.9 billion over five years, raised to a total of up to about $22.4 billion by September 2025; that same month, Nebius announced a $17.4 billion, five-year contract with Microsoft for capacity in Vineland, New Jersey.
neocloud, GPU cloud, GPU-as-a-service, AI cloud provider
Follow the research
One letter a week. First, the week's underwriting note, when one is published: a dated, argued decision, signed by Paul Bernardski. Then what the observatory published on cyber risk, political risk and insurance. No account, no tracking.
No spam, one-click unsubscribe, no data ever sold.