CoreWeave executives stated that AI demand continues to accelerate and is expanding beyond GPUs to infrastructure layers such as CPUs and storage, with agentic AI emerging as the core driver.
The primary bottleneck has now shifted to physical data center supply, such as power-constrained facilities. The company is restructuring its architecture to accommodate rising inference demands, and the Vera Rubin platform is expected to enter a volume ramp-up phase in 2027.
Demand for AI computing power continues to expand without showing signs of peaking and is now spreading further across the broader infrastructure layer.
$CoreWeave (CRWV.US)$Brannin McBee, Co-Founder and Chief Development Officer, and Nick Robbins, Vice President of Corporate Development and Investor Relations, recently told media in an interview that AI demand is "finding new ways to intensify every day," showing no signs of slowing down. The executives noted that as the wave of agentic AI accelerates, market demand for complementary resources like CPUs and storage is rising significantly relative to GPUs, prompting a fundamental rearchitecture of data centers overall.
On the core bottleneck currently constraining AI infrastructure expansion, McBee explicitly identified it as the “powered shell”—data center facilities that have already been equipped with power infrastructure—rather than GPU chips or HBM memory. This assessment carries direct implications for evaluating the pace of AI infrastructure investment.
Demand Continues to Accelerate, with Agentic AI Emerging as the New Engine
$CoreWeave (CRWV.US)$Executives pinpointed the fourth quarter of last year as the starting point of this latest acceleration in AI demand. McBee said that through deep engineering engagements with clients at the time, the company had already anticipated a concentrated market launch of agentic AI products in the first quarter of this year. "From a product perspective in the AI market, Q1 marked a major inflection point for inference and AI consumption—and it’s still accelerating."
Robbins offered an even more direct characterization of current demand dynamics: “It’s finding new ways to intensify every day.”
This insight stems from CoreWeave’s unique position within the AI ecosystem. According to Robbins, CoreWeave is currently the only independent cloud service provider simultaneously serving leading AI organizations including Anthropic, OpenAI, Meta, Google, Microsoft, and NVIDIA. This vantage point enables the company to gain forward-looking insights into technological evolution and proactively plan its infrastructure deployments accordingly.
Architectural Reconfiguration: CPU and Storage Demand Rising Significantly Relative to GPUs
The rise of agentic AI and inference models is reshaping the hardware allocation logic within data centers.
McBee stated,$CoreWeave (CRWV.US)$the company has been operating CPU resources since 2023, but the current trend shows a clear increase in the relative share of CPU and storage demand compared to GPUs. "As agents and inference truly take off within models, storage requirements are also rising significantly relative to previous generations, and I believe this trend will persist."
Robbins revealed that CoreWeave fundamentally redesigned its standard data center blueprint last year to allocate more space for storage and CPUs.
He confirmed that in the future, numerous NVIDIA Vera CPU racks will be deployed side by side with Vera Rubin GPU servers.
Regarding CPU supplier selection, Robbins stated that the current fleet is predominantly AMD-based, but as customer requirements evolve, NVIDIA Vera CPUs are expected to become a significant early adoption focus, noting there is “substantial interest in Vera CPUs.” McBee added that over 98% of CoreWeave’s revenue is contract-driven, with customers explicitly specifying the infrastructure configurations they require: “They are defining what we build.”
Power infrastructure—the most pressing bottleneck for expansion
When asked about the biggest current constraint, McBee directly cited “power infrastructure,” highlighting labor shortages among electricians as one complicating factor. He noted that CoreWeave currently operates 49 live sites and has accumulated extensive hands-on experience managing supply chain challenges: “We know which vendors to work with and which ones to avoid.”
Robbins addressed concerns regarding HBM memory costs and shortages, explaining that the company’s business model effectively insulates it from price volatility—by locking in customer pricing at the time GPU purchase orders are signed, thereby safeguarding margins and enabling smooth cost pass-through to clients. He also noted, “Component availability is not currently the biggest bottleneck—it’s power infrastructure. But at some point in the future, that answer could flip.”
Vera Rubin mass production: 2027 will be the main event
Regarding the mass production timeline for NVIDIA’s next-generation Vera Rubin (VR) platform, Robbins provided a relatively clear forecast.
he said,$CoreWeave (CRWV.US)$the company is already the world’s first vendor to have fully deployed and validated VR racks, with VR servers expected to begin delivery in the second half of this year, though large-scale production ramp-up will extend throughout 2027.
Robbins compared this rollout cadence to that of the previous-generation GB200/GB300 platforms—GB-series systems emerged in 2025, but true mass production occurred in 2026. “I expect VR to follow a very similar pattern over the next 12 to 18 months.”
Building a moat through execution capability and ecosystem depth
Facing ultra-large-scale cloud providers (Azure, AWS,$Alphabet-A (GOOGL.US)$) and other emerging cloud service providers (SpaceX,$NEBIUS (NBIS.US)$、$Oracle (ORCL.US)$) competition,$CoreWeave (CRWV.US)$Executives attribute their differentiated advantage to three dimensions: execution speed, performance, and ecosystem depth.
Citing third-party validation data, McBee stated that nine of the world’s top ten AI labs (excluding those in China) use the CoreWeave platform, and AI research firm SemiAnalysis has awarded it a unique Platinum rating. He believes NVIDIA prioritizes GPU allocation to CoreWeave due to its strong trust in the company’s engineering execution capabilities, noting, 'This is a supplier that has deep confidence in our track record and engineering capabilities.'
Robbins explained the competitive logic from a customer segmentation perspective: against hyperscale cloud providers, CoreWeave wins with extremely rapid deployment and stable operations; for research institutions, it competes on peak performance and per-token efficiency; and for enterprise clients, it leverages its superior inference and development tool orchestration layer to help them transform data into models and agents, thereby enabling cross-selling of CoreWeave’s cloud services.
Editor/melody