CoreWeave (NASDAQ:CRWV) co-founder and Chief Development Officer Brannin McBee said the company continues to see “overwhelming and insatiable demand” for AI infrastructure, with growth coming from hyperscalers, AI labs and enterprise customers.
Speaking with Mark Murphy, executive director at JPMorgan, McBee discussed CoreWeave’s demand outlook, financing strategy, customer mix and supply constraints. Murphy opened the session by noting what he described as an “extraordinary” period for the company, citing full-year fiscal 2025 revenue of $5.1 billion, nearly $100 billion in backlog and $40 billion in new bookings in a single quarter.
AI Demand Broadens Across Customer Segments
McBee said CoreWeave was “very excited” to announce Anthropic as a customer last quarter and said that “nine of the top 10 global non-China AI labs” are on the company’s platform. He described that as important both for customer diversification and as evidence of the platform’s performance.
Enterprise demand, he said, is also growing quickly, though those deals are typically smaller than CoreWeave’s largest contracts. McBee said CoreWeave added more than double the number of logos in the fourth quarter than it had in any prior quarter, and that those logos were coming from enterprise customers.
“Enterprise is rapidly growing on our platform,” McBee said, adding that the company sees broad interest in inference workloads from that segment.
Inference Drives Next Wave of Growth
McBee identified inference as the primary force behind the next phase of AI infrastructure demand. While training was central to building AI models, he said the industry is now focused on monetizing those models, which requires substantial infrastructure.
He said CoreWeave builds AI infrastructure that can be used for both training and inference, rather than designing separate systems for each use case. Customers can move seamlessly between the two, McBee said, including using the same infrastructure for training one hour and inference the next.
McBee said the company’s prior disclosure that inference accounted for materially more than 50% of power draw reflected both the maturation of Hopper-based infrastructure and the arrival of new customers that only need inference. He singled out financial services as one area of demand, saying that sector represents more than $10 billion of CoreWeave’s backlog.
He also said demand remains healthy for multiple generations of Nvidia GPUs, including Ampere, Hopper and Blackwell, and said customers are choosing technology based on workload fit rather than only seeking the newest generation.
Company Says It Remains Focused on Nvidia GPUs
Asked about the recently discussed Blackstone-Google TPU cloud deal, McBee characterized it as another demand signal for AI infrastructure. He said CoreWeave’s customers continue to ask for GPUs, specifically Nvidia GPUs, and said the company is not seeing customers compare TPU pricing against GPU pricing in its pipeline.
McBee said Blackstone and Google remain “massive and fantastic partners” of CoreWeave. He noted that Blackstone participated in CoreWeave’s first delayed-draw term loan transactions and said Google is a large GPU customer, describing it as a multi-billion-dollar consumer of CoreWeave’s GPU platform.
McBee said CoreWeave would consider operating other semiconductor technologies if customers asked for them at sufficient scale. However, he said the company’s operational stack is not dependent on the underlying hardware and that CoreWeave has demonstrated flexibility by moving from Hopper to Grace Blackwell architectures.
Contracts, Margins and Financing
McBee said CoreWeave’s enterprise contracts generally resemble those with AI labs and hyperscalers, with four- to six-year durations and similar margin targets. He said the company targets a 25% contribution margin at the special-purpose-vehicle level for operating clusters, regardless of the underlying GPU SKU.
On margins, McBee said first-quarter performance represented a trough and that the company expects to expand from there as infrastructure ramps in the second and third quarters. He said deployments now coming online are tied to contractual commitments, giving CoreWeave visibility into revenue, costs and margin profiles.
McBee also discussed CoreWeave’s financing model, saying the company views itself as a debt-financed business because its infrastructure is supported by physical assets and take-or-pay contracts. He said CoreWeave uses asset-level financing for large contracts and infrastructure, while parent-level financing fills remaining gaps through instruments that may include convertibles, high yield and equity issuance.
He said the company’s cost of capital has fallen substantially as it has built an execution track record. McBee cited DDTL 5, announced the same day, as a non-investment-grade example completed at roughly SOFR plus 450 basis points and about 70% loan-to-cost. He said the $3.1 billion to $3.5 billion facility drew $19 billion of demand.
Supply Constraints Expected to Persist
McBee said he does not see supply and demand balancing before the end of the decade. He said the current bottleneck is not electricity itself, but “powered shell” capacity, or the ability to consume electricity at the rack level.
He identified electricians, transformers, backup batteries and other components as constraints in global supply chains that were not built to scale at the pace of AI demand. CoreWeave has more than 43 sites in operation, McBee said, and has spent years navigating supply-chain disruptions.
McBee closed by emphasizing that securing power or signing leases does not automatically translate into revenue. He said the market may underappreciate the difficulty of deploying GPUs and turning power commitments into billable compute capacity.
“There is a chasm of execution risk between contracting for power and delivering billable GPU hours,” McBee said.
About CoreWeave (NASDAQ:CRWV)
CoreWeave is a U.S.-based provider of GPU-accelerated cloud infrastructure designed to support compute-intensive workloads such as artificial intelligence, machine learning, visual effects rendering and other high-performance computing applications. The company supplies access to large fleets of modern GPUs and complementary infrastructure that enable customers to train and deploy large models, run inference at scale, and process graphics-heavy workloads with low latency and high throughput.
CoreWeave’s product offering includes on-demand and dedicated GPU instances, bare-metal servers, private clusters and managed services tailored for enterprise and developer use.
