Cerebras Hot Seat: Sean Lie on Why Cerebras Builds a Big Chip

Why did Cerebras build a wafer-scale processor, and why can’t traditional GPU makers easily replicate it? In this episode of Hot Seat with Cerebras, CTO and Co-Founder Sean Lie sits down to answer the community's most pressing hardware and architecture questions—from silicon yield and defect self-healing to on-chip SRAM memory bandwidth, programming 900,000 cores, and data center deployment. Timestamps: 0:00 - Introduction & Hot Seat kick-off 0:36 - Why Cerebras built a wafer-scale chip 2:15 - Why can't incumbents build a bigger chip? 4:10 - Solving the wafer yield & defect puzzle 8:08 - Running massive AI models with on-chip SRAM 12:42 - Is Cerebras hard to program? (Software stack & compilers) 16:24 - Deploying wafer-scale systems in standard data centers 18:05 - Global data centers & wrap-up Learn more about Cerebras Wafer-Scale technology: https://cerebras.ai