The original article, titled “An Interview with CoreWeave Executives: AI Demand Seems to ‘Intensify’ Every Day,” was written by Tae Kim and translated by Peggy for BlockBeats. The interview features CoreWeave co-founder and Chief Development Officer Brannin McBee, together with Nick Robbins, vice president of corporate development and investor relations. Their discussion focuses on the current state of AI demand and the neocloud market. The central message from the executives is direct: AI demand seems to intensify in new ways every day, while the real infrastructure constraint has expanded beyond whether GPUs are available.
AI infrastructure pressure is moving beyond the GPU question
CoreWeave is described as an innovative early market leader in the neocloud category. Founded in 2017, the company provides large-scale GPU computing power to startups and large enterprises. It is also the only cloud service provider to receive the highest “platinum” rating from AI research firm SemiAnalysis. Its position in the AI infrastructure chain gives it a view across multiple kinds of customers, including OpenAI, Anthropic, Meta, Google, Microsoft and Nvidia, as well as research labs, enterprise users and hyperscale cloud providers.
McBee said CoreWeave saw the real beginning of the agentic AI demand wave in the fourth quarter of last year. At that time, the company was having engineering-level conversations with customers about products they expected to bring to market in the first quarter of this year. He said this kind of deeply connected engineering relationship is an important lens for understanding customer demand because it lets CoreWeave see trends ahead of time rather than respond only after changes have already occurred. From an AI product perspective, McBee described the first quarter as a major inflection point for reasoning and AI consumption, with that acceleration still continuing.
When Tae Kim asked Robbins about the current state of AI demand and whether there had been no sign of slowing in recent weeks, Robbins answered: “It seems to intensify every day in new ways.” That statement frames the broader interview. Demand did not cool after the earlier wave of GPU buying; instead, agentic AI, reasoning models and enterprise AI applications are continuing to raise the requirements for next-generation infrastructure.
Vera CPUs, Vera Rubin servers and storage enter the design discussion
On the rising demand for CPUs relative to GPUs during 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 is only now starting to add CPUs. The question is what customers need and whether that need is increasing on a relative basis. His answer was clear: it is. He added that as agentic and reasoning capabilities truly emerge in models, storage demand is also increasing compared with earlier generations, and he expects that trend to continue.
Robbins said the answer to Kim’s question about Vera CPU racks being deployed alongside Nvidia GPU servers is yes. He said customers will absolutely see a large number of Vera CPUs placed next to a large number of Vera Rubin servers. Last year, CoreWeave fundamentally redesigned its base data center approach to leave room for more storage and more CPUs alongside GPUs. Robbins linked that decision to CoreWeave’s unusual position in the ecosystem: he said the company is the only independent cloud service provider serving all of the most advanced technology users, with Anthropic, OpenAI, Meta, Google, Microsoft and Nvidia among its customers.
That customer base creates what Robbins called a helpful flywheel or positive feedback loop. CoreWeave can understand where customers are taking the technology and then plan around that direction. Asked whether the company will primarily use Nvidia Vera CPUs in the future, Robbins said the answer depends on the workload. CoreWeave is customer-demand driven. The company has disclosed that it expects to be an early and important adopter of Vera CPUs, but its fleet today is still mostly AMD. Over time, that mix can change according to customer requirements. Robbins added that customer interest in Vera CPU is very strong.
McBee used the 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 infrastructure customers want. Customers are very explicit about the configurations they need, and everything is customer-driven. In his words, the customer defines what CoreWeave builds.
How CoreWeave describes its position against neoclouds and hyperscalers
Kim also asked about competition 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 view differentiation through third-party validation. Excluding China, nine of the world’s top ten AI labs use CoreWeave’s platform. SemiAnalysis has consistently placed CoreWeave alone at the highest level in performance. McBee said he does not believe CoreWeave receives its GPU allocations because of a personal relationship with Jensen. Rather, he said suppliers have deep confidence in the company’s execution record and engineering capability, and believe CoreWeave can best represent their products’ capabilities globally.
Robbins broke the customer wins into different categories. He said CoreWeave wins hyperscale cloud customers because it is very strong at execution, can build these systems extremely quickly and runs them very well. It wins research lab customers because it provides the highest-performing version of the technology and the best efficiency per token. It wins enterprise customers because the infrastructure works well and because the company has built what Robbins described as an excellent, best-in-class orchestration layer, which is part of the basis for recognitions such as the platinum rating.
Robbins added that an increasingly important layer has been built across inference and development tools. This helps enterprises put AI into production. According to Robbins, CoreWeave is building and delivering products that help less technically mature enterprises turn data into models and then turn those models into agents that can run internally. In that process, the company can also cross-sell CoreWeave cloud services.
Powered shells, HBM costs and the Vera Rubin ramp
On current bottlenecks, McBee pointed to powered shells, meaning data center shells with available power. More precisely, he pointed to the components inside those shells. Kim specifically mentioned electricians, and McBee said that was exactly right, describing the area as complex. He also emphasized that CoreWeave already has 49 such sites live and operating. The company is not placing its hopes on one or two sites; it has done this 49 times. That creates a deep execution record and a large body of knowledge about how to manage supply chain issues, including which suppliers in the chain are suitable partners and which are not.
Asked about the cost and shortage of HBM memory and whether customers need to bear price increases, Robbins answered yes. He said CoreWeave’s business model is designed so that when the company signs a GPU purchase order and determines the cost it will pay, it also locks in the GPU price charged to customers. More broadly, that means the server price, and the server price clearly includes HBM cost. Robbins said acquiring components is not currently the largest bottleneck. The largest bottleneck is the powered shell, though at some point in the future the answer can move back and forth.
On the expected deployment ramp for Vera Rubin in the second half of the year, Robbins said CoreWeave was clearly the first company in the world to start and fully validate VR, meaning Vera Rubin cabinets. He said the same was true last year with GB200 and GB300. Robbins expects VR to begin appearing later this year. He expects a truly large-scale and very strong deployment ramp to run through all of 2027. He compared the cadence with GB: GB began appearing in 2025, but the real large-scale ramp is running through 2026. In other words, a meaningful amount was deployed by the end of last year, but this year is the true large-scale deployment year for GB. Robbins expects VR to follow a very similar rhythm over the next 12 to 18 months.