CoreWeave co-founder and Chief Development Officer Brannin McBee and Vice President of Corporate Development and Investor Relations Nick Robbins discussed the state of AI demand and the neocloud market in an interview with Key Context. Their comments present CoreWeave as a company sitting in the middle of the AI infrastructure chain: it serves leading customers such as OpenAI, Anthropic, Meta, Google, Microsoft and Nvidia, while also seeing demand from research labs, enterprise customers and hyperscale cloud providers.
AI demand moved into a new phase in the fourth quarter
CoreWeave was founded in 2017 and provides large-scale GPU computing capacity to startups and large enterprises. The company is described as an early market leader in the neocloud category. According to the source article, it is the only cloud service provider to receive the highest Platinum rating from AI research firm SemiAnalysis.
When asked when the wave of demand for agentic AI began to break out, Brannin McBee said CoreWeave saw the real beginning in the fourth quarter of last year. At that time, the company was speaking with customers at an engineering level about products they expected to bring to market in the first quarter of this year. He said this engineering relationship is important to how CoreWeave views customer demand, because it allows the company to see trends before it is forced to react to them after the fact.
From the product side of the AI market, McBee described the first quarter as a major inflection point for inference and AI consumption, adding that the acceleration has continued. Nick Robbins gave a concise assessment of the current demand environment. Asked whether there had been no sign of slowing in recent weeks compared with a few months earlier, he said demand seems to intensify every day in new ways.
CPU and storage are becoming part of AI infrastructure design
The interview also turned to the rising demand for CPUs relative to GPUs in the agentic AI wave. McBee said CoreWeave has been running CPUs since 2023 and has always had a full cloud product. In his view, the question is not whether CoreWeave has only just started adding CPUs, but what customers need and whether that need is rising on a relative basis. His answer was that it clearly is. He also said storage demand is rising compared with earlier generations as agentic and reasoning capabilities become more important in models.
Robbins answered directly when asked whether rows of Nvidia Vera CPU racks would be deployed next to Nvidia GPU servers. He said large numbers of Vera CPUs will be seen beside large numbers of Vera Rubin servers. Last year, CoreWeave fundamentally redesigned its baseline data center plan to leave room for more storage and more CPU capacity next to GPUs. He tied that change to CoreWeave’s position in the ecosystem, saying the company is the only independent cloud provider serving all of the most advanced technology users named in the interview.
On whether CoreWeave will mainly use Nvidia Vera CPUs in the future, Robbins said it depends on the specific workload. He described the company’s approach as customer-demand driven. CoreWeave has disclosed that it expects to be an early and important adopter of Vera CPUs. At present, its fleet is mainly AMD, but that configuration can change over time according to customer demand, and Robbins said customer interest in Vera CPUs is very strong.
McBee used that point to explain how CoreWeave’s contracts work. He said more than 98% of the company’s revenue is contract-driven. CoreWeave is not guessing what kind of infrastructure customers want; customers tell the company very clearly which configurations they need. In his framing, customers define what CoreWeave builds.
CoreWeave’s competitive position in neocloud
The interview compared CoreWeave with neocloud companies such as SpaceX, Nebius and Oracle, as well as hyperscale cloud providers such as Azure, AWS and Google. McBee said he prefers to discuss differentiation through third-party validation. Excluding China, nine of the world’s top ten AI labs use CoreWeave’s platform, and SemiAnalysis has consistently placed the company alone at the highest level for performance.
McBee also rejected the idea that CoreWeave receives GPU allocations because of a personal relationship with Jensen. He said the allocations reflect suppliers’ confidence in CoreWeave’s execution record and engineering capability, and their belief that the company can best demonstrate product capabilities on a global basis.
Robbins said CoreWeave wins hyperscale cloud customers because it is good at execution, can build systems very quickly and can keep them running well. It wins research lab customers by providing the strongest technical versions and the best efficiency per token. It wins enterprise customers because the infrastructure operates well and because CoreWeave has built a strong orchestration layer, which is one of the reasons for recognitions such as the Platinum rating.
Robbins added that CoreWeave has also built what he described as the most mature layer among AI cloud providers across inference and developer tools, helping enterprises put AI into production. That means building and delivering products that help less technically mature enterprises turn data into models and then into agents that can run internally, while also creating opportunities to cross-sell CoreWeave cloud services.
Powered data center shells are the current bottleneck
When asked about the current infrastructure bottleneck, McBee identified powered shells, meaning data center shells with power already in place. More precisely, he pointed to the components inside those shells, and agreed that electricians are part of the issue. He described it as a complex field. At the same time, he emphasized that CoreWeave already has 49 such sites online and running, so the company is not relying on one or two locations.
McBee said that record gives CoreWeave substantial experience in handling supply chain problems and in knowing which suppliers are suitable to work with and which are not. Robbins then addressed HBM memory costs and shortages. He said CoreWeave’s business model locks in the GPU price charged to customers at the same time the company signs GPU purchase orders and determines what it will pay. More broadly, he described this as the server price, which includes HBM costs.
Robbins said obtaining components is not the biggest bottleneck at the moment; the biggest bottleneck is the powered shell. On the Vera Rubin deployment ramp, he said CoreWeave is the first company globally to start and fully validate VR, meaning Vera Rubin racks. He said the company did the same last year with GB200 and GB300. He expects VR to start appearing later this year, with a truly large-scale and very strong deployment ramp running through 2027.
Robbins compared that expected pace with the GB cycle. GB began appearing in 2025, but its real large-scale ramp has run through 2026; he said quite a few units had already been deployed by the end of last year, while this year is the true large-scale deployment year for GB. He expects VR to follow a very similar rhythm over the next 12 to 18 months.