CoreWeave (CRWV-US) identifies powered data centers as AI bottleneck, not chips

CoreWeave (CRWV-US) executives state "powered shells," or data center buildings with power, are the primary bottleneck for AI infrastructure expansion, not GPU chips or HBM memory. This redefines the critical constraint amid accelerating AI demand, particularly for AI agents, according to co-founder Brannin McBee and VP Nick Robbins in recent interviews. The cloud service provider observes no slowdown in AI demand, with CPU and memory requirements rising significantly relative to GPUs since Q1 2026. This trend, driven by the emergence of AI agent products, is prompting CoreWeave to fundamentally redesign its data center architectures to support next-generation AI workloads. CoreWeave, an independent cloud provider serving major AI players, anticipates NVIDIA's Vera Rubin platform deliveries in H2 2026, with large-scale ramp-up expected throughout 2027. The company's business model effectively passes HBM cost increases to clients, mitigating price fluctuation risks and protecting profit margins.