Why Cerebras CEO Andrew Feldman Built The World's Largest Computer Chip

Cerebras CEO Andrew Feldman reveals how his company built the world's largest computer chip—58 times bigger than any previous chip, roughly the size of a dinner plate—by embracing wafer-scale integration. Unlike traditional AI chips that rely on networking multiple smaller chips, Cerebras packed its massive wafer with ultra-fast memory, eliminating the bottleneck that slows down inference. This architectural breakthrough makes their chips 15 times faster than leading GPUs and up to 1,000 times faster on certain tasks. The result? A new paradigm in AI performance that’s not just faster but dramatically more energy-efficient. Despite massive demand—evidenced by a $20 billion deal with OpenAI and a major AWS partnership—growth is now constrained not by manufacturing, but by data center capacity. Feldman also argues that speed is critical across all AI work, from simple queries to complex agentic workflows, and that open-source models are already significantly cheaper per unit of intelligence, setting up a fierce battle between closed and open AI. He dismisses CUDA’s dominance as outdated, noting that two of the three leading models today run without it. As the AI economy matures, he foresees financial markets evolving to include derivatives for compute capacity, while national security concerns remain a real but manageable challenge. The episode exposes a deeper truth: the future of AI isn’t just about smarter models, but about smarter infrastructure.