Show notes
“So, there are plenty of failure points, and when you have hundreds of thousands or millions of something, something will eventually fail,” Peter Salanki, co-founder and CTO of CoreWeave, tells Bloomberg Intelligence Senior Technology Analyst Anurag Rana. “Instead of throwing out half the potential capacity, we say that we expect some of these to fail. Then we build systems, automation and processes around handling those failures gracefully.” In this episode of Tech Disruptors, the pair discuss why AI infrastructure requires a fundamentally different architecture to traditional CPU-based cloud. Salanki also explains how CoreWeave is addressing training, inference and agentic workloads while navigating token costs, Nvidia chip demand and power constraints.
