Interviews with 13 CoreWeave customers and prospects suggest the company won much of its business because hyperscalers could not deliver GPUs fast enough, and that its customers still spread their spending across other clouds, SiliconANGLE reported Saturday. The research, led by analyst Dave Vellante, asks whether the neocloud can keep those customers as GPU supply loosens.

What Happened

Vellante's team ran the interviews with research firm Qualitate, totaling more than seven hours, with respondents in the United States, Canada, the Netherlands, the United Kingdom and Germany. Nine work at organizations classified as large enterprises. Seven respondents use CoreWeave significantly today, three are evaluating it, one ran a successful pilot without a rollout, and two decided against it, one choosing Azure and the other its own hardware.

Ten of the 13 said specific problems getting GPU capacity or products from hyperscalers were part of their buying decision. A healthcare IT director said CoreWeave delivered capacity in weeks, where AWS needed months, and that it beat Lambda Labs mainly on enterprise support, cluster availability at scale and flexibility.

RelatedUK Cyber Agency Publishes Post-Quantum TLS Settings and Urges Move to TLS 1.3

All seven current users run inference as well as training. One healthcare customer's mix moved from 90% training and 10% inference in the first year to 60-40 today, with about an even split expected within 12 months. All seven also keep hyperscaler relationships. One large technology buyer puts 10% to 25% of its GPU spending on CoreWeave, while another healthcare buyer places more than half there.

Why It Matters

The financials explain why retention matters so much. CoreWeave reported second-quarter revenue of $2.6 billion, up 112% from a year earlier, and a revenue backlog of $104 billion, with a further $25 billion added in the first month of the current quarter. But in the first half it generated $3.7 billion in operating cash flow against $14.1 billion in capital spending, a gap of about $10.5 billion.

Revenue is also concentrated. Three customers accounted for 72% of second-quarter revenue, and 98% came from committed take-or-pay contracts. Those contracts give the company predictable income as it takes on debt and lease obligations to fund its buildout. CoreWeave's latest 10-Q warns that a move toward pay-as-you-go or other consumption-based pricing could affect cash-flow predictability, margins and its financial condition.

The interviews point to where that pressure could come from. A German enterprise evaluating options found Nebius very similar for basic inference. A healthcare director said CoreWeave has SOC 2 Type 2 certification but lacks logging as granular as AWS CloudTrail and a real-time compliance dashboard, gaps that matter for regulated workloads.

Build Versus Rent

Several respondents weighed running their own GPUs. Estimates of the utilization needed to justify on-premises hardware ranged from 50% to 80%. A healthcare director said a total-cost-of-ownership model showed about 80% utilization was needed just to break even, which ruled it out. That math currently favors renting, but it also means customers can shift work as their usage becomes steadier.

What to Watch

Watch how many evaluations and pilots turn into production contracts, and whether CoreWeave's free cash flow improves as it offers more flexible pricing. Renewal rates on the large take-or-pay deals and the growth of inference workloads will show whether customers stay once GPUs are easier to get elsewhere. For healthcare and other regulated buyers, compliance tooling such as detailed audit logging may decide whether CoreWeave grows beyond overflow capacity.