An interview with CoreWeave executives offers a detailed view into the current AI infrastructure cycle. Speaking with Tae Kim, CoreWeave co-founder and chief development officer Brannin McBee and vice president of corporate development and investor relations Nick Robbins described demand for AI compute as continuing to rise rather than cooling after the earlier wave of GPU buying. Robbins summarized the state of the market in direct terms, saying AI demand seems to intensify every day in new ways.
CoreWeave was founded in 2017 and provides large-scale GPU compute to startups and large enterprises. It is described in the source article as an early market leader in the neocloud category and as the only cloud provider to receive the highest Platinum rating from AI research firm SemiAnalysis. The company sits in the middle of the AI infrastructure chain: it serves leading customers including OpenAI, Anthropic, Meta, Google, Microsoft and Nvidia, while also seeing demand from research labs, enterprise users and hyperscale cloud providers.
Demand accelerated with agentic AI and reasoning workloads
McBee said the real beginning of the agentic AI demand wave appeared in the fourth quarter of last year. At that time, CoreWeave was holding engineering-level conversations with customers about products they expected to bring to market in the first quarter of this year. He described the company’s engineering relationship with customers as deeply interconnected, allowing CoreWeave to see trends ahead of time rather than respond only after changes have already taken place.
From the product side of the AI market, McBee said the first quarter became a major inflection point for reasoning and AI consumption, and that acceleration has continued. This is the context behind the executives’ central message: the next phase of demand is not only about whether GPUs are available. The workload mix itself is changing as agentic AI and reasoning models become more important, and that change is lifting the role of CPUs and storage alongside GPUs.
CPU and storage needs are rising beside GPUs
Asked about 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 offered a complete cloud product. In his view, the question is not whether the company is just starting to add CPUs, but what customers need and whether that requirement is rising in relative terms. His answer was clear: it is rising. He also said storage demand is increasing compared with earlier generations as agentic and reasoning capabilities become more central within models.
Robbins said customers should expect to see large numbers of Vera CPUs deployed next to large numbers of Vera Rubin servers. He explained that CoreWeave fundamentally redesigned its underlying data center plan last year to leave room for more storage and more CPUs beside GPUs. According to Robbins, this decision came from CoreWeave’s specific position in the ecosystem. He said the company is the only independent cloud provider serving all of the most advanced technology users, and that no other independent AI cloud provider can name Anthropic, OpenAI, Meta, Google, Microsoft and Nvidia as customers in the same way.
On whether CoreWeave will mainly use Nvidia Vera CPUs in the future, Robbins said the answer depends on the specific workload. The company describes its approach as customer-driven. Robbins said CoreWeave has already disclosed that it expects to be an early and important adopter of Vera CPU. At the same time, its fleet is currently still mainly AMD, and that mix will change over time according to customer requirements. He added that customer interest in Vera CPU is very strong.
McBee used that point to explain how CoreWeave’s contracts work. More than 98% of the company’s revenue is contract-driven, he said. CoreWeave is not guessing what infrastructure customers want. Customers tell the company very explicitly what configuration they need, and those customer requirements define what CoreWeave builds.
Competition is framed around execution, performance and orchestration
The interview also addressed the competitive landscape. CoreWeave competes against 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 describe the company’s differentiation through third-party validation. Excluding China, nine of the world’s top ten AI labs use CoreWeave’s platform, he said. SemiAnalysis has also consistently placed CoreWeave alone at the highest performance level.
McBee said the company’s GPU allocations are not the result of a personal relationship with Jensen. Instead, he argued that suppliers have deep confidence in CoreWeave’s execution record and engineering capability, and believe the company can represent their products well on a global scale. Robbins added that CoreWeave wins hyperscale cloud customers because it is strong at execution: it can build systems very quickly and keep them running well.
Robbins broke down the customer base further. Research labs choose CoreWeave, he said, because the company offers the highest-performing version of the technology and the best efficiency per token. Enterprises choose the company because the infrastructure works well and because CoreWeave has built a strong orchestration layer, which is one of the reasons for recognition such as the Platinum rating. He also said the company has built a more mature layer of capabilities around inference and development tools to help enterprises put AI into production.
That enterprise work includes helping less technically mature companies turn data into models, and then turn those models into agents that can run internally. Robbins said this also gives CoreWeave the ability to cross-sell its cloud services during that process.
Powered data center shells are the main bottleneck
When asked about current bottlenecks, McBee pointed first to powered shells, meaning data center shells that already have power. More precisely, he said the constraint lies in the components inside those shells. The interviewer specifically mentioned electricians, and McBee agreed that this is part of a complex area. He stressed, however, that CoreWeave already has 49 such sites online and operating. The company is not relying on one or two sites; it has repeated the process 49 times.
McBee said that execution history has given CoreWeave a deep record and significant knowledge about how to deal with supply chain issues. It has learned which suppliers in that chain are suitable partners and which are not. That is why the infrastructure race described in the interview is broader than chip procurement alone. It includes access to powered data center capacity, server deployment, supply chain execution and the ability to optimize cost per token for customers.
The interview also touched on HBM memory cost and shortage concerns. Robbins said CoreWeave’s business model is structured so that when it signs GPU purchase orders and determines how much it will pay, it also locks in the GPU price charged to customers. More broadly, that means the server price, and the server price includes HBM costs. Robbins said component access is not currently the largest bottleneck; the largest bottleneck is the powered shell. He also noted that this answer can change back and forth at some point in the future.
On Vera Rubin deployment, Robbins said CoreWeave was the first company in the world to start and fully validate VR, or Vera Rubin, racks. He said the company had also done this last year with GB200 and GB300. Robbins expects VR to begin appearing later this year, while the truly large-scale and strong deployment ramp should run throughout 2027. He compared the timeline with GB: GB began appearing in 2025, but its real large-scale ramp has run through 2026. In his view, VR will follow a very similar pattern over the next 12 to 18 months.