Episode Details

AI demand is racing far ahead of supply, and Andrew Feldman says the real story is infrastructure, not a bubble. In this condensed version of the full 20VC episode, Harry Stebbings sits down with the Cerebras CEO to unpack the bottlenecks shaping AI data centers, token costs, memory shortages, and semiconductor geopolitics—cut from the full conversation into a quick listen that saves you time. Feldman explains why a $25B backlog, multi-year fab timelines, and scarce packaging capacity are keeping the industry supply-constrained, and why faster compute still matters more than cheaper tokens. He also breaks down how AI is changing enterprise workflows, where legal and security teams slow adoption, and why organized data becomes the next competitive edge. Finally, he weighs in on whether US companies should sell leading-edge chips to China, arguing for tighter export controls and more domestic semiconductor capacity. Listen now to get the key ideas in minutes.