OpenAI’s Compute Chief: We Can’t Build Fast Enough | Sachin Katti
OpenAI’s Compute Chief: We Can’t Build Fast Enough | Sachin Katti
The MAD Podcast: How AI Gets Built — with Matt Turck
番組の概要欄（原文）
<p>Is the AI industry actually overbuilding, or is the physical world moving too slowly to keep up? In this episode of the MAD Podcast, OpenAI's Head of Industrial Compute, Sachin Katti, takes us inside the "belly of the beast" of what may be the largest infrastructure project in human history. We explore the staggering physical reality of the AI boom—from $50 billion supercomputers and liquid-cooled data centers that "turn electrons into tokens," to overhauling the U.S. power grid and exploring nuclear energy. Sachin also pulls back the curtain on OpenAI's Stargate strategy, their move into custom silicon with Project Jalapeno, and the mind-bending reality that AI is now beginning to design the very chips that will power its own future.</p><p><br></p><p>(00:00) — Cold open: “One of the largest things humanity has ever built”<br>(00:30) — Welcome: Sachin Katti, Head of Industrial Compute at OpenAI<br>(01:44) — Is this the biggest infrastructure buildout in history?<br>(03:41) — Why OpenAI is building a new industrial muscle<br>(04:54) — What an AI data center actually is<br>(05:27) — “Factories turning electrons into tokens”<br>(06:35) — Why AI data centers need liquid cooling everywhere<br>(08:10) — The power problem: grids, generation, transmission, substations<br>(10:43) — Behind-the-meter power and gas turbines<br>(11:02) — Why nuclear “can’t come soon enough”<br>(11:49) — Jalapeño: why OpenAI is designing its own AI chips<br>(13:19) — Tokens per watt: the new metric that matters<br>(13:38) — Why inference may now dominate AI compute<br>(14:58) — Is OpenAI overbuilding compute?<br>(16:47) — Why OpenAI thinks the bigger risk is not building fast enough<br>(17:55) — Communities, jobs, water, and the local data-center debate<br>(21:16) — How OpenAI chooses data-center sites<br>(22:25) — What “industrial compute” means inside OpenAI<br>(25:59) — Sachin’s path: Stanford, startups, Intel, OpenAI<br>(28:05) — OpenAI’s compute portfolio: Microsoft, hyperscalers, neoclouds<br>(29:37) — Stargate explained<br>(31:21) — Abilene, Oracle, and the next wave of AI data centers<br>(32:48) — How massive AI compute gets financed<br>(34:05) — How OpenAI designed Jalapeño so quickly<br>(35:59) — AI is starting to help design AI chips<br>(36:20) — MRC: the networking problem behind 100,000 GPUs<br>(38:47) — Bottlenecks: transformers, turbines, electricians, supply chains<br>(40:29) — Guaranteed capacity: intelligence as a supply unit<br>(42:08) — Will AI data centers move to space?</p>