CoreWeave Identifies Power Infrastructure as AI Infrastructure Bottleneck, Not GPU or HBM

According to recent media interviews, CoreWeave (CRWV-US) executives stated that the key limitation for AI infrastructure expansion has shifted from chip availability to physical data center power infrastructure. CoreWeave co-founder and Chief Development Officer Brannin McBee pointed out that the "powered shell"—completed data center buildings with power configuration—is the critical bottleneck, not GPU chips or HBM memory.

The company noted that as AI agent adoption accelerates, demand for CPU and memory resources is rising relative to GPU demand. CoreWeave operates 49 sites globally and generates over 98% of revenue from contractual customers. The company has completed validation of Nvidia's Vera Rubin platform and expects server deliveries to begin in the second half of 2026, with large-scale production ramping through 2027.